

<!-- formatList: 

 Array
(
    [5 users] =&gt; Array
        (
            [slug] =&gt; 5 users
            [name] =&gt; 5 users
            [canonical] =&gt; five users
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [a---b test] =&gt; Array
        (
            [slug] =&gt; a---b test
            [name] =&gt; a---b test
            [canonical] =&gt; a-b test
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [a---b testing] =&gt; Array
        (
            [slug] =&gt; a---b testing
            [name] =&gt; a---b testing
            [canonical] =&gt; a-b test
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [a-b test] =&gt; Array
        (
            [slug] =&gt; a-b test
            [name] =&gt; A&amp;ndash;B test
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A&amp;ndash;B test is a form of automated {{usability evaluation}}, used particularly on the web. It involves releasing two (or more) minor variants of an application, A and B, for example two different colour schemes or positions for a {{call-to-action button}}. Metrics are gathered automatically, such as {{click-through rate}} or {{time on page}}, and these are used to determine which of the variants is best.
            [has_links] =&gt; 
        )

    [a-b testing] =&gt; Array
        (
            [slug] =&gt; a-b testing
            [name] =&gt; a-b testing
            [canonical] =&gt; a-b test
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [acm chi] =&gt; Array
        (
            [slug] =&gt; acm chi
            [name] =&gt; ACM CHI
            [canonical] =&gt; 
            [see] =&gt; chi conference
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [acm conference] =&gt; Array
        (
            [slug] =&gt; acm conference
            [name] =&gt; ACM conference
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; ACM is one of the major international organisations for computing. ACM conferences such as {{CHI}} are often among the most prestigious in their areas.
            [has_links] =&gt; 1
        )

    [alternative hypothesis] =&gt; Array
        (
            [slug] =&gt; alternative hypothesis
            [name] =&gt; alternative hypothesis
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In {{hypothesis testing}} the alternative hypothesis, often written H&lt;sub&gt;1&lt;/sub&gt;, is usually the thing you would like to show is true. For example, that your updated interface design is better than the old one.
            [has_links] =&gt; 
        )

    [alternative statistics] =&gt; Array
        (
            [slug] =&gt; alternative statistics
            [name] =&gt; alternative statistics
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; One of the responses to the {{statistical crisis}} has been to look towards different kinds of statistical methods including {{Bayesian statistics}} and {{simulation methods}}. Collectively these are sometimes called 'new statistics', 'alternative statistics', or even 'alt stats'. The term is also sometimes used to include lesser used aspects of more traditional statistics such as {{confidence intervals}} and measures of {{effect size}}.
            [has_links] =&gt; 
        )

    [american statistical association] =&gt; Array
        (
            [slug] =&gt; american statistical association
            [name] =&gt; American Statistical Association (ASA)
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This is the world's largest statistical organisation and second oldest after the {{Royal Statistical Society}}.
            [has_links] =&gt; 1
        )

    [analysis of variance] =&gt; Array
        (
            [slug] =&gt; analysis of variance
            [name] =&gt; analysis of variance
            [canonical] =&gt; anova
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [anova] =&gt; Array
        (
            [slug] =&gt; anova
            [name] =&gt; ANOVA (Analysis of Variance)
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This is a form of statistical analysis used for {{multivariate data}} that is {{Normally distributed}}. It is quite robust, in the sense that even if the data is not completely Normal, the results are fairly reliable. You might use this, for example, if you have performed an experiment measuring response times with several users, each performing several tasks with two alternative interactive systems; in this case you have three variables: user x task x system. You are probably interested in which system is better overall, but the different tasks may vary in difficulty and the users may differ in expertise. The ANOVA can disentangle these effects, giving you {{main effects}}: averages for each user, task and system; as well as {{interaction effects}} such as whether some tasks are better suited to one system or the other.
            [has_links] =&gt; 
        )

    [applied mathematics] =&gt; Array
        (
            [slug] =&gt; applied mathematics
            [name] =&gt; applied mathematics
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The branch of mathematics that deals with real world problems.
            [has_links] =&gt; 
        )

    [approximate distribution] =&gt; Array
        (
            [slug] =&gt; approximate distribution
            [name] =&gt; approximate distribution
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A {{distribution}} that is close enough to another distribution to be of use in calculations. The distribution being used for the approximation is usually in some ways more mathematically tractable or easier to deal with. For example, the {{Normal distribution}} is often used as an approximation for other distributions, even {{discrete distributions}} such as the {{Binomial distribution}}. 
            [has_links] =&gt; 
        )

    [approximately normal] =&gt; Array
        (
            [slug] =&gt; approximately normal
            [name] =&gt; approximately Normal
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A distribution is &lt;strong&gt;approximately Normal&lt;/strong&gt; when the {{Normal distribution}} can be used as an {{approximate distribution}}. This is common when the number of samples or parts making up a distribution grows; for example, if you have 100 coin tosses the resulting {{Binomial distribution}} is, for most purposes, approximately Normal. This is often helpful as many statistical tests rely on the data being {{Normally distributed}}. The {{Central Limit Theorem}} explains why it is so common to see approximately Normal distributions.
            [has_links] =&gt; 
        )

    [arithmetic mean] =&gt; Array
        (
            [slug] =&gt; arithmetic mean
            [name] =&gt; arithmetic mean
            [canonical] =&gt; mean
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [arithmetical data] =&gt; Array
        (
            [slug] =&gt; arithmetical data
            [name] =&gt; arithmetical data
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A form of {{discrete data}} where the values can be meaningfully added or averaged. Contrast with {{categorical data}} and {{ordinal data}}.
            [has_links] =&gt; 
        )

    [asa] =&gt; Array
        (
            [slug] =&gt; asa
            [name] =&gt; asa
            [canonical] =&gt; american statistical association
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [asymmetric distribution] =&gt; Array
        (
            [slug] =&gt; asymmetric distribution
            [name] =&gt; asymmetric distribution
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Where a distribution curve is not symmetrical: an example would be a distribution that rises quite rapidly from zero, but then tails off more slowly.
            [has_links] =&gt; 
        )

    [average] =&gt; Array
        (
            [slug] =&gt; average
            [name] =&gt; average
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The term {{average}} most commonly refers to the arithmetic {{mean}}, but can also refer to the {{median}}, especially for {{asymmetric distributions}} where the {{median}} is typically a little smaller than the {{mean}}, which can be influenced by a few large {{outliers}}. For example, 'average income' usually refers to the median, not the {{mean}}.
            [has_links] =&gt; 
        )

    [balance orders] =&gt; Array
        (
            [slug] =&gt; balance orders
            [name] =&gt; balance orders
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A way of mitigating {{order effects}} by, for example, having half the subjects in an experiment perform a task in the order A&ndash;B and half in the order B&ndash;A.
            [has_links] =&gt; 
        )

    [base period] =&gt; Array
        (
            [slug] =&gt; base period
            [name] =&gt; base period
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; For temporal data, the starting point in time, also known as base year. This is a special case of a {{basepoint}}.
            [has_links] =&gt; 
        )

    [base rate] =&gt; Array
        (
            [slug] =&gt; base rate
            [name] =&gt; base rate
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The base rate is the underlying level of prevalence of some attribute of a population. For example, the base rate of a disease is the proportion of people who normally have the illness. Where such a base rate is known it can be used as {{prior probability}} in {{Bayesian methods}}. The base rate is important in assessing how likely the underlying attribute is for a particular individual. If you ignore the base rate you may make a poor assessment of the situation, sometimes called the base rate fallacy or base rate neglect. For example, imagine a test for a rare form of cancer that is 100% accurate in detecting if you the cancer and 99% accurate when you don't. You find you have tested positive. At first this seems very worrying; however, if the base rate is that only 1 in a million people have the cancer but the other 999,999 don't,  then for every person who has the cancer and tests positive there are 1% of 999,999, that is around 10,000, who test positive and are in the clear. It may be worth having further tests, but you are no more likely to have the disease than to be killed on the road that year.
            [has_links] =&gt; 1
        )

    [baseline] =&gt; Array
        (
            [slug] =&gt; baseline
            [name] =&gt; baseline
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The choice of where to place the bottom of a graph (y-axis): at zero, or at some other value, a {{false baseline}}.
            [has_links] =&gt; 
        )

    [basepoint] =&gt; Array
        (
            [slug] =&gt; basepoint
            [name] =&gt; basepoint
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The choice of where to place the left-to-right start point of a graph (x-axis): at zero, or at some other value, a {{false basepoint}}.

            [has_links] =&gt; 
        )

    [bayes factor] =&gt; Array
        (
            [slug] =&gt; bayes factor
            [name] =&gt; Bayes factor
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The &lt;strong&gt;Bayes factor&lt;/strong&gt; is used in {{Bayesian statistics}} to measure the relative probability of alternatives based on some form of observed data. For a simple H&lt;sub&gt;0&lt;/sub&gt; vs H&lt;sub&gt;1&lt;/sub&gt; alternative (e.g. human vs Martian), the Bayes factor is the ratio of the odds for the two alternatives: P(H&lt;sub&gt;1&lt;/sub&gt; | observed data) / P(H&lt;sub&gt;0&lt;/sub&gt; | observed data). However, if H&lt;sub&gt;0&lt;/sub&gt; vs H&lt;sub&gt;1&lt;/sub&gt; is more complex (e.g. mean = 0 vs any other value), then calculating the Bayes factor involves averaging over the alternative possibilities weighted by the {{prior probability distribution}}.
            [has_links] =&gt; 
        )

    [bayes rule] =&gt; Array
        (
            [slug] =&gt; bayes rule
            [name] =&gt; Bayes rule
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Bayes rule is a mathematical equation that allows you to calculate {{conditional probabilities}} when you know conditional probabilities the 'other way round'. For example, if you know the probability of a patient exhibiting various symptoms if they have a particular illness, it allows you to calculate the probability that they have the illness given the symptoms they are showing. In the case of two mutually exclusive states A and B (e.g. A=&quot;has a cold&quot;, B=&quot;does not have a cold&quot;), and an observation C (e.g. &quot;has a runny nose&quot;), then Bayes rule is P(A|C) = P(C|A)P(A) / ( P(C|A)P(A) + P(C|A)P(B) ). Note that for this you need to know the {{prior probability}} of each occurrence, P(A) and P(B); when these are known from data they are also known as the {{base rate}}.
            [has_links] =&gt; 
        )

    [bayesian analysis] =&gt; Array
        (
            [slug] =&gt; bayesian analysis
            [name] =&gt; bayesian analysis
            [canonical] =&gt; bayesian methods
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [bayesian approach] =&gt; Array
        (
            [slug] =&gt; bayesian approach
            [name] =&gt; bayesian approach
            [canonical] =&gt; bayesian methods
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [bayesian evidence] =&gt; Array
        (
            [slug] =&gt; bayesian evidence
            [name] =&gt; Bayesian evidence
            [canonical] =&gt; bayesian methods
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [bayesian inference] =&gt; Array
        (
            [slug] =&gt; bayesian inference
            [name] =&gt; Bayesian inference
            [canonical] =&gt; 
            [see] =&gt; bayesian reasoning
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [bayesian methods] =&gt; Array
        (
            [slug] =&gt; bayesian methods
            [name] =&gt; Bayesian methods
            [canonical] =&gt; 
            [see] =&gt; bayesian reasoning
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [bayesian posterior] =&gt; Array
        (
            [slug] =&gt; bayesian posterior
            [name] =&gt; Bayesian posterior probability
            [canonical] =&gt; 
            [see] =&gt; posterior distribution
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [bayesian posterior probability] =&gt; Array
        (
            [slug] =&gt; bayesian posterior probability
            [name] =&gt; bayesian posterior probability
            [canonical] =&gt; bayesian posterior
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [bayesian reasoning] =&gt; Array
        (
            [slug] =&gt; bayesian reasoning
            [name] =&gt; Bayesian reasoning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Bayesian reasoning is the use of {{Bayes rule}} to allow inferences to be made about the probability of things based on evidence. It relies crucially on already knowing the {{prior probability}} of different alternatives before the evidence is available. Bayesian reasoning can be used for internal algorithms, for example to make adaptive or {{intelligent interfaces}}. It can also be used in a form of statistical reasoning, {{Bayesian statistics}}, which can be used as an alternative to traditional {{hypothesis testing}} or other statistical analyses.
            [has_links] =&gt; 
        )

    [bayesian statistical inference] =&gt; Array
        (
            [slug] =&gt; bayesian statistical inference
            [name] =&gt; bayesian statistical inference
            [canonical] =&gt; bayesian inference
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [bayesian statistical methods] =&gt; Array
        (
            [slug] =&gt; bayesian statistical methods
            [name] =&gt; bayesian statistical methods
            [canonical] =&gt; bayesian statistics
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [bayesian statistics] =&gt; Array
        (
            [slug] =&gt; bayesian statistics
            [name] =&gt; Bayesian statistics
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Bayesian statistics uses {{Bayesian reasoning}} as an alternative to {{traditional statistics}}, such as {{significance testing}} or {{confidence intervals}}. Crucially this demands that you make precise your {{prior belief}} about the outcomes {{encoded as a probability}}. In some cases your belief will be based on existing data or evidence; in these cases {{Bayes rule}} gives a strong mathematical foundation, yielding a precise {{posterior probability}} given the additional evidence. More commonly the prior is a guesstimate or subjective assessment and the {{posterior probability}} is not really a probability, but rather evidence-based belief. 
            [has_links] =&gt; 
        )

    [bayesian stats] =&gt; Array
        (
            [slug] =&gt; bayesian stats
            [name] =&gt; bayesian stats
            [canonical] =&gt; bayesian statistics
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [bayesian test] =&gt; Array
        (
            [slug] =&gt; bayesian test
            [name] =&gt; Bayesian test
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A type of statistical technique using Bayesian methods to address similar problems to traditional statistical {{significance testing}}.
            [has_links] =&gt; 
        )

    [belief encoded as probability] =&gt; Array
        (
            [slug] =&gt; belief encoded as probability
            [name] =&gt; belief encoded as probability
            [canonical] =&gt; encoding as probability
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [between-subjects design] =&gt; Array
        (
            [slug] =&gt; between-subjects design
            [name] =&gt; between-subjects design
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; An experiment design where there are many subjects, but each only performs in a single condition. Contrast with a {{within-subjects design}}.
            [has_links] =&gt; 
        )

    [between-subjects experiment] =&gt; Array
        (
            [slug] =&gt; between-subjects experiment
            [name] =&gt; between-subjects experiment
            [canonical] =&gt; between-subjects design
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [bias] =&gt; Array
        (
            [slug] =&gt; bias
            [name] =&gt; bias
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
Bias is about systematic effects that skew your results in one way or another.
            [has_links] =&gt; 
        )

    [big data] =&gt; Array
        (
            [slug] =&gt; big data
            [name] =&gt; big data
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Vast amounts of data that are now available and allow us to study phenomena in a way that would have been impossible only a few years ago.
            [has_links] =&gt; 
        )

    [bimodal distribution] =&gt; Array
        (
            [slug] =&gt; bimodal distribution
            [name] =&gt; bimodal distribution
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A bimodal distribution has two main peaks or {{modes}}. This often happens when there is a {{mixed population}} consisting of two distinct groups with different characteristics. Educators may be familiar with this in courses with more formal or mathematical materials; there is often a group that 'gets it' and has marks clustered towards the high grades, while the rest have marks clustered towards the lower end of the range.
            [has_links] =&gt; 
        )

    [binary data] =&gt; Array
        (
            [slug] =&gt; binary data
            [name] =&gt; binary data
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A special case of discrete data with only two states, such as yes/no answers, or present/not present.
            [has_links] =&gt; 
        )

    [binary variable] =&gt; Array
        (
            [slug] =&gt; binary variable
            [name] =&gt; binary variable
            [canonical] =&gt; binary data
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [binomial] =&gt; Array
        (
            [slug] =&gt; binomial
            [name] =&gt; Binomial
            [canonical] =&gt; binomial distribution
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [binomial distribution] =&gt; Array
        (
            [slug] =&gt; binomial distribution
            [name] =&gt; Binomial distribution
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A distribution resulting from the sum of independent two-outcome events such as coin tosses.
            [has_links] =&gt; 
        )

    [bonferroni correction] =&gt; Array
        (
            [slug] =&gt; bonferroni correction
            [name] =&gt; Bonferroni correction
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A method of dividing the target {{p-level}} by the number of tests, used when there have been multiple tests to reduce the chance of thinking that random effects are real. For example, if you run 10 tests there is an approximately 50:50 chance of seeing at least one come out at 5% significance level by chance. In order to have a real 5% level overall, you therefore test each one at 0.5% (5%/10). It is a form of {{multiplicity adjustment}}.
            [has_links] =&gt; 
        )

    [borrowed methods] =&gt; Array
        (
            [slug] =&gt; borrowed methods
            [name] =&gt; borrowed methods
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Methods derived from  other disciplines, for example psychology, where controlled experiments typically have a single cause and are in highly controlled environments. Care must be taken, because interaction studies, even in lab settings, almost always have multiple causes and open situations.
            [has_links] =&gt; 
        )

    [bounded] =&gt; Array
        (
            [slug] =&gt; bounded
            [name] =&gt; bounded
            [canonical] =&gt; bounded data
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [bounded below] =&gt; Array
        (
            [slug] =&gt; bounded below
            [name] =&gt; bounded below
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Data which is finite at the low end of the range but may be unbounded at the high end, e.g. when a coin is tossed, the number of heads until the first tail.
            [has_links] =&gt; 
        )

    [bounded data] =&gt; Array
        (
            [slug] =&gt; bounded data
            [name] =&gt; bounded data
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Data which has a limited range of responses with either a fixed maximum or minimum. This may be {{discrete data}}: for example, the number of heads in 6 coin tosses can only be between 0 and 6. You cannot have -3 heads or 7 heads ({{bounded above}} and {{bounded below}}). Similarly, for {{continuous data}} the user {{response time}} cannot be negative ({{bounded below}}).
            [has_links] =&gt; 
        )

    [bounded values] =&gt; Array
        (
            [slug] =&gt; bounded values
            [name] =&gt; bounded values
            [canonical] =&gt; bounded data
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [bounds] =&gt; Array
        (
            [slug] =&gt; bounds
            [name] =&gt; bounds
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; These are the maximum {{upper bound}} and minimum {{lower bound}} values that it is possible for a value to take. For example the bounds for the roll of a die are 1 (lower bound) and 6 (upper bound). Not all variables have bounds: for example, when {{coin tossing}} the number of tails until you get a head is {{bounded below}} by 0 (you can't have a negative number of tails), but {{unbounded above}}, so it has a lower bound, but no upper bound.
            [has_links] =&gt; 
        )

    [categorical data] =&gt; Array
        (
            [slug] =&gt; categorical data
            [name] =&gt; categorical data
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Data which has {{discrete values}} that cannot easily be compared, added, or averaged. Also known as nominal data. Contrast with {{arithmetical data}} and {{ordinal data}}.
            [has_links] =&gt; 
        )

    [cauchy distribution] =&gt; Array
        (
            [slug] =&gt; cauchy distribution
            [name] =&gt; Cauchy distribution
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A bell-shaped distribution often used for unbounded value ranges in {{Bayesian statistics}}. Similar to a {{Normal distribution}} but with fatter edges (tails). It is a {{symmetric distribution}} with {{infinite variance}}. The formula for the simple case with {{mean}} 0 is 1/&amp;pi;(1+x&lt;sup&gt;2&lt;/sup&gt;).
            [has_links] =&gt; 1
        )

    [causality] =&gt; Array
        (
            [slug] =&gt; causality
            [name] =&gt; causality
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; From data we may observe that two phenomena are related, even that one always precedes the other. However, it can sometimes be hard to establish causality - which is the cause of the other, the reason it happens or has the value it does. Sometimes two phenomena may be linked by a common cause rather than one being the cause of the other. For example, from data we might note that people who do a lot of exercise are also more healthy, and so think that exercise causes good health, but equally well it might be that people who are in good health are more likely to take exercise, or that there is some aspect of people's upbringing or genes that leads to both. Understanding {{mechanism}} can sometimes help establish causality.
            [has_links] =&gt; 
        )

    [central limit theorem] =&gt; Array
        (
            [slug] =&gt; central limit theorem
            [name] =&gt; Central Limit Theorem
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A mathematical result, which says that the average ({{arithmetic mean}}) of lots of items or the sum of lots of very small things behaves more and more like a {{Normal distribution}} as the number of small items gets larger.
            [has_links] =&gt; 
        )

    [cherry picking] =&gt; Array
        (
            [slug] =&gt; cherry picking
            [name] =&gt; cherry picking
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A common problem in statistics, when researchers deliberately or accidentally ignore results that are not to their liking and only report those that are advantageous. Cherry picked results may mislead people (including the researchers themselves) or make them falsely believe something is true.
            [has_links] =&gt; 
        )

    [chi] =&gt; Array
        (
            [slug] =&gt; chi
            [name] =&gt; CHI
            [canonical] =&gt; chi conference
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [chi conference] =&gt; Array
        (
            [slug] =&gt; chi conference
            [name] =&gt; CHI conference
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Annual ACM Conference on Human Factors in Computing Systems
            [has_links] =&gt; 
        )

    [citation count] =&gt; Array
        (
            [slug] =&gt; citation count
            [name] =&gt; citation count
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The number of times a paper has been cited by others.
            [has_links] =&gt; 
        )

    [citation data] =&gt; Array
        (
            [slug] =&gt; citation data
            [name] =&gt; citation data
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This is the data arising from counting or recording the way academic papers reference each other. It is a form of {{network data}} where papers are linked in a directional way as one paper cites another. A simple metric of this is the {{citation count}}, the number of times a paper has been cited by others. However, it is important to note that, in common with other forms of network data, the citation count usually follows a {{power law}} with a {{long tail}} and is not suitable for tests and analysis that rely on the data being {{approximately Normal}}.
            [has_links] =&gt; 
        )

    [claims analysis] =&gt; Array
        (
            [slug] =&gt; claims analysis
            [name] =&gt; claims analysis
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A method developed by Carroll and Rosson, which seeks to better integrate activities directed at understanding with those directed at design. DOI: &lt;doi&gt;10.1145/146802.146834&lt;/doi&gt;
            [has_links] =&gt; 
        )

    [closed formula] =&gt; Array
        (
            [slug] =&gt; closed formula
            [name] =&gt; closed formula
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A tractable mathematical description.
            [has_links] =&gt; 
        )

    [closed world] =&gt; Array
        (
            [slug] =&gt; closed world
            [name] =&gt; closed world
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A world where we assume we know everything and there are no unknown external factors. {{Bayesian statistics}} effectively uses a closed world assumption as it requires all unknown parameters to be assigned a {{prior probability}}. In contrast, {{traditional statistics}} adopts an {{open world}} model where some parameters are treated as completely unknown.
            [has_links] =&gt; 
        )

    [cognitive load] =&gt; Array
        (
            [slug] =&gt; cognitive load
            [name] =&gt; cognitive load
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The amount of mental resources (e.g. memory, attention) needed for a task or activity.
            [has_links] =&gt; 
        )

    [coin tosses] =&gt; Array
        (
            [slug] =&gt; coin tosses
            [name] =&gt; coin tosses
            [canonical] =&gt; coin tossing
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [coin tossing] =&gt; Array
        (
            [slug] =&gt; coin tossing
            [name] =&gt; coin tossing
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Simple coin tossing experiments can help you to understand random phenomena and statistical procedures. When doing statistics for real, you are having to disentangle some underlying effect (say the ease of use of an interface) from effectively random or interfering effects (such as the different levels of skill of your participants). This is difficult. With coin tossing you know the coins are unbiased, 50:50 heads:tails, and hence any odd effects you see are purely due to randomness. The number of heads or tails when you toss a fixed number of coins follow the {{Binomial distribution}}.
            [has_links] =&gt; 1
        )

    [computer simulation] =&gt; Array
        (
            [slug] =&gt; computer simulation
            [name] =&gt; computer simulation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; When you emulate some real world phenomenon in a computer. As used in statistics this will usually involve random values with repeated runs of the simulation in order to obtain average values or distributions of values. This can be used in {{simulation methods}} to calculate the {{likelihood}} of particular measurements even when there is no simple mathematical formula.
            [has_links] =&gt; 
        )

    [conditional probability] =&gt; Array
        (
            [slug] =&gt; conditional probability
            [name] =&gt; conditional probability
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This is the probability of one event (say A) &lt;emph&gt;given&lt;emp&gt; knowledge of another event (B). It is usually written as P(A|B), read as &quot;the probability of A given B&quot;. For example, while the probability that it will rain on a typical day might be 50%, on a cloudy day this might be 80%; that is P(rain|cloudy) = 0.8.
            [has_links] =&gt; 
        )

    [confidence interval] =&gt; Array
        (
            [slug] =&gt; confidence interval
            [name] =&gt; confidence interval
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Confidence intervals give a lower and upper bound for the true value of an effect. They allow you to be able to answer questions such as &quot;What is the maximum likely value of this?&quot;, or &quot;How close to no difference is there between these two conditions?&quot;. They use the same information and mathematics as are used to generate the {{p-values}} in a {{significance test}}. You can choose to set higher or lower confidence levels, for example the 95% confidence interval for a value might be [0.3,1.1] and the 99% confidence interval [-0.1,1.5]. In general the higher the level of confidence you ask for, the wider the range. Like all statistics these bounds are {{uncertain}}. However, if you do lots of studies and calculate 95% confidence intervals for them all, then over time the proportion of true values that lie within the 95% confidence intervals will be at least 95% ... although you can't tell which they are!
            [has_links] =&gt; 
        )

    [confidence intervals] =&gt; Array
        (
            [slug] =&gt; confidence intervals
            [name] =&gt; confidence intervals
            [canonical] =&gt; confidence interval
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [confirmation bias] =&gt; Array
        (
            [slug] =&gt; confirmation bias
            [name] =&gt; confirmation bias
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Confirmation bias is when we unintentionally only seek out evidence or take notice of results that reinforce our existing views. With {{traditional statistics}} this might lead us not to publish results that we think are wrong, perhaps even removing data points that we feel must have been errors of some kind. In {{Bayesian statistics}} confirmation bias can cause even more problems, as the choice of {{prior probability}} is effectively one's own beliefs {{encoded as a probability}} and hence the output of the process, the {{posterior probability}}, may simply reflect one's existing prejudices.
            [has_links] =&gt; 
        )

    [confounding variables] =&gt; Array
        (
            [slug] =&gt; confounding variables
            [name] =&gt; confounding variables
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; When there is some form of relationship or correlation between the various factors of which you are measuring aspects. For example, if you are comparing the speed of use of two systems, hoping that your new system is faster, a confounding variable might be user expertise: if the users of one system are more expert than those of the other, this may create a difference in {{response time}} that is not due to the system. One way to deal with such a situation would be to create a {{balanced design}}, where the expertise of the users of each system is the same. When this is not possible, one can attempt to {{model}} the effect of expertise and correct for it.
            [has_links] =&gt; 
        )

    [conscious mind] =&gt; Array
        (
            [slug] =&gt; conscious mind
            [name] =&gt; conscious mind
            [canonical] =&gt; conscious thinking
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [conscious thinking] =&gt; Array
        (
            [slug] =&gt; conscious thinking
            [name] =&gt; conscious thinking
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; These are the thoughts that you are explicitly aware of having: what you would answer if asked &quot;What are you thinking of now?&quot;. The conscious mind is extremely powerful, allowing us to think rationally and logically, but it is also limited. In particular, your conscious system of thought likes to have a single model of the world and is really bad at probability. Contrast with {{subconscious reactions}}.
            [has_links] =&gt; 
        )

    [constant area visualisation] =&gt; Array
        (
            [slug] =&gt; constant area visualisation
            [name] =&gt; constant area visualisation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A visualisation for hierarchical data in which the area of each part reflects the number or size of the items it represents (e.g. a {{TreeMap}} or {{PieTree}})
            [has_links] =&gt; 
        )

    [context effects] =&gt; Array
        (
            [slug] =&gt; context effects
            [name] =&gt; context effects
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Elements of the context in which an experiment is conducted that may affect the subjects' responses and therefore influence the results of the study, for example the mood of the researcher or even the weather. 
            [has_links] =&gt; 
        )

    [continuous] =&gt; Array
        (
            [slug] =&gt; continuous
            [name] =&gt; continuous
            [canonical] =&gt; continuous data
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [continuous data] =&gt; Array
        (
            [slug] =&gt; continuous data
            [name] =&gt; continuous data
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A set of numerical data where the values can be anywhere in a range, including fractional values, for example task completion time, or height. Contrast with {{discrete data}} which can only take on specific values. 
            [has_links] =&gt; 
        )

    [continuous distribution] =&gt; Array
        (
            [slug] =&gt; continuous distribution
            [name] =&gt; continuous distribution
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A probability distribution resulting from a set of {{continuous data}}, such as time or height values. The {{Normal distribution}} is an example of a continuous distribution. Because the continuous data can take on any of the infinite number of fractional values, the probability of any precise value is effectively zero. Instead, the probability of a range of values is calculated as the area under the curve of the distribution function over the range. 
            [has_links] =&gt; 
        )

    [continuous distributions] =&gt; Array
        (
            [slug] =&gt; continuous distributions
            [name] =&gt; continuous distributions
            [canonical] =&gt; continuous distribution
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [controlled environments] =&gt; Array
        (
            [slug] =&gt; controlled environments
            [name] =&gt; controlled environments
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In physics and other sciences, scientists use very pure substances with very precise instruments to reduce the variability in a study. In HCI we can sometimes emulate this, controlling conditions to reduce the variability in our study ({{reduce noise}}). For example, in a user test we might ask people to use a mobile phone to perform the same tasks in the same location with similar levels of sound and distractions. The downside of this is that it can sometimes reduce {{ecological validity}}, making the resulting conditions less like the real world.
            [has_links] =&gt; 
        )

    [controlled experiment] =&gt; Array
        (
            [slug] =&gt; controlled experiment
            [name] =&gt; controlled experiment
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Controlled experiments follow a strict protocol, a predefined number of trials, often in a laboratory setting. In HCI, controlled experiments may influence results. For example, experts may behave more like novices in the artificial  environment of a lab experiment, being more careful about their actions and monitoring the results. Typically, controlled experiments have to sacrifice {{ecological validity}}, being less like the equivalent real-world activity. Contrast with {{ethnographic methods}} and {{in-the-wild}} studies. 
            [has_links] =&gt; 
        )

    [controlled laboratory experiments] =&gt; Array
        (
            [slug] =&gt; controlled laboratory experiments
            [name] =&gt; controlled laboratory experiments
            [canonical] =&gt; controlled experiment
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [controlled tasks] =&gt; Array
        (
            [slug] =&gt; controlled tasks
            [name] =&gt; controlled tasks
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Restricting tasks to enable more valid study of individual variables: for example, participants are in a real situation, but instead of doing what they like, they are given a scripted task to perform. While some realism is lost in terms of the chosen task, this increases the level of {{ecological validity}} in the environment.
            [has_links] =&gt; 
        )

    [coping mechanisms] =&gt; Array
        (
            [slug] =&gt; coping mechanisms
            [name] =&gt; coping mechanisms
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The resources that people draw on when they find ways to manage a difficult task, whether consciously or unconsciously. While this is useful for getting a job done, for the designer it can mean that usability problems may be hidden. {{Distractor tasks}} are one way to deal with this.
            [has_links] =&gt; 
        )

    [correlation coefficient] =&gt; Array
        (
            [slug] =&gt; correlation coefficient
            [name] =&gt; correlation coefficient
            [canonical] =&gt; 
            [see] =&gt; pearson correlation coefficient
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [count data] =&gt; Array
        (
            [slug] =&gt; count data
            [name] =&gt; count data
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Data obtained by simply counting something, such as the number of people, errors, or clicks of a button. This is often found in data collected by surveys, such as opinion polls and questionnaires, or in discrete data in a user study, such as counting kinds of errors.
            [has_links] =&gt; 
        )

    [counterfactual] =&gt; Array
        (
            [slug] =&gt; counterfactual
            [name] =&gt; counterfactual
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A form of reasoning where one has to work with states of the world about which one is agnostic or which one even believes to be false.
            [has_links] =&gt; 
        )

    [covariance] =&gt; Array
        (
            [slug] =&gt; covariance
            [name] =&gt; covariance
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The covariance is the average product of two variables after their means have been subtracted: cov(x,y) = ( &amp;Sigma; (x&lt;sub&gt;i&lt;/sub&gt; - &amp;mu;&lt;sub&gt;x&lt;/sub&gt;) &amp;times (y&lt;sub&gt;i&lt;/sub&gt; - &amp;mu;&lt;sub&gt;y&lt;/sub&gt;) ) / N. It is a generalisation of the {{variance}} of a single variable. It measures how large is the common variability of the two variables; if they are unrelated it is near zero. For example, if you measured the heights and weights of people you would expect the covariance to be positive. The size of the covariance is also influenced by the level of variability of the two individual variables: for example, the covariance between length and weight of sparrows would be a lot smaller than the same calculation for whales. The {{correlation coefficient}} corrects for this by dividing by the product of the {{standard deviation}}&nbsp;of the two variables.
            [has_links] =&gt; 
        )

    [critical value] =&gt; Array
        (
            [slug] =&gt; critical value
            [name] =&gt; critical value
            [canonical] =&gt; critical values
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [critical values] =&gt; Array
        (
            [slug] =&gt; critical values
            [name] =&gt; critical values
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In the days before computers, when all calculations were performed by hand, and even when computers were less powerful than they are today, you would check the {{significance level}} of a result by looking up the value in a table that listed the critical value of a given statistic for 5%, 1%, etc. For example, the critical value for the {{Student's t distribution}} with 10 {{degrees of freedom}} at a 5% significance level is 2.23. Suppose we performed a t test comparing the heights of 6 people from Cardiff and 6 from London (df = 6+6-2 = 10): if the t value were greater than this critical value of 2.23 we would conclude that London and Cardiff heights were different at the 5% significance level.
            [has_links] =&gt; 
        )

    [data analyst] =&gt; Array
        (
            [slug] =&gt; data analyst
            [name] =&gt; data analyst
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The person who is performing the analysis of data, typically a professional such as a researcher or data scientist. The outcomes of the analysis may often be summarised or visualised to make them more comprehensible for the {{data consumer}}.
            [has_links] =&gt; 
        )

    [data consumer] =&gt; Array
        (
            [slug] =&gt; data consumer
            [name] =&gt; data consumer
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The person using some form of data collected, or visualised: for example, a client, management, the general public, or an academic reading an illustrated article.
            [has_links] =&gt; 
        )

    [data journals] =&gt; Array
        (
            [slug] =&gt; data journals
            [name] =&gt; data journals
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; These are academic journals that primarily publish documented data. This is in contrast to traditional journals, which may publish data, but only as some form of addendum to a paper that is primarily about the analysis and interpretation of the data.
            [has_links] =&gt; 
        )

    [day effects] =&gt; Array
        (
            [slug] =&gt; day effects
            [name] =&gt; day effects
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A particular kind of {{context effect}} when you do the same experiment on two different days and get different effects. This may be due, for example, to the weather or the mood of the researcher on a particular day.
            [has_links] =&gt; 
        )

    [default bayesian factors] =&gt; Array
        (
            [slug] =&gt; default bayesian factors
            [name] =&gt; default Bayesian factors
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Bayesian statistics requires some form of prior. Sometimes this is based on past data or is in the form of a researcher's beliefs {{encoded as a probability}}, but where there are no clear grounds for either of these, default Bayesian factors may be used. Typically this will be a {{uniform distribution}} over {{bounded values}}: for example, assigning a prior probability of 1/6 for each of the potential faces of a die. In the case of {{unbounded data}} the {{Cauchy distribution}} is a frequent default choice although this also requires a central value and spread parameter.
            [has_links] =&gt; 
        )

    [degrees of freedom] =&gt; Array
        (
            [slug] =&gt; degrees of freedom
            [name] =&gt; degrees of freedom
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; There are various quite complex (and mathematically precise) ways to think about degrees of freedom. However, for most purposes when analysing data it means the number of independent data items minus the number of values you have {{fitted}} to the data. For example, if you have 100 (x,y) data items and then perform a simple {{linear regression}} the software will fit a slope and intercept of the {{regression line}} &ndash; that is you have fitted two numbers, hence the remaining {{degrees of freedom}} is 100&ndash;2 = 98. &lt;br&gt;Informally (but very close to the vector space mathematics) you can think of this as having 100 pieces of randomness to start with; two of these are used up in the two fitted values, so you have 98 left.
            [has_links] =&gt; 
        )

    [demonic intervention] =&gt; Array
        (
            [slug] =&gt; demonic intervention
            [name] =&gt; demonic intervention
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Creating extreme experimental situations to {{increase effect}} size.
            [has_links] =&gt; 
        )

    [demonic interventions!] =&gt; Array
        (
            [slug] =&gt; demonic interventions!
            [name] =&gt; demonic interventions!
            [canonical] =&gt; demonic intervention
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [dependent variable] =&gt; Array
        (
            [slug] =&gt; dependent variable
            [name] =&gt; dependent variable
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The thing that is being measured and out of the direct control of the experimenter, for example response time or error rate. Usually shown on the vertical (y) axis. Contrast with {{independent variable}}.
            [has_links] =&gt; 
        )

    [design space] =&gt; Array
        (
            [slug] =&gt; design space
            [name] =&gt; design space
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The range of alternative design choices that can be combined in different ways, for example synchronous vs asynchronous, domain specific vs generic. The design space may be represented in a matrix of different semi-independent  choices, but also may be more complex.
            [has_links] =&gt; 
        )

    [dichotomous readings of hypothesis testing] =&gt; Array
        (
            [slug] =&gt; dichotomous readings of hypothesis testing
            [name] =&gt; dichotomous readings of hypothesis testing
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The simplistic binary distinction of {{significant}} vs {{non-significant}}, which stems from the days when computation was performed by hand. No longer necessary with automated computation since specific {{p-values}} are readily available.
            [has_links] =&gt; 
        )

    [discrete] =&gt; Array
        (
            [slug] =&gt; discrete
            [name] =&gt; discrete
            [canonical] =&gt; discrete data
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [discrete data] =&gt; Array
        (
            [slug] =&gt; discrete data
            [name] =&gt; discrete data
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A set of numerical data where the values are at specific points rather than anywhere along a range, for example errors: a single user may have 12 or 13 errors but cannot have 12.5. Contrast with {{continuous data}}. 
            [has_links] =&gt; 
        )

    [discrete distributions] =&gt; Array
        (
            [slug] =&gt; discrete distributions
            [name] =&gt; discrete distributions
            [canonical] =&gt; discrete data
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [discrete values] =&gt; Array
        (
            [slug] =&gt; discrete values
            [name] =&gt; discrete values
            [canonical] =&gt; discrete data
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [distractor task] =&gt; Array
        (
            [slug] =&gt; distractor task
            [name] =&gt; distractor task
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; An additional task given during an experimental situation, which aims to saturate some aspect of the user&rsquo;s cognitive abilities, so that differences in the {{cognitive load}} of the systems or conditions being studied become apparent; for example asking a subject to count backward as they perform the {{primary task}}.
            [has_links] =&gt; 
        )

    [distractor tasks] =&gt; Array
        (
            [slug] =&gt; distractor tasks
            [name] =&gt; distractor tasks
            [canonical] =&gt; distractor task
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [distribution] =&gt; Array
        (
            [slug] =&gt; distribution
            [name] =&gt; distribution
            [canonical] =&gt; 
            [see] =&gt; probability distribution
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [distributions] =&gt; Array
        (
            [slug] =&gt; distributions
            [name] =&gt; distributions
            [canonical] =&gt; distribution
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [dna evidence] =&gt; Array
        (
            [slug] =&gt; dna evidence
            [name] =&gt; DNA evidence
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Forensic method to discover whether a person is connected to a crime. Each person&rsquo;s DNA is almost unique; however, DNA testing is imperfect and has the possibility of error.
            [has_links] =&gt; 
        )

    [dna match] =&gt; Array
        (
            [slug] =&gt; dna match
            [name] =&gt; DNA match
            [canonical] =&gt; dna evidence
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [ecological validity] =&gt; Array
        (
            [slug] =&gt; ecological validity
            [name] =&gt; ecological validity
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The ecological validity of a test, study or experiment is the extent to which it is like the real world. Performing a {{controlled experiment}} in a laboratory would usually have low ecological validity, whereas an {{ethnographic study}}, observing people in their own workplace, would have high ecological validity.
            [has_links] =&gt; 
        )

    [ecologically meaningful] =&gt; Array
        (
            [slug] =&gt; ecologically meaningful
            [name] =&gt; ecologically meaningful
            [canonical] =&gt; ecological validity
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [ecologically valid] =&gt; Array
        (
            [slug] =&gt; ecologically valid
            [name] =&gt; ecologically valid
            [canonical] =&gt; ecological validity
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [effect size] =&gt; Array
        (
            [slug] =&gt; effect size
            [name] =&gt; effect size
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Effect size is how large the effect is, as opposed to how likely it is to be valid or reproducible. For example, if we had a huge data set we might be able to assess that a really tiny effect is {{statistically significant}}. It is important to give some indication of the effect size so that you can assess the importance of the result. Note that some measures called 'effect size' are actually about the ratio between the raw size of the effect and other factors. Effect size is one of the three elements in the {{noise-effect-number triangle}} that can be manipulated to increase the {{statistical power}} of a study or experiment. 
            [has_links] =&gt; 
        )

    [effective sample size] =&gt; Array
        (
            [slug] =&gt; effective sample size
            [name] =&gt; effective sample size
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; If the elements of a sample are not independent, then the raw sample size may not be the best value to use when thinking about the data. The effective size of a sample when these factors are taken into account may be smaller than the raw sample size. For example, if we asked 100 random people attending a football match which team they supported, we could use this to estimate the popularity of the teams based on a sample size of 100. However, if we had instead obtained the sample by taking 50 people each from two coaches, then the effective sample size would probably be 2 rather than 100, as it is likely that everyone on a coach supports the same team.   
            [has_links] =&gt; 
        )

    [elderly users] =&gt; Array
        (
            [slug] =&gt; elderly users
            [name] =&gt; elderly users
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Older users often have different characteristics from younger ones: for example, on average poorer eyesight or problems with short-term memory. However, this said, there is also a wide variation between different users of any age group and, of course, similarities between ages; for example, many younger people have poor eyesight.
            [has_links] =&gt; 
        )

    [emergent patterns] =&gt; Array
        (
            [slug] =&gt; emergent patterns
            [name] =&gt; emergent patterns
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Patterns that become apparent during a study and may warrant further investigation.
            [has_links] =&gt; 
        )

    [empirical] =&gt; Array
        (
            [slug] =&gt; empirical
            [name] =&gt; empirical
            [canonical] =&gt; empirical methods
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [empirical data] =&gt; Array
        (
            [slug] =&gt; empirical data
            [name] =&gt; empirical data
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Data obtained by some form of experiment or measurement of the real world.
            [has_links] =&gt; 
        )

    [empirical distribution] =&gt; Array
        (
            [slug] =&gt; empirical distribution
            [name] =&gt; empirical distribution
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A distribution derived from actual data, for example showing the number of people in each of several categories. Contrast with a {{theoretical distribution}}.
            [has_links] =&gt; 
        )

    [empirical evaluation] =&gt; Array
        (
            [slug] =&gt; empirical evaluation
            [name] =&gt; empirical evaluation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This is when a system or aspect of a system is evaluated through actual data, whether gathered in a {{laboratory experiment}} or {{in the wild}}. Alternatives would be to use some form of {{expert evaluation}} or {{theoretical analysis}}.
            [has_links] =&gt; 
        )

    [empirical methods] =&gt; Array
        (
            [slug] =&gt; empirical methods
            [name] =&gt; empirical methods
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A way of working out statistical properties, primarily used when there is no standard test. Some form of computer simulation is run, based on mathematical probabilities or observed data. By running many simulations, we can observe the likelihood of the results we have observed. 
            [has_links] =&gt; 
        )

    [encoded as a probability] =&gt; Array
        (
            [slug] =&gt; encoded as a probability
            [name] =&gt; encoded as a probability
            [canonical] =&gt; encoding as probability
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [encoding as probability] =&gt; Array
        (
            [slug] =&gt; encoding as probability
            [name] =&gt; encoding (belief) as probability
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Bayesian statistics requires a precise {{prior probability}}. Often this is obtained by quantifying a prior belief into an estimated number, which is then treated as if it is a real probability.
            [has_links] =&gt; 
        )

    [epistemologically random] =&gt; Array
        (
            [slug] =&gt; epistemologically random
            [name] =&gt; epistemologically random
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Random in one's state of knowledge. For example, if you have tossed a coin, but not yet looked at it, you would still treat it as a 50:50 odds if asked to bet, but actually it is already in one state or the other, you just don't know which. Contrast with {{ontologically random}}. 
            [has_links] =&gt; 
        )

    [error bars] =&gt; Array
        (
            [slug] =&gt; error bars
            [name] =&gt; error bars
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; An indicator used on a graph or histogram to show the the variability of the average (the {{standard error}} of the mean). This gives some idea of the accuracy of the average measure. It is usually in the form of small 'T' shaped bars extending above and below the central point on the graph.
            [has_links] =&gt; 
        )

    [error rate] =&gt; Array
        (
            [slug] =&gt; error rate
            [name] =&gt; error rate
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The error rate for a task is the frequency or number of errors that a user encounters whilst performing the task. This is one of the metrics used to evaluate the usability of a system.
            [has_links] =&gt; 
        )

    [error rates] =&gt; Array
        (
            [slug] =&gt; error rates
            [name] =&gt; error rates
            [canonical] =&gt; error rate
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [estimate] =&gt; Array
        (
            [slug] =&gt; estimate
            [name] =&gt; estimate
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In statistics, an estimate can have its ordinary day-to-day meaning: a value which is a best guess for some parameter or variable. However, it can also have a more precise meaning: an estimate calculated via some mathematical or computational method. For example, if we are interested in the average height of people in Wales, we might choose 100 people in the country at random and then work out the average of the sample. This sample average could then be used as an estimate of the overall average height of people in Wales.
            [has_links] =&gt; 
        )

    [estimate of the variability] =&gt; Array
        (
            [slug] =&gt; estimate of the variability
            [name] =&gt; estimate of the variability
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Random phenomena have {{variability}}. In some cases we know what this is based on existing literature, previous studies or theoretical calculations, but at other times we need to estimate the size of the variability, usually from data gathered in an experiment or study. For example, we might calculate the {{interquartile range}} of our measured data and use that as an estimate of the variability of the underlying phenomenon.
            [has_links] =&gt; 
        )

    [estimated standard deviation] =&gt; Array
        (
            [slug] =&gt; estimated standard deviation
            [name] =&gt; estimated standard deviation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A special case of an {{estimate of the variability}} of data is the {{estimated standard deviation}}. For example, if we know that data follows the {{Normal distribution}} (say people's heights), we might work out the {{standard deviation of the population}} for a sample, and then use this as an estimate of the true {{standard deviation}}.
            [has_links] =&gt; 
        )

    [ethnographic methods] =&gt; Array
        (
            [slug] =&gt; ethnographic methods
            [name] =&gt; ethnographic methods
            [canonical] =&gt; ethnography
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [ethnographic study] =&gt; Array
        (
            [slug] =&gt; ethnographic study
            [name] =&gt; ethnographic study
            [canonical] =&gt; ethnography
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [ethnography] =&gt; Array
        (
            [slug] =&gt; ethnography
            [name] =&gt; ethnography
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A form of {{qualitative research}} based strongly on observation of real situations. A particular focus of {{ethnography}} is to understand a situation from the point of view of the people involved, to become encultured into their patterns of thought, rather than to impose one's own models, theories or values. Experienced ethnographers typically resist any attempt to claim {{generalisability}} of their work, emphasising the unique nature of each situation and context. However, in practice they are very good at bringing out aspects of the situation that inform similar situations elsewhere.
            [has_links] =&gt; 
        )

    [experimental error] =&gt; Array
        (
            [slug] =&gt; experimental error
            [name] =&gt; experimental error
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; No interesting experiment gives a precise and unambiguous result; there is always a margin of error. For example, if you sample 100 people out of the workforce of a large company, by chance you might choose those who are unusually tall or short, or those who are unusually happy or unhappy with the systems they use. Furthermore, there may be limits to the accuracy of the measurement devices you use, although this is more of a problem in physical rather than human sciences. Where the error is due to random effects, and these are understood, you may be able to estimate the margin of error, for example the {{standard error}} of the mean.
            [has_links] =&gt; 
        )

    [experimenter effects] =&gt; Array
        (
            [slug] =&gt; experimenter effects
            [name] =&gt; experimenter effects
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A particular type of {{context effect}} where something about the experimenter (for example their mood) affects the responses of the subjects to the tasks.
            [has_links] =&gt; 
        )

    [expert evaluation] =&gt; Array
        (
            [slug] =&gt; expert evaluation
            [name] =&gt; expert evaluation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Evaluation carried out by a usability expert rather than an actual user. These are often used for early assessment of a system to avoid the costs of a full {{usability evaluation}}. Note that the term 'expert' here refers to usability or UX expertise, not expertise in the domain of the system. Examples of expert evaluation include {{cognitive walkthrough}} and {{heuristic evaluation}}.
            [has_links] =&gt; 
        )

    [expert slip] =&gt; Array
        (
            [slug] =&gt; expert slip
            [name] =&gt; expert slip
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; An expert slip is a form of error that is more frequently performed by an expert. Whilst novices may make errors due to lack of knowledge or practice, expert slips are often about performing tasks quickly or with less attention. They are typically infrequent, but may be more damaging and may be noticed far later as experts do not usually need to check the results of their actions as often as novices.
            [has_links] =&gt; 
        )

    [explanation] =&gt; Array
        (
            [slug] =&gt; explanation
            [name] =&gt; explanation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; By analysing data, visualising it, calculating various measures or performing statistical tests, we can establish that a phenomenon occurs, &lt;emph&gt;what&lt;/emph&gt; is true. &lt;strong&gt;Explanation&lt;/strong&gt; tries to work out &lt;emph&gt;why&lt;/emph&gt; it happens and &lt;emph&gt;how&lt;/emph&gt; it works &amp;ndash; deep understanding.
            [has_links] =&gt; 
        )

    [explanation stage] =&gt; Array
        (
            [slug] =&gt; explanation stage
            [name] =&gt; explanation stage
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; One of three main uses of empirical work during research, which often relate to the stages or goals of a research project. The other two are {{exploration}} and {{validation}}. The explanation stage is about working out why an observation is true, and how it happens in detail. The method may be theory based or qualitative and helps to establish {{mechanism}}.
            [has_links] =&gt; 
        )

    [exploration stage] =&gt; Array
        (
            [slug] =&gt; exploration stage
            [name] =&gt; exploration stage
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; One of three main uses of empirical work during research, which often relate to the stages or goals of a research project. The other two are {{explanation}} and {{validation}}. Exploration is principally about identifying the questions you want to ask. Techniques for exploration are often open-ended. They may be {{qualitative}}: {{ethnography}}, {{in-depth interviews}}, or detailed {{observation of behaviour}} whether {{in the lab}} or {{in the wild}}. However, this is also a stage that might involve (relatively) {{big data}}, for example if you have deployed software with logging, or have conducted a large-scale but open-ended {{survey}}. Data analysis may then be used to uncover patterns, which in turn suggest research questions. Note, you may not need this as a stage of research if you are beginning with an existing hypothesis, perhaps from previous phases of your own research, questions arising from other published work, or based on your own experiences.
            [has_links] =&gt; 
        )

    [external non-independence] =&gt; Array
        (
            [slug] =&gt; external non-independence
            [name] =&gt; external non-independence
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; When the choice of subjects in a study is connected to the topic being studied: an obvious example might be surveying preferences for MacOS or Windows in an Apple Store; less obvious might be using a mobile app-based survey on a topic that is likely to be age related. Overlooking external non-independence may lead to unintentional bias in results.
            [has_links] =&gt; 
        )

    [extrapolation] =&gt; Array
        (
            [slug] =&gt; extrapolation
            [name] =&gt; extrapolation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Using results that are in a studied range to make predictions about potential effects of factors outside the range.
            [has_links] =&gt; 
        )

    [extreme statistics] =&gt; Array
        (
            [slug] =&gt; extreme statistics
            [name] =&gt; extreme statistics
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Extreme measures such as {{maximum}} and {{minimum}} are affected greatly by a small number of values. One consequence of this is that they have relatively complex mathematics. More importantly, they are not {{robust}}, that is they tend to vary a lot and if you do the same study or experiment again you are likely to get very different results. In contrast, measures such as the {{mean}}, {{median}} or {{quartiles}} tend to be far more stable.
            [has_links] =&gt; 
        )

    [extreme values] =&gt; Array
        (
            [slug] =&gt; extreme values
            [name] =&gt; extreme values
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This term is used in two senses in the book. &lt;br&gt;The first describes anomalous outcomes in your data. For example, if 10 coins are tossed there is about a 1 in 50 chance of getting only 1 head or 1 tail. Really extreme values in results are relatively unlikely. If they are unexpected they are called {{outliers}}, and may arise by chance or from faulty equipment; if they are valid data points but just happen to be extreme, they are part of the random nature of the phenomenon being studied. &lt;br&gt; The second usage refers to the various parameters in a distribution or model. Even if you expect real values to be more central, experimenting ... playing! ... with more extreme values can often give you a sense of the data. For example, experiments where you toss (virtual) coins with very high bias or coin-to-coin correlation.
            [has_links] =&gt; 1
        )

    [factor effects] =&gt; Array
        (
            [slug] =&gt; factor effects
            [name] =&gt; factor effects
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; When there is some form of relationship or correlation between the various factors of which you are measuring aspects. Such factors are called {{confounding variables}}. For example, preferred social media is likely to be related to musical taste due to the different age groups using different applications.
            [has_links] =&gt; 
        )

    [factorial] =&gt; Array
        (
            [slug] =&gt; factorial
            [name] =&gt; factorial
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The product of all the numbers from 1 up to a given number. The notation is [given number]!, for example 6! (6 factorial) is 6x5x4x3x2x1.
            [has_links] =&gt; 
        )

    [fair coin] =&gt; Array
        (
            [slug] =&gt; fair coin
            [name] =&gt; fair coin
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A coin where the probability of throwing a head or a tail is the same, 0.5. 
            [has_links] =&gt; 
        )

    [fair estimate] =&gt; Array
        (
            [slug] =&gt; fair estimate
            [name] =&gt; fair estimate
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This is the probability of one event (say A) &lt;emph&gt;given&lt;emp&gt; knowledge of another event (B). It is usually written as P(A|B), read as &quot;the probability of A given B&quot;. For example, while the probability that it will rain on a typical day might be 50%, on a cloudy day this might be 80%; that is P(rain|cloudy) = 0.8.
            [has_links] =&gt; 
        )

    [false baseline] =&gt; Array
        (
            [slug] =&gt; false baseline
            [name] =&gt; false baseline
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; When the scale of a {{baseline}} starts at a value other than zero. This can be very useful where the range of values would make changes in a graph difficult to see, but without care it can also be misleading, making insignificant differences appear large and exaggerating effects.
            [has_links] =&gt; 
        )

    [false baselines] =&gt; Array
        (
            [slug] =&gt; false baselines
            [name] =&gt; false baselines
            [canonical] =&gt; false baseline
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [false basepoint] =&gt; Array
        (
            [slug] =&gt; false basepoint
            [name] =&gt; false basepoint
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; When the x-axis of a scale of a {{basepoint}} starts at a value other than zero. This is common for time series &ndash; what time do you start from? It can be useful for highlighting changes, but may also be misleading sometimes.
            [has_links] =&gt; 
        )

    [false negative] =&gt; Array
        (
            [slug] =&gt; false negative
            [name] =&gt; false negative
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Where there may be a real effect, but an experiment or study is not sensitive enough to detect it. {{Statistical power}} is about avoiding false negatives. A false negative conclusion is known in statistics as a {{Type II error}}. Contrast with a {{false positive}}.
            [has_links] =&gt; 
        )

    [false negatives] =&gt; Array
        (
            [slug] =&gt; false negatives
            [name] =&gt; false negatives
            [canonical] =&gt; false negative
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [false origin] =&gt; Array
        (
            [slug] =&gt; false origin
            [name] =&gt; false origin
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Where the start point of a graph is not the &lsquo;true origin&rsquo; of the data (0,0), but starts at some other value on one or both axes. The {{baseline}} is the y coordinate of the false origin and the {{basepoint}} is the x coordinate. There are often good reasons for choosing a false origin, {{false baseline}} or {{false basepoint}} in order to explore or demonstrate a small effect, but without care this can also lead to misinterpretation or even be used for deliberate misinformation.
            [has_links] =&gt; 
        )

    [false positive] =&gt; Array
        (
            [slug] =&gt; false positive
            [name] =&gt; false positive
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This is where you wrongly believe something to be true. In traditional {{significance testing}} the choice of the {{significance level}} is precisely the probability of a false positive. For example, a p&lt;5% means that there is a 1 in 20 chance of getting a false positive result. Note, though, that this only applies to the raw results; if some results are not published (the {{file drawer effect}}) the false positive result of published results may be higher. A false positive conclusion is known in statistics as a {{Type I error}}. Contrast with a {{false negative}}.
            [has_links] =&gt; 
        )

    [false positive result] =&gt; Array
        (
            [slug] =&gt; false positive result
            [name] =&gt; false positive result
            [canonical] =&gt; false positive
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [false positives] =&gt; Array
        (
            [slug] =&gt; false positives
            [name] =&gt; false positives
            [canonical] =&gt; false positive
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [feedback effects] =&gt; Array
        (
            [slug] =&gt; feedback effects
            [name] =&gt; feedback effects
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Where the outputs or past results of a process or phenomenon act as inputs to its future behaviour. These effects can be positive or negative. Snowflake growth is an example of {{positive feedback}} where any small growth grows further. In contrast, a negative feedback process smooths out changes, for example any protuberances on a pebble tend to be eroded more as it is turned over on the seashore, so that the pebble becomes rounded over time. Feedback effects may cause {{nonlinearity}, because some initial change can make huge differences.
            [has_links] =&gt; 
        )

    [file drawer effect] =&gt; Array
        (
            [slug] =&gt; file drawer effect
            [name] =&gt; file drawer effect
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; For every {{positive result}} published there are many more {{negative results}} ({{non-significant}}) that are withheld. Such selective publishing may be a deliberate attempt to deceive, but is more often a combination of ignorance and {{bias}}. This is exacerbated by journal reviewing practices that typically make it hard or impossible to publish {{non-significant results}}.
            [has_links] =&gt; 
        )

    [file tree] =&gt; Array
        (
            [slug] =&gt; file tree
            [name] =&gt; file tree
            [canonical] =&gt; file tree visualisation
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [file tree visualisation] =&gt; Array
        (
            [slug] =&gt; file tree visualisation
            [name] =&gt; file tree visualisation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This is an outline-style {{visualisation}} similar to that found in many operating system file-system views. A similar visualisation can be used for other forms of {{hierarchical data}}.
            [has_links] =&gt; 
        )

    [fine positioning tasks] =&gt; Array
        (
            [slug] =&gt; fine positioning tasks
            [name] =&gt; fine positioning tasks
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A task requiring fine motor skills and direct gaze, for example selecting a small target with an on-screen pointer.
            [has_links] =&gt; 
        )

    [finite data] =&gt; Array
        (
            [slug] =&gt; finite data
            [name] =&gt; finite data
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Numerical data that is {{bounded}} at both ends of its range of values.
            [has_links] =&gt; 
        )

    [finite variance] =&gt; Array
        (
            [slug] =&gt; finite variance
            [name] =&gt; finite variance
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; For any real set of data the {{variance}} is bound to be a finite number, but for some {{theoretical distributions}}, notably {{power-law distributions}}, if you mathematically calculate variance it is infinite. This does impact practical statistics. Normally if you take larger and larger samples the measured variances will nearly always shift closer and closer to some finite value. However, if the underlying distribution is not finite, for example the number of contacts in a social network, then the measured variances of larger and larger samples will tend to become larger and larger. 
            [has_links] =&gt; 
        )

    [fishing expedition] =&gt; Array
        (
            [slug] =&gt; fishing expedition
            [name] =&gt; fishing expedition
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This is where, consciously or unconsciously, you look through data or results to find the ones that are particularly favourable.
            [has_links] =&gt; 
        )

    [fitts' law] =&gt; Array
        (
            [slug] =&gt; fitts' law
            [name] =&gt; Fitts' Law
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; {{Paul Fitts}} found that, for simple pointing tasks, {{task completion time}} is proportional to the {{Index of Difficulty (IoD)}}, which is essentially the logarithm of the distance to target divided by the target size: IoD = log2 ( distance to target/target size ). While Fitts' original research was on physical one-dimensional pointing alternating between two metal strips, Fitts&rsquo; Law has been found to hold for many kinds of pointing tasks, with a wide variety of devices, over multiple orders of magnitude. Although Fitts&rsquo; Law holds in many situations it is not universal; crucially it depends on situations where there is {{hand&ndash;eye coordination}} and where movement error is proportional to distance moved. 
            [has_links] =&gt; 1
        )

    [five users] =&gt; Array
        (
            [slug] =&gt; five users
            [name] =&gt; five users (the myth)
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Often it is said that five users are enough for {{usability testing}}. This had some truth in the particular original context of {{iterative software development}}, where only enough issues needed to be uncovered for the next cycle of re-development. However, it is certainly not sufficient for statistical purposes.
            [has_links] =&gt; 1
        )

    [five users are enough] =&gt; Array
        (
            [slug] =&gt; five users are enough
            [name] =&gt; five users are enough
            [canonical] =&gt; five users
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [fixed effect] =&gt; Array
        (
            [slug] =&gt; fixed effect
            [name] =&gt; fixed effect
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A fixed effect is one that you choose, which would stay the same if you repeated the study. It is typically something the researcher wants to find out about. Contrast with a {{random effect}}. 
            [has_links] =&gt; 
        )

    [font size] =&gt; Array
        (
            [slug] =&gt; font size
            [name] =&gt; font size
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The size of characters in a {{font}} is usually measured in points, where a printer's point is 1/72 of an inch. So the tall characters in a &lt;span style=&quot;font-size:12pt&quot;&gt;12pt font&lt;/span&gt; are around 1/6 of an inch high, whereas a &lt;span style=&quot;font-size:14pt&quot;&gt;14pt font&lt;/span&gt; is about 1/5 of an inch and &lt;span style=&quot;font-size:7pt&quot;&gt;7pt font&lt;/span&gt; is about 1/10 of an inch. However, the exact size does vary between fonts. On screen the physical size of pixels (the {{screen resolution}}) varies. Old displays often used to have around 70 pixels per inch, so a point and a pixel were nearly the same, but now that varies a lot more, particularly for phone displays and retina displays. The device tries to choose a pixel-height that closely matches the paper dimensions of the desired font. If the precise screen size of characters matters, the font size may be measured in pixels.
            [has_links] =&gt; 
        )

    [font sizes] =&gt; Array
        (
            [slug] =&gt; font sizes
            [name] =&gt; font sizes
            [canonical] =&gt; font size
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [formalist] =&gt; Array
        (
            [slug] =&gt; formalist
            [name] =&gt; formalist
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A pragmatic way of thinking about probability: it doesn&rsquo;t matter what probability &lsquo;really&rsquo; is, so long as it satisfies the right mathematical rules. In particular, {{Bayesian statistics}} encodes beliefs as values between 0 and 1, which satisfy the rules of probability ({{Kolomogrov laws}}). This {{belief encoded as probability}} is sometimes called {{plausibility}} or {{reasonable expectation}}). Contrast with {{frequentist}} and {{idealist}}. 

            [has_links] =&gt; 
        )

    [formative assessment] =&gt; Array
        (
            [slug] =&gt; formative assessment
            [name] =&gt; formative assessment
            [canonical] =&gt; formative evaluation
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [formative evaluation] =&gt; Array
        (
            [slug] =&gt; formative evaluation
            [name] =&gt; formative evaluation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Focused on making the system better, formative evaluation is performed on the design artefacts (sketch, prototype, or experimental system) during the cycles of design&ndash;build&ndash;
test. Its form varies from {{expert evaluation}} to large-scale {{user testing}}, and its primary purpose is to uncover usability or experience problems for the next cycle. Contrast with {{summative evaluation}}.
            [has_links] =&gt; 
        )

    [formative evaluations] =&gt; Array
        (
            [slug] =&gt; formative evaluations
            [name] =&gt; formative evaluations
            [canonical] =&gt; formative evaluation
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [formative stages] =&gt; Array
        (
            [slug] =&gt; formative stages
            [name] =&gt; formative stages
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Stages of {{software development}} where {{formative evaluation}} is carried out.
            [has_links] =&gt; 
        )

    [frequentist] =&gt; Array
        (
            [slug] =&gt; frequentist
            [name] =&gt; frequentist
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A down-to-earth way of looking at statistics: if a frequentist says the chance of a coin landing heads is 50:50, they mean that if they keep on tossing the coin again and again and again, on average, after many many tosses, the ratio of heads to tails will be about 50:50. Contrast with {{formalist}} and {{idealist}}.
            [has_links] =&gt; 
        )

    [gaussian distribution] =&gt; Array
        (
            [slug] =&gt; gaussian distribution
            [name] =&gt; Gaussian distribution
            [canonical] =&gt; 
            [see] =&gt; normal distribution
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [generative artefact] =&gt; Array
        (
            [slug] =&gt; generative artefact
            [name] =&gt; generative artefact
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A thing, such as a toolkit, design methodology or property, that is used to create or to generate other things. For example, with properties we may try to design a system that is synchronous, or one with consistency. Typically there are many systems that satisfy the property (or could be produced using the toolkit or methodology), and if we asked different practitioners, each one would take the same property and create a different final system. This makes it very hard to {{evaluate}} a generative artefact, but that does not mean it is impossible to {{validate}} them.
            [has_links] =&gt; 
        )

    [generative artefacts] =&gt; Array
        (
            [slug] =&gt; generative artefacts
            [name] =&gt; generative artefacts
            [canonical] =&gt; generative artefact
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [good-enough solution] =&gt; Array
        (
            [slug] =&gt; good-enough solution
            [name] =&gt; good-enough solution
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The point in software development where costs start to outweigh any benefits gained by continuing to refine the system.
            [has_links] =&gt; 
        )

    [h_0] =&gt; Array
        (
            [slug] =&gt; h_0
            [name] =&gt; H&lt;sub&gt;0&lt;/sub&gt;
            [canonical] =&gt; 
            [see] =&gt; null hypothesis
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [h_1] =&gt; Array
        (
            [slug] =&gt; h_1
            [name] =&gt; H&lt;sub&gt;1&lt;/sub&gt;
            [canonical] =&gt; 
            [see] =&gt; alternative hypothesis
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [hark] =&gt; Array
        (
            [slug] =&gt; hark
            [name] =&gt; HARK
            [canonical] =&gt; 
            [see] =&gt; hypothesis after the results are known
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [harking] =&gt; Array
        (
            [slug] =&gt; harking
            [name] =&gt; HARKing
            [canonical] =&gt; 
            [see] =&gt; hypothesis after the results are known
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [hawthorne effect] =&gt; Array
        (
            [slug] =&gt; hawthorne effect
            [name] =&gt; Hawthorne effect
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This when subjects in a study or experiment behave differently because they know they are being observed.  For example in an educational or work setting, subjects may pay more attention to a task when they know they are being watched and recorded.  This may simply life or depress oerall performace, but is particlarly problematic when it casues {{bias}}, in particular, if subjects are aware that the pwerson organising the study or observing the study has an interest in a particular outcome.  In {{user studies}} this often occurs when new software is being trialed.  Subjects may try harder at the favoured system, or give better subjcetive assessments, just to be nice.
            [has_links] =&gt; 1
        )

    [hci] =&gt; Array
        (
            [slug] =&gt; hci
            [name] =&gt; HCI
            [canonical] =&gt; 
            [see] =&gt; human-computer interaction
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [hci studies] =&gt; Array
        (
            [slug] =&gt; hci studies
            [name] =&gt; HCI studies
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Research in {{HCI}} often involves some form of study with people: sometimes using a {{prototype}}, sometimes in some form of {{focus group}} or {{interview}}, sometimes in a more experimental situation closer to pure psychology. The methods used are often similar to the {{user studies}} or {{user experience studies}} used by {{usability practitioners}} during {{user interface development}}, but the aims are different.
            [has_links] =&gt; 
        )

    [human-computer interaction] =&gt; Array
        (
            [slug] =&gt; human-computer interaction
            [name] =&gt; Human&amp;ndash;Computer Interaction
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This is the discipline focused on understanding of the way people use and are impacted by computer technology and the way that technology can be designed to better serve human needs. This includes single users interacting directly with devices, but also the impact of computers at the level of work groups, families, organisations and communities.
            [has_links] =&gt; 1
        )

    [hypothesis after the results are known] =&gt; Array
        (
            [slug] =&gt; hypothesis after the results are known
            [name] =&gt; hypothesis after the results are known
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; When you notice something interesting in data it is tempting to do a statistical test of this. However, there is a clear {{selection bias}}, by definition one only notices patterns that are in the data and it is therefore not so surprising that these patterns emerge as (apparently) statistically significant.  One viewpoint is that one should *never* HARK, instead use the observation as a basis to frame a {{hypothesis}} for a fresh study or experiment. However, soemtimes this is not possible or practical, for example when analysing a pre-existing or very hard to replicate dataset.  In such cases one has to be very careful to assess how nay other patterns would have been be equally or more surprising and factor this into your results.. 
            [has_links] =&gt; 1
        )

    [hypothesis testing] =&gt; Array
        (
            [slug] =&gt; hypothesis testing
            [name] =&gt; hypothesis testing
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; You start with a number of predetermined options (hypotheses) and then use statistical techniques to determine which of them has more or less evidence for being true. There are forms of hypothesis testing used in {{Bayesian statistics}}, but the term is more usually associated with {{traditional statistics}}, and in particular with what is sometimes called {{null hypothesis significance testing}} or {{NHST}}. Here you start with a {{null hypothesis}} (usually what you would like to be false, such as your system not being any better) and have an {{alternative hypothesis}} (typically what you would like to be true, such as your system being better!). You then use a {{significance test}} to see if the data is consistent with the null hypothesis, and if not you 'reject' the null and instead conclude that the alternative is likely to be true.
            [has_links] =&gt; 
        )

    [idealist] =&gt; Array
        (
            [slug] =&gt; idealist
            [name] =&gt; idealist
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This way of thinking about statistics is based on the philosophy of Plato, who saw the actual events of the world as mere reflections of deeper ideals. The toss of the actual coin in some ways is &lsquo;just&rsquo; an example of an {{ideal coin toss}}. If an idealist tosses a coin five times in a row, even if it happens to come up heads each time, they probably still believe that it is &lsquo;really&rsquo; a 50:50 phenomenon. Contrast with {{formalist}} and {{frequentist}}.
            [has_links] =&gt; 
        )

    [image recognition] =&gt; Array
        (
            [slug] =&gt; image recognition
            [name] =&gt; image recognition
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Software that uses algorithms to interpret image data, either static pictures or video. For example, it might be used to recognise faces in a picture so that they can be more easily labelled, or to recognise hand gestures using the computer's camera. Before {{GPS}} was common in phones, some applications would match the image on the phone with known landmarks to enable  location-specific services such as historical information or geo annotation. Similar techniques are also common in {{augmented reality}} systems that overlay digital imagery over a live feed of the real world. The accuracy of the image recognition can affect the {{user experience}}.
            [has_links] =&gt; 
        )

    [in the wild] =&gt; Array
        (
            [slug] =&gt; in the wild
            [name] =&gt; in the wild
            [canonical] =&gt; in-the-wild
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [in-the-wild] =&gt; Array
        (
            [slug] =&gt; in-the-wild
            [name] =&gt; in-the-wild
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; An {{ethnographic study}} is carried out in the normal context of the subjects of the study, for example observing people in their own workplace. It is a form of {{in the wild}} study; these have high {{ecological validity}} but hold their own specific challenges, such as {{replication}} and control of {{variability}}.
            [has_links] =&gt; 1
        )

    [in-the-wild studies] =&gt; Array
        (
            [slug] =&gt; in-the-wild studies
            [name] =&gt; in-the-wild studies
            [canonical] =&gt; in-the-wild
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [income distribution] =&gt; Array
        (
            [slug] =&gt; income distribution
            [name] =&gt; income distribution
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Income distribution data is an example of {{continuous data}} (weekly income may be a fraction of the base currency amount), an {{empirical distribution}} (it is derived from real data, not a theoretical model), {{bounded below}} (there are no negative incomes), {{unbounded above}} (although there is always some maximum income at any point in time, there is often no fundamental limit), an {{asymmetric distribution}} (the shape is very skewed with a large number of smaller incomes and small number of larger ones ), and a {{long-tail distribution}} (there are a small number of extremely large incomes). Furthermore, for practical purposes it effectively has {{infinite variance}} (larger and larger samples typically have larger and larger variances).
            [has_links] =&gt; 
        )

    [income distributions] =&gt; Array
        (
            [slug] =&gt; income distributions
            [name] =&gt; income distributions
            [canonical] =&gt; income distribution
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [increase effect] =&gt; Array
        (
            [slug] =&gt; increase effect
            [name] =&gt; increase effect
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; There are several strategies to gain {{statistical power}} by manipulating the sensitivity of a study. One way is to attempt to increase the size of the measured effect using techniques such as a {{targeted user group}}, {{distractor tasks}}, {{targeted tasks}} and {{demonic interventions}}. However, care must be taken not to mislead with these techniques as they typically have low generalisability. See the {{noise-effect-number triangle}} for other ways to increase {{statistical power}}.
            [has_links] =&gt; 
        )

    [increase number] =&gt; Array
        (
            [slug] =&gt; increase number
            [name] =&gt; increase number
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The standard approach to gain {{statistical power}}. Used on its own this usually requires a very large increase in the number of subjects or trials. In a more nuanced approach researchers can also {{increase effects}} and {{reduce noise}}, the other two parts of the {{noise-effect-number triangle}}.
            [has_links] =&gt; 
        )

    [independence] =&gt; Array
        (
            [slug] =&gt; independence
            [name] =&gt; independence
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Broadly, when one measurement or factor does not give information about another. There are several kinds of independence: {{independence of measurements}}, {{independence of factor effects}}, and {{independence of sample composition}}. Contrast with {{non-independence}}.

            [has_links] =&gt; 
        )

    [independence condition] =&gt; Array
        (
            [slug] =&gt; independence condition
            [name] =&gt; independence condition
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; One of the conditions for the {{Central Limit Theorem}}, which determines whether the average of large numbers of small things ends up being {{approximately Normal}}. The things being averaged or added up need to be nearly independent, on order that one of them does not become overwhelming as part of the whole.
            [has_links] =&gt; 
        )

    [independent] =&gt; Array
        (
            [slug] =&gt; independent
            [name] =&gt; independent
            [canonical] =&gt; independence
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [independent evidence] =&gt; Array
        (
            [slug] =&gt; independent evidence
            [name] =&gt; independent evidence
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This is where elements of evidence (for example data or measurements) are gathered from entirely distinct sources, or where they refer to aspects of the same data source that are not in any way related. For example, we might perform a user study on a new piece of educational software in two schools in completely different cities. It is far easier to reason about independent evidence than it is when there is some form of connection between the evidence. For example, if the new software was tested on two classes in the same school, the second class might have heard about it from their friends in the first class and so their satisfaction scores might merely reflect the views of the first class. &lt;br&gt;This can be a particular problem in {{Bayesian statistics}} if the {{prior probability}} is based on multiple evidence, some of which have been influenced by a common source. It is possible to attempt to model this common factor, but that can be complex.
            [has_links] =&gt; 
        )

    [independent variable] =&gt; Array
        (
            [slug] =&gt; independent variable
            [name] =&gt; independent variable
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The thing that is being varied experimentally, for example distance to target, or is given by the world, for example a date. Usually shown on the horizontal (x) axis. Contrast with the {{dependent variable}}, the thing that is being measured. For example, in a {{pointing task}} experiment, the independent variables might be the size of a target and the distance to it, whereas the dependent variables would be the time to reach the target and the {{error rate}}.
            [has_links] =&gt; 
        )

    [index of difficulty] =&gt; Array
        (
            [slug] =&gt; index of difficulty
            [name] =&gt; index of difficulty (IoD)
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The logarithm of the distance to target divided by the target size: IoD = log&lt;sub&gt;2&lt;/sub&gt; ( distance to target/target size ). Used in {{Fitts' Law}}.
            [has_links] =&gt; 
        )

    [index of difficulty (iod)] =&gt; Array
        (
            [slug] =&gt; index of difficulty (iod)
            [name] =&gt; index of difficulty (iod)
            [canonical] =&gt; index of difficulty
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [individual differences] =&gt; Array
        (
            [slug] =&gt; individual differences
            [name] =&gt; individual differences
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Differences between subjects that may influence the effect being studied. For some studies these individual differences can mask the thing being examined (create {{noise}}). Ways to reduce the impact of individual differences in such situations include using a very {{narrow user group}}, having the same users perform on different systems (a {{within-subjects design}}), and choosing participants of the same age and experience to {{reduce noise}}. Each of these has implications for the {{validity}} of results. Within {{interaction design}} and {{user experience design}} it is crucial to remember that people are different. Sometimes you have to create a 'one size fits all' application, but if you can understand differences between users this can lead to better designs, whether different applications targeted at different user groups, {{customisation}} options within a single application, or {{intelligent interfaces}} that adapt themselves to the user.
            [has_links] =&gt; 
        )

    [infinite variance] =&gt; Array
        (
            [slug] =&gt; infinite variance
            [name] =&gt; infinite variance
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; For any real set of data the {{variance}} is bound to be a finite number, but for some {{theoretical distributions}}, notably {{power-law distributions}}, if you mathematically calculate variance, it is infinite. This does impact practical statistics. Normally if you take larger and larger samples the measured variances will nearly always shift closer and closer to some finite value. However, if the underlying distribution is not finite, for example the number of contacts in a social network, then the measured variances of larger and larger samples will tend to become larger and larger. 
            [has_links] =&gt; 
        )

    [intelligent interface] =&gt; Array
        (
            [slug] =&gt; intelligent interface
            [name] =&gt; intelligent interface
            [canonical] =&gt; intelligent interfaces
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [intelligent interfaces] =&gt; Array
        (
            [slug] =&gt; intelligent interfaces
            [name] =&gt; intelligent interfaces
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; These are systems that use some form of {{artificial intelligence}} or {{machine learning}} as part of {{user interaction}}. Many of us will see examples of this daily as web sites adapt content and especially adverts according to models based on our past behaviour. More advanced systems might try to help us to complete tasks or maybe automate repeated tasks based on examples.
            [has_links] =&gt; 1
        )

    [inter-related factors] =&gt; Array
        (
            [slug] =&gt; inter-related factors
            [name] =&gt; inter-related factors
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Factors that influence the effect being studied, such as the experience of users being related to their age; also changes in a design which affect other parts of the design and make it difficult to study individual aspects.
            [has_links] =&gt; 
        )

    [interference effect] =&gt; Array
        (
            [slug] =&gt; interference effect
            [name] =&gt; interference effect
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A type of {{order effect}} where a user does less well in a second test because they have got used to the detailed way things were done in an earlier test.
            [has_links] =&gt; 
        )

    [internal non-independence] =&gt; Array
        (
            [slug] =&gt; internal non-independence
            [name] =&gt; internal non-independence
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; When subjects are likely to be similar to one another, but in no particular direction with regard to your question, for example in a survey of adjacent people in a football match queue. If they all support the same team it might simply mean that they all travelled on the same bus.
            [has_links] =&gt; 
        )

    [interpolate] =&gt; Array
        (
            [slug] =&gt; interpolate
            [name] =&gt; interpolate
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Inferring information about values between the available measured values, for example an  assumption that the reading speed of an 11-point font falls between the reading speed data you have for the 10-point and 12-point variants. Contrast with {{extrapolation}}.
            [has_links] =&gt; 
        )

    [interquartile range] =&gt; Array
        (
            [slug] =&gt; interquartile range
            [name] =&gt; interquartile range
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This is a measure of {{variation}} calculated as the difference between the {{upper quartile}} and {{lower quartile}}. That is, half of the data items will fall within the interquartile range. It is typically a far more {{robust}} measure than the more common {{standard deviation}}, but less tractable for calculations.
            [has_links] =&gt; 
        )

    [interval] =&gt; Array
        (
            [slug] =&gt; interval
            [name] =&gt; interval
            [canonical] =&gt; interval data
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [interval data] =&gt; Array
        (
            [slug] =&gt; interval data
            [name] =&gt; interval data
            [canonical] =&gt; arithmetical data
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [iod] =&gt; Array
        (
            [slug] =&gt; iod
            [name] =&gt; iod
            [canonical] =&gt; index of difficulty (iod)
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [iterative development] =&gt; Array
        (
            [slug] =&gt; iterative development
            [name] =&gt; iterative development
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A software design cycle that includes repeated stages of evaluation of evolving designs, prototypes or systems. The two main kinds of evaluation are {{formative evaluation}} at each stage to improve the design and {{summative evaluation}} at the end to assess whether the final design is sufficiently good.
            [has_links] =&gt; 
        )

    [iterative evaluation] =&gt; Array
        (
            [slug] =&gt; iterative evaluation
            [name] =&gt; iterative evaluation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The {{formative evaluation}} in {{iterative development}}
            [has_links] =&gt; 
        )

    [iterative software development] =&gt; Array
        (
            [slug] =&gt; iterative software development
            [name] =&gt; iterative software development
            [canonical] =&gt; iterative development
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [job of statistics] =&gt; Array
        (
            [slug] =&gt; job of statistics
            [name] =&gt; job of statistics
            [canonical] =&gt; 
            [see] =&gt; the job of statistics
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [kolomogrov laws] =&gt; Array
        (
            [slug] =&gt; kolomogrov laws
            [name] =&gt; Kolomogrov laws
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A collection of three axioms developed by {{Andrey Kolmogorov}} in 1933, which summarise the laws of {{probability}}. The three axioms state that (i) the probability of any set of events is greater than or equal to zero (you can't have a negative chance of doing something); (ii) the probability of the set of all events is one (the probability that something will happen is one); and (iii) the probability of the union of disjoint sets of events is the sum of the probability of each set (probabilities add up when there is no overlap). For example, if we throw a (possibly loaded) die, the sets of events are subsets of {1,2,3,4,5,6}. Axiom (i) says, inter alia, that Prob({1,2}) &gt; 0 (meaning the probability of either a 1 or 2); axiom (ii) says that Prob({1,2,3,4,5,6})=1; and axiom (iii) that, inter alia, Prob( Union({1,2},{3,6}) ) =Prob( {1,2,3,6}) = Prob({1,2}) + Prob({3,6}).
            [has_links] =&gt; 1
        )

    [larger samples] =&gt; Array
        (
            [slug] =&gt; larger samples
            [name] =&gt; larger samples
            [canonical] =&gt; 
            [see] =&gt; sample size
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [learning effect] =&gt; Array
        (
            [slug] =&gt; learning effect
            [name] =&gt; learning effect
            [canonical] =&gt; learning effects
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [learning effects] =&gt; Array
        (
            [slug] =&gt; learning effects
            [name] =&gt; learning effects
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A type of {{order effect}} where a user is better at a second test because they have already got used to the general ideas of an application in the first test. 
            [has_links] =&gt; 
        )

    [likelihood] =&gt; Array
        (
            [slug] =&gt; likelihood
            [name] =&gt; likelihood
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The {{conditional probability}} of a measurement given a hypothetical state of the world (or typically some specific parameters of the world). For example, the probability that six tosses of a coin will come out heads given the coin is fair is 1/64, so if this result occurs, the likelihood that the coin is fair is 1/64. This likelihood is used in slightly different ways by {{traditional statistics}} and {{Bayesian statistics}} to work back from the measurement to assess its evidence for or against the value of the underlying property or state of the world.
            [has_links] =&gt; 
        )

    [likert scale] =&gt; Array
        (
            [slug] =&gt; likert scale
            [name] =&gt; Likert scale
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A scale to measure degrees of agreement or satisfaction. The values are often coded as numbers and there is an order but otherwise no specific meaning to the values. This is {{ordinal data}}. &lt;br&gt; An example of survey question using a Likert scale might be:&lt;br&gt; The top level menu items were:  unintelligible &ndash; 1 &ndash; 2 &ndash; 3 &ndash; 4 &ndash; 5 &ndash; always easy to interpret.
            [has_links] =&gt; 
        )

    [linear regression] =&gt; Array
        (
            [slug] =&gt; linear regression
            [name] =&gt; linear regression
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A statistical process that creates a best fit line through data points. The goodness of fit of the line is often measured using the {{Pearson correlation coefficient}}.
            [has_links] =&gt; 
        )

    [linearity condition] =&gt; Array
        (
            [slug] =&gt; linearity condition
            [name] =&gt; linearity condition
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; One of the conditions for the {{Central Limit Theorem}}. It says that the way in which lots of near-independent factors combine together must have a linear nature such as adding together lots of small factors, or averaging lots of larger ones.
            [has_links] =&gt; 
        )

    [log-normal distribution] =&gt; Array
        (
            [slug] =&gt; log-normal distribution
            [name] =&gt; Log-Normal distribution
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This distribution is common where there is some form of multiplicative effect: for example, with {{task completion times}} we would expect longer times to have greater variability. Tests designed for the {{Normal distribution}} cannot be used, but the data can be transformed by taking the logarithm of the raw times, then the resulting {{transformed data}} is often {{approximately Normal}} and can be analysed using off-the-shelf statistics.
            [has_links] =&gt; 
        )

    [long tail] =&gt; Array
        (
            [slug] =&gt; long tail
            [name] =&gt; long tail
            [canonical] =&gt; long-tail distribution
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [long-tail distribution] =&gt; Array
        (
            [slug] =&gt; long-tail distribution
            [name] =&gt; long-tail distribution
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A {{distribution}} where there are very high (or very low) values but they occur with very low frequency. Examples include income distributions (small numbers of very rich individuals) and numbers of contacts in {{social network data}} (a small number of people with very large numbers of 'friends'). A specific example is a {{power-law distribution}}, which is often used as a {{model}} of {{network data}}.
            [has_links] =&gt; 
        )

    [lower] =&gt; Array
        (
            [slug] =&gt; lower
            [name] =&gt; lower
            [canonical] =&gt; lower bound
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [lower bound] =&gt; Array
        (
            [slug] =&gt; lower bound
            [name] =&gt; lower bound
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This is the smallest possible value that a variable or dataset can have. For example, when rolling a die, the lower bound is 1. In contrast, some data is {{unbounded below}}, that is there is no meaningful absolute minimum.
            [has_links] =&gt; 
        )

    [lower quartile] =&gt; Array
        (
            [slug] =&gt; lower quartile
            [name] =&gt; lower quartile
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The value such that 25% (one quarter) of the data values are less and 75% more. It is one of three {{quartiles}}, the others being the {{median}} (50% point) and {{upper quartile}} (75% point).
            [has_links] =&gt; 
        )

    [machine learning] =&gt; Array
        (
            [slug] =&gt; machine learning
            [name] =&gt; machine learning
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Various algorithms that allow a computer to learn from data in a way analogous to, but typically very different from, human or animal learning. This can be used as a research tool, or as part of some form of {{intelligent interface}}. Machine learning (ML) algorithms have existed for many years, but the vast amounts of data now available allow us to study phenomena in a way that would have been impossible only a few years ago. However, the analysis of such data using machine learning techniques can lead to unintentional {{cherry picking}}, as the ML will pick out the few patterns that work amongst a vast number that do not.
            [has_links] =&gt; 
        )

    [manipulate the sensitivity] =&gt; Array
        (
            [slug] =&gt; manipulate the sensitivity
            [name] =&gt; manipulate sensitivity
            [canonical] =&gt; 
            [see] =&gt; sensitivity
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [matched users] =&gt; Array
        (
            [slug] =&gt; matched users
            [name] =&gt; matched users
            [canonical] =&gt; matching subjects
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [matching subjects] =&gt; Array
        (
            [slug] =&gt; matching subjects
            [name] =&gt; matching subjects
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; When the same subject repeats a test under different conditions (e.g. evaluating two systems) there may be {{learning effects}}. One way around this is to have different subjects for each condition, but to have two or more subjects who are very similar (in gender, age, skills), and to allocate one from each pair (or matched group) to each condition. Matching subjects makes results more robust, requiring fewer assumptions, but typically has less {{statistical power}} than a {{within-subjects experiment}}.
            [has_links] =&gt; 
        )

    [mathematics] =&gt; Array
        (
            [slug] =&gt; mathematics
            [name] =&gt; mathematics
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; There are many branches of mathematics, but within {{statistics}} the most important part is {{probability theory}}, which enables us to reason explicitly about probabilities.
            [has_links] =&gt; 
        )

    [mathematics of probability] =&gt; Array
        (
            [slug] =&gt; mathematics of probability
            [name] =&gt; mathematics of probability
            [canonical] =&gt; mathematics
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [mean] =&gt; Array
        (
            [slug] =&gt; mean
            [name] =&gt; mean (&amp;mu;)
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; An average obtained by adding the numbers then dividing the sum by the number of numbers that were added. Contrast this with the {{median}} (the middle in the sense that 50% of values are below it) and the {{mode}}, the value with the largest frequency. The arithmetic mean is easy to deal with mathematically, but often the {{median}} is a more {{robust}} measure.
            [has_links] =&gt; 
        )

    [means] =&gt; Array
        (
            [slug] =&gt; means
            [name] =&gt; means
            [canonical] =&gt; mean
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [mechanism] =&gt; Array
        (
            [slug] =&gt; mechanism
            [name] =&gt; mechanism
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Mechanism is about understanding how and why something occurs: the steps and processes, which buttons were pressed, what screens viewed, what information was examined and how this all creates a larger phenomenon. It is essential to understand mechanism in order to draw lessons and make predictions that {{generalise}} and  {{extrapolate}} beyond particular data and situations.
            [has_links] =&gt; 
        )

    [median] =&gt; Array
        (
            [slug] =&gt; median
            [name] =&gt; median
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A central measure that is the mid-point at which precisely half the values are below and half above. It is typically more {{robust} than the arithmetic {{mean}}, but the latter is easier for certain kinds of calculation. The term {{average}} most commonly refers to the arithmetic {{mean}}, but can also refer to the {{median}}, especially for {{asymmetric distributions}}, where the {{median}} is typically a little smaller than the {{mean}}, because the latter can be influenced by a few large {{outliers}}. For example, 'average income' usually refers to the median, not the {{mean}}.
            [has_links] =&gt; 
        )

    [menu navigation] =&gt; Array
        (
            [slug] =&gt; menu navigation
            [name] =&gt; menu navigation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Using menus as an interaction mechanism to find one's way around a system.
            [has_links] =&gt; 
        )

    [menu selection] =&gt; Array
        (
            [slug] =&gt; menu selection
            [name] =&gt; menu selection
            [canonical] =&gt; menu navigation
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [meta-analysis] =&gt; Array
        (
            [slug] =&gt; meta-analysis
            [name] =&gt; meta-analysis
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A method of using multiple studies by different people, typically in the published literature, in order to cross-validate and find emerging patterns. To enable meta-analysis, methods and results of studies must be clearly reported. Making all study data and methodology available in the public domain is a good way to enable meta-analysis, with the proviso that ethical issues must be considered.
            [has_links] =&gt; 
        )

    [mode] =&gt; Array
        (
            [slug] =&gt; mode
            [name] =&gt; mode
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The most likely value: in an {{empirical distribution}} the data value with the highest frequency (the hump!), or in the {{theoretical distribution}} the value with maximum probability. Sometimes a distribution has more than one mode, for example in university courses the more mathematical courses often have one group of students towards the higher end (those that 'get it'), then a gap and then another group with lower marks. Distributions with more than one mode sometimes happen because there is a mixed group (e.g. expert and novice users), or because of {{positive feedback}} effects, for example when students who understand some concepts learn further concepts, whereas those who misunderstand one thing are confused by everything that follows.
            [has_links] =&gt; 
        )

    [model] =&gt; Array
        (
            [slug] =&gt; model
            [name] =&gt; model
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This is some sort of formal or informal description, {{mechanism}} or mathematical formula that allows one to reason about or predict {{real-world}} phenomena.
            [has_links] =&gt; 
        )

    [model the bias] =&gt; Array
        (
            [slug] =&gt; model the bias
            [name] =&gt; model the bias
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A calculation based on measurements may have {{bias}}, that is it may on average either over- or underestimate the {{real world effect}} that it is trying to estimate. Sometimes it is possible to work out how large this bias is and then correct the calculation to take that into account. For example, imagine you have a ladder with rungs 20cm apart. You throw a ball up towards the ladder, record which pair of rungs the ball falls between, then multiply the rung it goes over by 0.20 to give the ball height in metres. This will typically underestimate the height of the ball &ndash; it is a {biased estimate}}. However, we can model the bias: it is clear that the underestimation is about half the distance between rungs. Having modelled the bias, we can add 10cm to the original calculation and obtain a {{fair estimate}} of the ball height.
            [has_links] =&gt; 
        )

    [monty hall problem] =&gt; Array
        (
            [slug] =&gt; monty hall problem
            [name] =&gt; Monty Hall problem
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A probability problem derived from a 1950s quiz challenge, where participants had to guess which of three doors concealed a prize, but were allowed to change their their initial choice after the quizmaster (Monty Hall) had opened one of the remaining doors. An example of a problem where common sense may not help and careful calculation is needed.
            [has_links] =&gt; 1
        )

    [more subjects] =&gt; Array
        (
            [slug] =&gt; more subjects
            [name] =&gt; more subjects
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The most obvious way to increase the {{statistical power}} of an experiment is to have more subjects. This is one form of the {{increase number}} strategy when you want to {{gain power}}.
            [has_links] =&gt; 
        )

    [muller-lyer illusion] =&gt; Array
        (
            [slug] =&gt; muller-lyer illusion
            [name] =&gt; M&amp;uuml;ller&amp;ndash;Lyer illusion
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A well-known optical illusion: there are two horizontal parallel lines of equal length, the top one bounded by outward-facing arrows and the lower one by inward-facing arrows. The lower line looks longer, though it is the same length. The illusion was described by Franz Carl M&uuml;ller-Lyer, a German sociologist, in 1889.
            [has_links] =&gt; 1
        )

    [multiple causes] =&gt; Array
        (
            [slug] =&gt; multiple causes
            [name] =&gt; multiple causes
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In an {{ecologically valid}} {{HCI study}}, {{in the wild}} or even in the lab, there are usually multiple causes for effects; that is different possible explanations for the effects you are seeing, several of which are often true. This happens because controls are difficult to implement and individuals and contexts always vary. {{Claims analysis}} is a method developed by Carroll and Rosson that specifically embraces multiple causes of the success or failure of systems.
            [has_links] =&gt; 
        )

    [multiple statistics] =&gt; Array
        (
            [slug] =&gt; multiple statistics
            [name] =&gt; multiple statistics
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A form of {{cherry picking}} that tries different kinds of statistical technique till something comes out 'right'. For example, a non-parametric test, followed by a t-test, followed by a Bayesian test, and so on. Sometimes a change is justified, for example if it is clear that the wrong kind of test for the distribution was chosen initially, but the reasons for making such a change must be based on the data, not the results. 
            [has_links] =&gt; 
        )

    [multiple tests] =&gt; Array
        (
            [slug] =&gt; multiple tests
            [name] =&gt; multiple tests
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A form of {{cherry picking}} where multiple tests are performed on different aspects of the same data but only the successful ones are reported. For example, a few tests that show some effect may be selected and the rest ignored, or tests that show a desired effect may be selected while those that indicate the opposite are ignored.
            [has_links] =&gt; 
        )

    [multiplicity adjustments] =&gt; Array
        (
            [slug] =&gt; multiplicity adjustments
            [name] =&gt; multiplicity adjustments
            [canonical] =&gt; 
            [see] =&gt; multiplicity control
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [multiplicity control] =&gt; Array
        (
            [slug] =&gt; multiplicity control
            [name] =&gt; multiplicity control
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Techniques for adjusting for the effect of running {{multiple tests}} on data. If you do ten tests you are ten times as likely to see an apparently positive result by pure chance; if you do 1000 tests then it starts to become highly likely that something will look promising even if there is no real effect. Some form of multiplicity adjustment or multiplicity control is therefore critical to prevent {{false positive}} results.
            [has_links] =&gt; 
        )

    [narrow user group] =&gt; Array
        (
            [slug] =&gt; narrow user group
            [name] =&gt; narrow user group
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Using a very narrow user group can {{reduce noise}} by reducing {{individual differences}} in knowledge and skill, thereby increasing {{statistical power}}. This makes it easier to see the effect of a novel interaction technique, but it also reduces the validity for generalisation beyond that group.
            [has_links] =&gt; 
        )

    [negative binomial] =&gt; Array
        (
            [slug] =&gt; negative binomial
            [name] =&gt; negative binomial
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In a test where only two outcomes are possible, the negative Binomial distribution gives you the probability of the number of times one outcome occurs before the other outcome occurs a specified number of times: for example, when a coin is tossed, the number of heads that fall before you get two tails. A special case of this is the {{Poisson distribution}}, which is when you are waiting for the very first outcome.
            [has_links] =&gt; 1
        )

    [negative correlation] =&gt; Array
        (
            [slug] =&gt; negative correlation
            [name] =&gt; negative correlation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Where two things are related and one is less likely to happen if the other happens, or, for two numerical variables, when one tends to decrease as the other increases. In the case of numbers, a perfect negative correlation occurs if one variable (say x) is linked to the other (say y) by an exact linear formula such as y = 10 &ndash; 2x. However, the effects can also be non-linear, such as y = 10 &ndash; 2x&lt;sup&gt;3&lt;/sup&gt;  and of course in statistics there is usually a substantial dose of {{noise}}! When performing {{linear regression}} the {{correlation coefficient}} will be negative. 
            [has_links] =&gt; 
        )

    [negative results] =&gt; Array
        (
            [slug] =&gt; negative results
            [name] =&gt; negative results
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; These are the times when you are not able to state anything definitively. If, given a hypothesis, you are able to say with a high degree of certainty either &quot;yes&quot; it is true or (paradoxically) &quot;no&quot; it is not, these are both regarded as {{positive results}}. If you are unable to say &quot;yes&quot; or &quot;no&quot; then that is a negative result. In traditional {{hypothesis testing}} a positive result is when you have rejected the {{null hypothesis}} and hence decided you have evidence for the {{alternative hypothesis}}; a negative result is when the test is {{non significant}}. In this case there is no definitive &quot;no&quot;, merely &quot;yes&quot; and &quot;not proven&quot;. A negative result may mean there is no effect, or that your study did not have sufficient {{statistical power}}.
            [has_links] =&gt; 
        )

    [network data] =&gt; Array
        (
            [slug] =&gt; network data
            [name] =&gt; network data
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Data derived from social networks, such as friendship circles, or similar kinds of linked objects such as web pages. The ways in which these arise often involve {{positive feedback}}; for example, if a web page has many incoming links, people are more likely to find it, and hence more likely to add yet more links. For this kind of data the {{Normal distribution}} does not apply and the data is more likely to obey a {{power law}}. 
            [has_links] =&gt; 
        )

    [network phenomena] =&gt; Array
        (
            [slug] =&gt; network phenomena
            [name] =&gt; network phenomena
            [canonical] =&gt; network data
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [new statistics] =&gt; Array
        (
            [slug] =&gt; new statistics
            [name] =&gt; new statistics
            [canonical] =&gt; 
            [see] =&gt; alternative statistics
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [noise---effect---number triangle] =&gt; Array
        (
            [slug] =&gt; noise---effect---number triangle
            [name] =&gt; noise---effect---number triangle
            [canonical] =&gt; noise-effect-number triangle
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [noise-effect-number triangle] =&gt; Array
        (
            [slug] =&gt; noise-effect-number triangle
            [name] =&gt; noise&amp;ndash;effect&amp;ndash;number triangle
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Three factors that interact to influence the {{statistical power}} of an experiment or study: the amount of {{noise}}, the {{effect size}} and the number of {{subjects}} or {{trials}}. You are likely to get a {{positive result}} if the effect size is a lot greater than the size of the noise divided by the number of subjects/trials (effect &gt;&gt; noise/number). Hence a smaller noise, a larger effect or more subjects or trials all lead to increased power. Techniques for increasing power can target one or more of the elements of the noise-effect-number triangle.
            [has_links] =&gt; 
        )

    [nominal data] =&gt; Array
        (
            [slug] =&gt; nominal data
            [name] =&gt; nominal data
            [canonical] =&gt; 
            [see] =&gt; categorical data
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [non-independence] =&gt; Array
        (
            [slug] =&gt; non-independence
            [name] =&gt; non-independence
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; When one measurement or factor influences or gives information about another. Non-independence may increase {{variability}}, lead to misattribution of effects, or suggest completely the wrong effect. Contrast with {{independence}}.
            [has_links] =&gt; 
        )

    [non-independent] =&gt; Array
        (
            [slug] =&gt; non-independent
            [name] =&gt; non-independent
            [canonical] =&gt; non-independence
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [non-independent evidence] =&gt; Array
        (
            [slug] =&gt; non-independent evidence
            [name] =&gt; non-independent evidence
            [canonical] =&gt; non-independence
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [non-linear] =&gt; Array
        (
            [slug] =&gt; non-linear
            [name] =&gt; non-linear
            [canonical] =&gt; nonlinearity
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [non-linearity] =&gt; Array
        (
            [slug] =&gt; non-linearity
            [name] =&gt; non-linearity
            [canonical] =&gt; nonlinearity
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [non-parametric methods] =&gt; Array
        (
            [slug] =&gt; non-parametric methods
            [name] =&gt; non-parametric methods
            [canonical] =&gt; nonparametric statistics
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [non-parametric statistic] =&gt; Array
        (
            [slug] =&gt; non-parametric statistic
            [name] =&gt; non-parametric statistic
            [canonical] =&gt; nonparametric statistics
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [non-parametric statistics] =&gt; Array
        (
            [slug] =&gt; non-parametric statistics
            [name] =&gt; non-parametric statistics
            [canonical] =&gt; nonparametric statistics
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [non-parametric test] =&gt; Array
        (
            [slug] =&gt; non-parametric test
            [name] =&gt; non-parametric test
            [canonical] =&gt; nonparametric statistics
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [non-significance] =&gt; Array
        (
            [slug] =&gt; non-significance
            [name] =&gt; non-significance
            [canonical] =&gt; non-significant result
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [non-significant] =&gt; Array
        (
            [slug] =&gt; non-significant
            [name] =&gt; non-significant
            [canonical] =&gt; non-significant result
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [non-significant result] =&gt; Array
        (
            [slug] =&gt; non-significant result
            [name] =&gt; non-significant result
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; When the results from your experiment or study do not reach a pre-defined measure of {{significance}}, that is where they could simply have occurred randomly with a given probability. Often the {{p-value}} used for this is a standard one such as 5% or 1%, largely because in days past it was only possible to calculate these in books of statistical tables. It is very important to remember that non-significant does &lt;strong&gt;not&lt;/strong&gt; mean {{no effect}}, but simply not sufficient evidence to be confident that there is a real effect &ndash; {{not proven}}.
            [has_links] =&gt; 
        )

    [non-significant results] =&gt; Array
        (
            [slug] =&gt; non-significant results
            [name] =&gt; non-significant results
            [canonical] =&gt; non-significant result
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [nonlinear] =&gt; Array
        (
            [slug] =&gt; nonlinear
            [name] =&gt; nonlinear
            [canonical] =&gt; nonlinearity
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [nonlinearity] =&gt; Array
        (
            [slug] =&gt; nonlinearity
            [name] =&gt; nonlinearity
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This is when the relationship between things cannot be represented by a simple linear formula such as y=3+2x; that is where there is some sort of bend or curve in the graph of the data. In statistics this tends to be harder to see because there is usually a lot of noise in the data. Note particularly that when performing {{linear regression}}, two variables might have a large {{correlation coefficient}} but the relationship can still be non-linear.
            [has_links] =&gt; 
        )

    [nonparametric methods] =&gt; Array
        (
            [slug] =&gt; nonparametric methods
            [name] =&gt; nonparametric methods
            [canonical] =&gt; nonparametric statistics
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [nonparametric statistic] =&gt; Array
        (
            [slug] =&gt; nonparametric statistic
            [name] =&gt; nonparametric statistic
            [canonical] =&gt; nonparametric statistics
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [nonparametric statistics] =&gt; Array
        (
            [slug] =&gt; nonparametric statistics
            [name] =&gt; nonparametric statistics
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; One of two broad classes of statistical technique, where there is no known {{theoretical distribution}} or it is too complex, and universal properties such as the ordering of values must be relied upon instead. Contrast with {{parametric statistics}}.
            [has_links] =&gt; 1
        )

    [nonparametric test] =&gt; Array
        (
            [slug] =&gt; nonparametric test
            [name] =&gt; nonparametric test
            [canonical] =&gt; nonparametric statistics
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [normal approximation] =&gt; Array
        (
            [slug] =&gt; normal approximation
            [name] =&gt; Normal approximation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; When we approximate some other distribution using the {{Normal distribution}}. For example, if we have tossed 100 coins and counted the number of heads, this is strictly a {{Binomial distribution}}; however, we might approximate this by a Normal distribution with {{mean}} of 50 and {{standard deviation}} of 0.05.
            [has_links] =&gt; 
        )

    [normal distribution] =&gt; Array
        (
            [slug] =&gt; normal distribution
            [name] =&gt; Normal distribution
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This is the ubiquitous bell-shaped curve. Some phenomena such as heights seem to naturally follow this distribution, whilst others, including coin-toss scores (the {{Binomial distribution}}), end up looking {{approximately Normal}} for sufficiently many tosses. This {{theoretical distribution}} is also known as a {{Gaussian distribution}}.
            [has_links] =&gt; 
        )

    [normal distributions] =&gt; Array
        (
            [slug] =&gt; normal distributions
            [name] =&gt; normal distributions
            [canonical] =&gt; normal distribution
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [normally distributed] =&gt; Array
        (
            [slug] =&gt; normally distributed
            [name] =&gt; Normally distributed
            [canonical] =&gt; normal distribution
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [not independent] =&gt; Array
        (
            [slug] =&gt; not independent
            [name] =&gt; not independent
            [canonical] =&gt; non-independence
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [not proven] =&gt; Array
        (
            [slug] =&gt; not proven
            [name] =&gt; not proven
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The state of not being able to make a strong statement either for or against the effect, system, or theory being tested. If results are non-significant, it can never be reasoned that a hypothesis is false, only that it is not statistically proven. This is why {{statistical power}} is so important, to increase the likelihood that an experiment or study is likely to reveal an effect if it is present. &lt;br&gt;In Scottish law, courts can return three possible verdicts: &quot;guilty&quot;, &quot;not guilty&quot; and &quot;not proven&quot;, where &quot;not guilty&quot; is a strong statement of innocence, whereas &quot;not proven&quot; is when the court does not feel there is sufficient evidence to make a definitive judgement in either direction. In contrast, English courts only have the options &quot;guilty&quot; or &quot;not guilty&quot;, where the latter encompasses both strong evidence of innocence and lack of evidence. Traditional {{hypothesis testing}} is more like the English court, whereas using {{confidence intervals}} or certain kinds of {{Bayesian statistics}} (where applicable) may allow a strong negative &quot;no difference&quot; to be distinguished from the weak &quot;don't know&quot;.


            [has_links] =&gt; 
        )

    [not significant] =&gt; Array
        (
            [slug] =&gt; not significant
            [name] =&gt; not significant
            [canonical] =&gt; non-significant result
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [null hypothesis] =&gt; Array
        (
            [slug] =&gt; null hypothesis
            [name] =&gt; null hypothesis
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In {{hypothesis testing}} the null hypothesis, often written H&lt;sub&gt;0&lt;/sub&gt;, is usually the thing you would like to show is false. For example, that your updated interface design has not made any improvement.
            [has_links] =&gt; 
        )

    [numeric data] =&gt; Array
        (
            [slug] =&gt; numeric data
            [name] =&gt; numeric data
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Data that can be counted or measured as a number, such as completion time in an experiment, or number of heads in coin tosses.
            [has_links] =&gt; 
        )

    [numeric values] =&gt; Array
        (
            [slug] =&gt; numeric values
            [name] =&gt; numeric values
            [canonical] =&gt; numeric data
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [observed outcome] =&gt; Array
        (
            [slug] =&gt; observed outcome
            [name] =&gt; observed outcome
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This is a term for the actual data or evidence seen, as opposed to a predicted value or probabilistic range of possibilities.
            [has_links] =&gt; 
        )

    [odds ratio] =&gt; Array
        (
            [slug] =&gt; odds ratio
            [name] =&gt; odds ratio
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The odds ratio is the general term for the ratio of a particular outcome given two different conditions. For example how much more likely a person is to get type 2 diabetes if they are obese, compared with if they are of a lower weight. The {{Bayes factor}} is a particular example of an odds ratio, where it is used as a measure of the strength of evidence between hypotheses.
            [has_links] =&gt; 
        )

    [one-tailed test] =&gt; Array
        (
            [slug] =&gt; one-tailed test
            [name] =&gt; one-tailed test
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A test where data is only compared in one direction. For example, when comparing two systems you may only care about seeing if the new system is &lt;em&gt;better&lt;/em&gt;. If they are the same or if the new system is worse, these are equally failures of the new design. So, you care that the new system has a  &lt;em&gt;faster&lt;/em&gt; {{response time}} than the old one (one direction), or if it has a &lt;em&gt;lower&lt;/em&gt; {{error rate}} (one direction). Contrast with {{two-tailed test}}. 
            [has_links] =&gt; 
        )

    [ontologically random] =&gt; Array
        (
            [slug] =&gt; ontologically random
            [name] =&gt; ontologically random
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; (An effect) that is random in its actual state of being, such as the radioactive decay of an atom, contrasted with (an effect) that is {{epistemologically random}} &mdash; random in your state of knowledge. For example, if you are about to toss a coin, the future result of the toss is ontologically random; however, once you have tossed the coin but not yet looked at it, it is epistemologically random: you would still treat it as a 50:50 odds if asked to bet, but actually it is already on one state or the other, you just don't know which.
            [has_links] =&gt; 
        )

    [open scholarship] =&gt; Array
        (
            [slug] =&gt; open scholarship
            [name] =&gt; open scholarship
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The placement of research into the public domain: instructions, software, raw study data, and derived data. This enables {{replication}} and {{meta-analysis}}, as well as new research that tests alternative theories, or mines the data for emergent patterns. Consideration must be given to ethical issues such as privacy and transparency, practical issues such as documentation, and cultural issues around the use of data in academic disciplines.
            [has_links] =&gt; 
        )

    [open situations] =&gt; Array
        (
            [slug] =&gt; open situations
            [name] =&gt; open situations
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Where we do not have control of the context within which a system under study is being used. It is a feature of {{in the wild}} research or studies of deployed systems in actual use.
            [has_links] =&gt; 
        )

    [opinion poll] =&gt; Array
        (
            [slug] =&gt; opinion poll
            [name] =&gt; opinion poll
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A form of survey or {{count data}} where members of the public are asked for their judgments or intentions, for example voting intentions before an election, or opinion of a government policy.
            [has_links] =&gt; 
        )

    [order effects] =&gt; Array
        (
            [slug] =&gt; order effects
            [name] =&gt; order effects
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; One of a number of ways in which measurements may be related to one another, which can occur where, for example, a test compares a user using a particular interface followed by the same user using a different interface. Types of order effects include: {{positive learning effects}}, where the user is better at the second interface because they have already become familiar with the application while using the first; and {{interference effects}}, where the user does less well in the second interface because they have got used to the features of the first.
            [has_links] =&gt; 
        )

    [ordinal data] =&gt; Array
        (
            [slug] =&gt; ordinal data
            [name] =&gt; ordinal data
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Data that has an order, but otherwise no specific meaning to the values; for example the degrees of agreement or satisfaction in a {{Likert scale}}. Sometimes this can be converted to {{binary data}} for ease of analysis, though this loses some information. Contrast with {{arithmetical data}} and {{categorical data}}.
            [has_links] =&gt; 
        )

    [outliers] =&gt; Array
        (
            [slug] =&gt; outliers
            [name] =&gt; outliers
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Unexpectedly extreme values in data, which may be caused by a fault in equipment, or other irrelevant effect, or may occur by chance. While they may be valid data points within the random nature of the phenomenon being studied, even a few outliers can have a disproportionate effect on statistics such as arithmetic mean and variance. For some purposes, better results are obtained by removing the most extreme outliers, but this can introduce the potential for {{cherry picking}}, so, where possible, a strategy for dealing with outliers should be determined prior to any analysis.
            [has_links] =&gt; 
        )

    [p-curve] =&gt; Array
        (
            [slug] =&gt; p-curve
            [name] =&gt; p-curve
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; If we conduct an experiment where there is no real effect, we would expect, by random chance, that we would get, around 5% of the time, a {{p-value}} of 0.05 or lower and, about 1% of the time, a {{p-value}} less than 0.01 &ndash; that is precisely what these {{significance levels}} mean: on average, a flat curve for p-values vs number of experiments with that value. In contrast, where there is a real effect we would hope to see a considerably higher proportion of lower {{p-values}}. Because of the {{file drawer effect}} not all the results of experiments are published, so typically we are unlikely to see the {{negative results}}. However, it has been suggested by Simonsohn et al. that the shape of the {{p-curve}}, that is the graph of results with each {{p-value}}, can give a sense of the health of a discipline and the impact of the {{file drawer effect}}. If this curve is flat, it suggests that there are few real results and most are pure chance; if it is highly curved, with less than a 5:1 ratio of 5% or more to 1% or more results, then it is likely that the discipline is using studies with sufficient {{statistical power}} and that most or many of the published results do in fact correspond to real effects.
            [has_links] =&gt; 1
        )

    [p-hacking] =&gt; Array
        (
            [slug] =&gt; p-hacking
            [name] =&gt; p-hacking
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Where a researcher manipulates their results in some way to obtain a {{p-value}} that is deemed {{significant}} and hence able to be published. This may be deliberate, or simply the result of poor practice or poor training. One example of {{p-hacking}} is to try out many, many different experiments and data analysis methods, and only report those that are {{significant}}, meaning that all the {{non-significant}} experiments are ignored; a practice called the {{file drawer effect}}. The impact of {{p-hacking}} is that there may be published results that are pure chance and do not correspond to a real effect; this increases the number of {{false positives}} in the literature, hence reducing the validity of the entire discipline.
            [has_links] =&gt; 
        )

    [p-level] =&gt; Array
        (
            [slug] =&gt; p-level
            [name] =&gt; p-level
            [canonical] =&gt; p-value
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [p-value] =&gt; Array
        (
            [slug] =&gt; p-value
            [name] =&gt; p-value
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The {{likelihood}} that the data we have measured, or the result we have obtained, could have arisen by chance given the {{null hypothesis}}. Note that because this is related to the probability of making an error, big p-values are bad and small p-values are good. &lt;br&gt;The null hypothesis is often some form of {{no effect}}, such as there being no improvement due to a new system being used. P-values are often quoted at specific levels such as 5%, 1% or 0.1% (p&lt;0.05, p&lt;0.01, or p&lt;0.001 respectively), but this is largely because they were difficult to calculate in the past and hence one had to look them up in a book of statistical tables where they had been (laboriously) precalculated. Recent advice from the {{Royal Statistical Society}} is to quote the exact p-value as a measure of evidence, rather than rely on magic values. &lt;br&gt;Note also that even if there is no effect, a p-value of 5% is likely to occur one time in twenty and a p-value of 1% one time in a hundred, that is one may occasionally get a {{false positive}} result, or {{Type I error}}. A low p-value means that a {{false positive}} is less likely.
            [has_links] =&gt; 1
        )

    [p-values] =&gt; Array
        (
            [slug] =&gt; p-values
            [name] =&gt; p-values
            [canonical] =&gt; p-value
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [pair subjects] =&gt; Array
        (
            [slug] =&gt; pair subjects
            [name] =&gt; pair subjects
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A special, but common, case of {{matching subjects}} where you match pairs of subjects with similar characteristics and make sure that one from each pair is assigned to two different conditions.
            [has_links] =&gt; 
        )

    [parameters] =&gt; Array
        (
            [slug] =&gt; parameters
            [name] =&gt; parameters
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; These are variables in {{models}} or {{theoretical distributions}} that typically need to be fitted to data or provided with values based on other knowledge. For example, the {{Binomial distribution}} has two parameters &ndash; Binomial(N.p). One parameter, N, is the number of trials and the other parameter, p, the probability of success of each ({{independent}}) trial. If we use this for 10 tosses of a {{fair coin}}, we can fill in these parameters based on our knowledge: N=10 and p=0.5. However, if we are uncertain whether the coin is biased, we may toss it a hundred times (N=100), but use the fact that it is a binomial distribution to fit p, that is to estimate the {{bias}} of the coin. &lt;br&gt;In a more {{HCI}}-specific example, {{Fitts' Law}} says that time to hit a target is A + B x IoD (where IoD is the {{index of difficulty}}), but the {{parameters}} A and B vary on the characteristics of the device used (e.g. mouse vs trackpad vs finger pointing).
            [has_links] =&gt; 
        )

    [parametric statistics] =&gt; Array
        (
            [slug] =&gt; parametric statistics
            [name] =&gt; parametric statistics
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A statistical technique where there is a known theoretical distribution, and only a few unknown {{parameters}} need to be estimated. For example, people's heights, where we may know that heights have a {{Normal distribution}}, but need to estimate average height and its {{standard deviation}}. Statisticians prefer to use parametric statistics where possible, because they are less likely to miss a real effect; the statistics have greater {{power}}. Contrast with {{nonparametric statistics}}.
            [has_links] =&gt; 
        )

    [pearson correlation coefficient] =&gt; Array
        (
            [slug] =&gt; pearson correlation coefficient
            [name] =&gt; Pearson correlation coefficient
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This measures the level of linear relationship between two variables. It lies in the range [-1,+1] where +1 denotes a perfect positive relationship and -1 a perfect negative relationship. It is calculated by working out the {{covariance}} (average product) of the two variables and dividing by the product of their individual {{standard deviation}}:  ( cov(x,y) / ( &amp;sigma;&lt;sub&gt;x&lt;/sub&gt; &amp;times; &amp;sigma;&lt;sub&gt;y&lt;/sub&gt;).
            [has_links] =&gt; 
        )

    [peripheral vision] =&gt; Array
        (
            [slug] =&gt; peripheral vision
            [name] =&gt; peripheral vision
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Where a person is not looking directly at something, but it is still visible out of their central vision. Human peripheral vision is not accurate enough for fine {{positioning tasks}}.
            [has_links] =&gt; 
        )

    [perturbation analysis ] =&gt; Array
        (
            [slug] =&gt; perturbation analysis 
            [name] =&gt; perturbation analysis 
            [canonical] =&gt; 
            [see] =&gt; sensitivity analysis
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [pietree] =&gt; Array
        (
            [slug] =&gt; pietree
            [name] =&gt; pietree
            [canonical] =&gt; pietree visualisation
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [pietree visualisation] =&gt; Array
        (
            [slug] =&gt; pietree visualisation
            [name] =&gt; PieTree
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The PieTree is an interactive {{constant area visualisation}} for hierarchical data with some form of numerical attribute including values at the nodes &ndash; for example in a website where there are visits to both the main page of a section and also to sub-pages. The visualisation starts off like a standard pie chart where each segment represents a branch of the tree (e.g. section of the site), except there is also a small circle in the middle representing the value at the root (e.g home page). If any of the segments is selected it breaks into smaller segments for each sub-branch, and so on. 
            [has_links] =&gt; 1
        )

    [plausibility] =&gt; Array
        (
            [slug] =&gt; plausibility
            [name] =&gt; plausibility
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In {{Bayesian statistics}} the 'probabilities' used for the {{prior distribution}} and hence the resulting {{posterior distribution}} are {{belief encoded as probability}} &ndash; that is using a number between 0 and 1 to represent a level of belief. To distinguish these from true probabilities they are called {{plausibility}} or {{reasonable expectation}}. For the purposes of mathematical calculation, {{plausibility}} satisfies the {{Kolomogrov Laws}} of probability and so can be treated mathematically in the same way even though its meaning is subtly different. 
            [has_links] =&gt; 
        )

    [pointing task] =&gt; Array
        (
            [slug] =&gt; pointing task
            [name] =&gt; pointing task
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A class of low-level device interaction task where a mouse, trackpad, stylus or finger is used  to move towards a target or follow some sort of path. This is typically to target an element onscreen in order to act on it, for example to drag it or select it. The time for positioning tasks often satisfies some variant of {{Fitts' Law}}. 
            [has_links] =&gt; 
        )

    [pointing tasks] =&gt; Array
        (
            [slug] =&gt; pointing tasks
            [name] =&gt; pointing tasks
            [canonical] =&gt; pointing task
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [poisson distribution] =&gt; Array
        (
            [slug] =&gt; poisson distribution
            [name] =&gt; Poisson distribution
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This {{probability distribution}} is used when you have some sort of {{binary data}} (such as the toss of a coin, or whether a die rolls a 6), a fixed number of trials (say tossing the coin 10 times) and each trial is {{independent}}. The Poisson distribution gives the probability of getting a certain number of 'successes' (e.g. head in a coin toss). The Poisson distribution has two parameters, N (the number of trials) and p (the probability that a single trial is a 'success'). The probability of n successes is then Comb(N,n)  p&lt;sup&gt;n&lt;/sup&gt; (1-p)&lt;sup&gt;N-n&lt;/sup&gt;, where Comb(N.n) is the number of combinations of n things out of N = N! / ( n! (N-n)! ). When N becomes large, the central portion of the Poisson distribution is {{approximately Normal}}.
            [has_links] =&gt; 1
        )

    [population] =&gt; Array
        (
            [slug] =&gt; population
            [name] =&gt; population
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A complete group of people one wants to know about, for example all the employees in a company, or all users of a product. By gathering data about a small {{sample}} from the population an estimate may be made about the population as a whole.
            [has_links] =&gt; 
        )

    [positive correlation] =&gt; Array
        (
            [slug] =&gt; positive correlation
            [name] =&gt; positive correlation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Where two things are related and one is more likely to happen if the other happens or if one thing tends to become greater when the other does. Contrast with {{negative correlation}}.
            [has_links] =&gt; 
        )

    [positive feedback] =&gt; Array
        (
            [slug] =&gt; positive feedback
            [name] =&gt; positive feedback
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Where effects accumulate and increase the overall effect. Examples are: snowflake growth where ice forms more easily on sharp tips, so any small growth grows further; earthquakes, where a very small release of tension in a fault can lead to a larger release in adjacent areas and culminate in an earthquake; citation rates, where the more citations a paper gains, the more it is seen and the more it is cited further. 
            [has_links] =&gt; 
        )

    [positive feedback effects] =&gt; Array
        (
            [slug] =&gt; positive feedback effects
            [name] =&gt; positive feedback effects
            [canonical] =&gt; positive feedback
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [positive learning] =&gt; Array
        (
            [slug] =&gt; positive learning
            [name] =&gt; positive learning
            [canonical] =&gt; 
            [see] =&gt; learning effects
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [positive result] =&gt; Array
        (
            [slug] =&gt; positive result
            [name] =&gt; positive result
            [canonical] =&gt; positive results
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [positive results] =&gt; Array
        (
            [slug] =&gt; positive results
            [name] =&gt; positive results
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Results that are {{significant}} and either prove or disprove a hypothesis. There is a cultural tendency for only positive results to be published &ndash; the {{file drawer effect}}. Contrast with a {{negative result}} where you are unable to make a strong statement other than {{not proven}}.
            [has_links] =&gt; 
        )

    [post-hoc hypotheses] =&gt; Array
        (
            [slug] =&gt; post-hoc hypotheses
            [name] =&gt; post-hoc hypotheses
            [canonical] =&gt; post-hoc hypothesis
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [post-hoc hypothesis] =&gt; Array
        (
            [slug] =&gt; post-hoc hypothesis
            [name] =&gt; post-hoc hypothesis
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A form of {{cherry picking}} where the initial hypothesis is replaced by one suggested by a pattern in the data. Although this may be less than ideal it is pragmatically necessary in some contexts. However, where a post-hoc hypothesis is used its use should be openly acknowledged and any potential impact on results should be assessed.
            [has_links] =&gt; 
        )

    [post-hoc hypothesis testing] =&gt; Array
        (
            [slug] =&gt; post-hoc hypothesis testing
            [name] =&gt; post-hoc hypothesis testing
            [canonical] =&gt; post-hoc hypothesis
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [posterior] =&gt; Array
        (
            [slug] =&gt; posterior
            [name] =&gt; posterior
            [canonical] =&gt; posterior distribution
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [posterior distribution] =&gt; Array
        (
            [slug] =&gt; posterior distribution
            [name] =&gt; posterior distribution
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In {{Bayesian methods}} this is the probability calculated after combining fresh evidence with the {{prior distribution}}. It is given by {{Bayes rule}}: Posterior(A&lt;sub&gt;f&lt;/sub&gt;|E)  = P(E|A&lt;sub&gt;f&lt;/sub&gt;)xPrior(A&lt;sub&gt;f&lt;/sub&gt;)/(&amp;Sigma;&lt;sub&gt;all i&lt;/sub&gt; P(E|A&lt;sub&gt;i&lt;/sub&gt;)xPrior(A&lt;sub&gt;i&lt;/sub&gt;)), where E is the evidence, and A&lt;sub&gt;i&lt;/sub&gt; ranges over all alternatives including the focus one A&lt;sub&gt;f&lt;/sub&gt;. Note that in {{Bayesian statistics}} the {{posterior distribution}} is typically not a real probability but a {{quantified belief}} or {{ plausibility}}.
            [has_links] =&gt; 
        )

    [posterior probabilities] =&gt; Array
        (
            [slug] =&gt; posterior probabilities
            [name] =&gt; posterior probabilities
            [canonical] =&gt; posterior distribution
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [posterior probability] =&gt; Array
        (
            [slug] =&gt; posterior probability
            [name] =&gt; posterior probability
            [canonical] =&gt; posterior distribution
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [posterior probability distribution] =&gt; Array
        (
            [slug] =&gt; posterior probability distribution
            [name] =&gt; posterior probability distribution
            [canonical] =&gt; posterior distribution
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [power (statistical)] =&gt; Array
        (
            [slug] =&gt; power (statistical)
            [name] =&gt; power (statistical)
            [canonical] =&gt; 
            [see] =&gt; statistical power
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [power law] =&gt; Array
        (
            [slug] =&gt; power law
            [name] =&gt; power law
            [canonical] =&gt; power-law distribution
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [power law distribution] =&gt; Array
        (
            [slug] =&gt; power law distribution
            [name] =&gt; power law distribution
            [canonical] =&gt; power-law distribution
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [power law distributions] =&gt; Array
        (
            [slug] =&gt; power law distributions
            [name] =&gt; power law distributions
            [canonical] =&gt; power-law distribution
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [power-law data] =&gt; Array
        (
            [slug] =&gt; power-law data
            [name] =&gt; power-law data
            [canonical] =&gt; power-law distribution
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [power-law distribution] =&gt; Array
        (
            [slug] =&gt; power-law distribution
            [name] =&gt; power-law distribution
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A form of {{long-tail distribution}} where the frequency is inversely proportional to some power: that is freq(n) = 1/n&lt;sup&gt;&amp;lambda;&lt;/sup&gt;. These occur in many forms of {{network data}} such as number of web page links, number of contacts in {{social networks}}, or number of citations of papers, and in other areas such as city sizes, wealth of individuals or companies, and size of earthquakes. They are often the result of {{positive feedback}} effects; for example, papers that are highly cited are more widely seen and hence attract more citations; or people with lots of money have more opportunities to make more. The power {{parameter}} &amp;lambda; varies between different kinds of data, but is often a little under 2 for web data. This is critical, because if the value is less than 2, the distribution does not have {{finite variance}}, one of the prerequisites for the {{Central Limit Theorem}}. The practical impact is that averages based on the data are not {{approximately Normal}} and hence you &lt;em&gt;must not&lt;/em&gt; use tests designed for the {{Normal distribution}}, even when the {{sample size}} is very large.
            [has_links] =&gt; 1
        )

    [power-law distributions] =&gt; Array
        (
            [slug] =&gt; power-law distributions
            [name] =&gt; power-law distributions
            [canonical] =&gt; power-law distribution
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [pre-register] =&gt; Array
        (
            [slug] =&gt; pre-register
            [name] =&gt; pre-register
            [canonical] =&gt; pre-registration
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [pre-registration] =&gt; Array
        (
            [slug] =&gt; pre-registration
            [name] =&gt; pre-registration
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A way to avoid the {{file drawer effect}} by recording a statement before starting an experiment to specify the hypothesis being tested and which statistical technique will be used. This forces justification of the reasons for any {{post-hoc hypothesis}}. It is standard practice in many areas of science, and becoming so in {{HCI}}.
            [has_links] =&gt; 
        )

    [prior] =&gt; Array
        (
            [slug] =&gt; prior
            [name] =&gt; prior
            [canonical] =&gt; prior distribution
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [prior belief] =&gt; Array
        (
            [slug] =&gt; prior belief
            [name] =&gt; prior belief
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The assumptions that are held to be true before starting an experiment. {{Bayesian reasoning}} tries to quantify these beliefs.
            [has_links] =&gt; 
        )

    [prior beliefs] =&gt; Array
        (
            [slug] =&gt; prior beliefs
            [name] =&gt; prior beliefs
            [canonical] =&gt; prior belief
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [prior distribution] =&gt; Array
        (
            [slug] =&gt; prior distribution
            [name] =&gt; prior distribution
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; When applying {{Bayes rule}} this is the probability one would assign to various parameters &lt;em&gt;before&lt;/em&gt; considering additional evidence. It is then combined with the evidence from the data to obtain the {{posterior probability}}. In {{Bayesian statistics}}, the values for this prior may not be a true probability but rather {{prior belief}} {{encoded as probability}}.
            [has_links] =&gt; 
        )

    [prior probabilities] =&gt; Array
        (
            [slug] =&gt; prior probabilities
            [name] =&gt; prior probabilities
            [canonical] =&gt; prior distribution
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [prior probability] =&gt; Array
        (
            [slug] =&gt; prior probability
            [name] =&gt; prior probability
            [canonical] =&gt; prior distribution
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [prior probability distribution] =&gt; Array
        (
            [slug] =&gt; prior probability distribution
            [name] =&gt; prior probability distribution
            [canonical] =&gt; prior distribution
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [priors] =&gt; Array
        (
            [slug] =&gt; priors
            [name] =&gt; priors
            [canonical] =&gt; prior distribution
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [probabilistic phenomena] =&gt; Array
        (
            [slug] =&gt; probabilistic phenomena
            [name] =&gt; probabilistic phenomena
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Events that have a likelihood of happening but are not predetermined.
            [has_links] =&gt; 
        )

    [probability distribution] =&gt; Array
        (
            [slug] =&gt; probability distribution
            [name] =&gt; probability distribution
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The function or histogram that for any data value or potential outcome of a measurement tells you its probability. This can be an {{empirical distribution}} based on measurements of real data, such as an {{income distribution}} that tells you what proportion of people earn any specific amount. Alternatively, it can be a {{theoretical distribution}} based on mathematical analysis, for example the {{Binomial distribution}} that tells you the probability of getting exactly n heads out of 20 coin tosses.
            [has_links] =&gt; 
        )

    [probability distributions] =&gt; Array
        (
            [slug] =&gt; probability distributions
            [name] =&gt; probability distributions
            [canonical] =&gt; probability distribution
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [proof by contradiction] =&gt; Array
        (
            [slug] =&gt; proof by contradiction
            [name] =&gt; proof by contradiction
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; An argument form in logic or mathematics, where what is believed to be false is asserted as if it were true, in order to reason from that to something that is clearly contradictory, and then use the contradiction to argue that what was first asserted must indeed be false. Compare with the {{null hypothesis}} in statistics.
            [has_links] =&gt; 
        )

    [public domain] =&gt; Array
        (
            [slug] =&gt; public domain
            [name] =&gt; public domain
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This is when the various materials used or produced in research are made available to everyone in some publicly accessible form. These materials include instructions, software (if applicable), and of course the raw study data, whether this is survey reports, video, or keystroke logging, as well as derived data, all the way to the data that lies behind the graphs in your published papers.
            [has_links] =&gt; 
        )

    [public sector borrowing] =&gt; Array
        (
            [slug] =&gt; public sector borrowing
            [name] =&gt; public sector borrowing
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Borrowing by government or government bodies as opposed to borrowing by companies and individuals (private sector).
            [has_links] =&gt; 
        )

    [publication bias] =&gt; Array
        (
            [slug] =&gt; publication bias
            [name] =&gt; publication bias
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; If only {{positive results}} are publshied, then it may appaer that they are more {{significant}} (in all senses of the word) than they really are.  This is becasue of you do enough experiments or studies, then by raodm chance one of them will come up 5% significant about 1 time in 20. Publication bias occurs artly because researchers self-select to report things that are most impressive, but also because journal and conference reviewers may not accept papers with {{negative results}}. In {{qualitative}} studies, there can be similar effects, if you look at enough people, or enough events you will see some interesting things, and may tend to report these leaing to a sense that they are common, whereas they may not be at all typical.
            [has_links] =&gt; 
        )

    [publish or perish] =&gt; Array
        (
            [slug] =&gt; publish or perish
            [name] =&gt; publish or perish
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A cultural definition of academic success that requires peer-reviewed publication of research results.
            [has_links] =&gt; 
        )

    [qualitative methods] =&gt; Array
        (
            [slug] =&gt; qualitative methods
            [name] =&gt; qualitative methods
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Research methods such as rich observations, ethnography, or post-experiment interviews, which are suited to exploratory research, revealing the processes and mechanisms by which something occurs and answering how and why questions. Contrast with {{quantitative methods}}.
            [has_links] =&gt; 
        )

    [qualitative research] =&gt; Array
        (
            [slug] =&gt; qualitative research
            [name] =&gt; qualitative research
            [canonical] =&gt; qualitative methods
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [quantified belief] =&gt; Array
        (
            [slug] =&gt; quantified belief
            [name] =&gt; quantified belief
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In {{Bayesian statistics}} you have to represent your level of belief in terms of a {{probability}}. This not a real probability, but it does obey the {{Kolomogrov laws}}, so behaves like probability in many circumstances. Crucially, when a posterior 'probability' is calculated this too is a measure of belief informed by additional evidence.
            [has_links] =&gt; 
        )

    [quantitative data] =&gt; Array
        (
            [slug] =&gt; quantitative data
            [name] =&gt; quantitative data
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Any data based on numbers or measurable values. This may include objectively measurable values such as {{task completion times}} or more subjective data such as {{opinion polls}}. Contrast with {{qualitative methods}}.
            [has_links] =&gt; 
        )

    [quantitative methods] =&gt; Array
        (
            [slug] =&gt; quantitative methods
            [name] =&gt; quantitative methods
            [canonical] =&gt; quantitative data
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [quartile] =&gt; Array
        (
            [slug] =&gt; quartile
            [name] =&gt; quartile
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The result of dividing a range into four sections with an equal number of items in each section when sorted. For example, if your (sorted) data values are 1.3, 1.7, 2.2, 2.5, 2.6, 2.6, 2.7, 2.9, 3.0, 3.1, 3.3, 3.4, then you divide these into four equal sections:  1.3, 1.7, 2.2  |  2.5, 2.6, 2.6 |  2.7, 2.9, 3.0 | 3.1, 3.3, 3.4; and the quartiles are therefore 2.35, 2.65, and 3.05. The middle of these is called the {{median}} and the other two are the lower and upper quartile respectively. Of the data, 25% of values are smaller than the {{lower quartile}}, 50% smaller than the {{median}} and 75% smaller than the {{upper quartile}}.
            [has_links] =&gt; 
        )

    [quartiles] =&gt; Array
        (
            [slug] =&gt; quartiles
            [name] =&gt; quartiles
            [canonical] =&gt; quartile
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [r] =&gt; Array
        (
            [slug] =&gt; r
            [name] =&gt; R
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; R is a programming language and environment designed for statistical computation. It is an open source project and free for anyone to use. Its user base has grown rapidly and it challenges {{SPSS}} as the &lt;emph&gt;de facto&lt;/emph&gt; statistical tool.
            [has_links] =&gt; 1
        )

    [random `noise'] =&gt; Array
        (
            [slug] =&gt; random `noise'
            [name] =&gt; random `noise'
            [canonical] =&gt; random noise
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [random effect] =&gt; Array
        (
            [slug] =&gt; random effect
            [name] =&gt; random effect
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A random effect is one that might vary if you repeated the same study; for example, the skill level of participants, or weather on the day of an experiment. Random effects need to be treated as an additional source of noise. Contrast with a {{fixed effect}}. 
            [has_links] =&gt; 
        )

    [random effects] =&gt; Array
        (
            [slug] =&gt; random effects
            [name] =&gt; random effects
            [canonical] =&gt; random effect
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [random noise] =&gt; Array
        (
            [slug] =&gt; random noise
            [name] =&gt; random noise
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Random effects that are irrelevant for whatever you are trying to measure or study, but add {{noise}}, making the results harder to interpret. For example, if you are measuring someone's weight, they may sway slightly, meaning the scales register slightly more or less than their actual weight; or if you ask someone whether they like a new interface feature their response may depend on how good their day has been. Part of the {{job of statistics}} is to see back, past this noise, to the underlying facts about the {{real world}}.
            [has_links] =&gt; 
        )

    [randomise orders] =&gt; Array
        (
            [slug] =&gt; randomise orders
            [name] =&gt; randomise orders
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A way to mitigate {{order effects}} by performing tests in a random order: for example some subjects try system A followed by system B,  and other subjects do the opposite.
            [has_links] =&gt; 
        )

    [real world] =&gt; Array
        (
            [slug] =&gt; real world
            [name] =&gt; real world
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; At a practical level, actual things and happenings as opposed to more theoretical or mathematical abstractions. However, what is 'real' in statistics has a variety of philosophical meanings, from concrete events to more idealised notions such as the 'typical user' of a system.
            [has_links] =&gt; 
        )

    [real-world] =&gt; Array
        (
            [slug] =&gt; real-world
            [name] =&gt; real-world
            [canonical] =&gt; real world
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [reasonable expectation] =&gt; Array
        (
            [slug] =&gt; reasonable expectation
            [name] =&gt; reasonable expectation
            [canonical] =&gt; 
            [see] =&gt; plausibility
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [reduce noise] =&gt; Array
        (
            [slug] =&gt; reduce noise
            [name] =&gt; reduce noise
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A way to increase {{statistical power}}. Variability due to factors that are not controlled or not known about is called noise. Noise can be reduced by (1) controlling conditions in a study to limit variability and (2) obtaining information about factors that may affect results and modelling these effects.
            [has_links] =&gt; 
        )

    [regression] =&gt; Array
        (
            [slug] =&gt; regression
            [name] =&gt; regression
            [canonical] =&gt; regression analysis
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [regression analysis] =&gt; Array
        (
            [slug] =&gt; regression analysis
            [name] =&gt; regression analysis
            [canonical] =&gt; 
            [see] =&gt; linear regression
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [regression line] =&gt; Array
        (
            [slug] =&gt; regression line
            [name] =&gt; regression line
            [canonical] =&gt; regression analysis
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [removing outliers] =&gt; Array
        (
            [slug] =&gt; removing outliers
            [name] =&gt; removing outliers
            [canonical] =&gt; 
            [see] =&gt; outliers
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [repeatability] =&gt; Array
        (
            [slug] =&gt; repeatability
            [name] =&gt; repeatability
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The extent to which the results of an experiment or study can be recreated by others. The language used for this varies, but the ASA distinguishes between {{reproducibility}} and {{replication}}. If others can use your data, etc., to obtain the same published result, that is {{reproducibility}}. For {{replication}}, the question is: if others follow your procedures with fresh subjects, etc., they will come to a similar result? Furthermore, some writers use the term 'repeatability' in a very specific sense: does the same scientist in the same laboratory with the same equipment get identical results? Some of these distinctions are far less clear in {{HCI}} and human sciences; for example, due to {{learning effects}} if the same subject repeats a task we do not expect a similar outcome.
            [has_links] =&gt; 1
        )

    [replicate] =&gt; Array
        (
            [slug] =&gt; replicate
            [name] =&gt; replicate
            [canonical] =&gt; replication
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [replication] =&gt; Array
        (
            [slug] =&gt; replication
            [name] =&gt; replication
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The extent to which a study or experiment can be replicated by others with the same results. True replication is difficult in HCI, where differences in subjects and contexts mean that even the most controlled experimental studies cannot be precisely repeated. A form of {{repeatability}}; contrast with {{reproducibility}}. 
            [has_links] =&gt; 1
        )

    [replication crisis] =&gt; Array
        (
            [slug] =&gt; replication crisis
            [name] =&gt; replication crisis
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The situation where many attempts to repeat scientific studies have failed to reproduce the original outcomes.
            [has_links] =&gt; 
        )

    [replichi] =&gt; Array
        (
            [slug] =&gt; replichi
            [name] =&gt; RepliCHI
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; An initiative of ACM CHI to increase {{repeatability}}, which led to the addition of a &ldquo;validation and refutation&rdquo; category in some subsequent conferences, and to long-term changes in the ACM {{CHI Conference}} publication guidelines.
            [has_links] =&gt; 
        )

    [representative] =&gt; Array
        (
            [slug] =&gt; representative
            [name] =&gt; representative
            [canonical] =&gt; representative sample
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [representative sample] =&gt; Array
        (
            [slug] =&gt; representative sample
            [name] =&gt; representative sample
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A small group who exhibit the typical characteristics of a larger group. Results from the study of a representative sample can be extrapolated to the larger group. Sometimes this sample may be chosen randomly, or may be deliberately chosen so as to ensure that important features such as gender, age or experience match those of the relevant {{population}} as a whole.
            [has_links] =&gt; 
        )

    [reproducibility] =&gt; Array
        (
            [slug] =&gt; reproducibility
            [name] =&gt; reproducibility
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 1
        )

    [residuals] =&gt; Array
        (
            [slug] =&gt; residuals
            [name] =&gt; residuals
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Many statistical procedures have some sort of underlying {{model}} for which you fit {{parameters}} based on your data. If there were no {{noise}} this might fit your data precisely, but in practice there is a difference between the value predicted by the model and the actual data points; this difference is called the residual. &lt;br&gt;For example, suppose you are performing a user study and have pre-tested the skill level of users with a value between 0 and 10. You measure their {{task completion time}} for a particular task and then calculate a {{linear regression}} to obtain a best fit line: time = 50 &ndash; 2*skill_level. Although this is a best fit line, it does not fit every user perfectly. Perhaps Aled has a task completion time of 43 seconds and a skill level of 8. The model would predict a time of 50 &ndash; 2*8 = 34 seconds; the difference of 9 (43&ndash;34) is the residual. 
            [has_links] =&gt; 
        )

    [response time] =&gt; Array
        (
            [slug] =&gt; response time
            [name] =&gt; response time
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The time it takes for someone to respond to some sort of event, notification, alarm or cue. This may be affected by the form in which the event is presented (e.g. a loud sound vs a subtle change at the edge of the screen), and by the cognitive state of the user (e.g. waiting attentively or highly focused on another task). In the case when some sort of decision is also needed, then other issues will also be important, such as the number of choices and preparedness or implicit biases of the user.
            [has_links] =&gt; 
        )

    [response times] =&gt; Array
        (
            [slug] =&gt; response times
            [name] =&gt; response times
            [canonical] =&gt; response time
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [restricted device] =&gt; Array
        (
            [slug] =&gt; restricted device
            [name] =&gt; restricted device
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A method to {{reduce noise}} by limiting the functionality of a device; for example, locking a mobile phone so that it can only use the test app, thereby ensuring that activity is directed toward the aspects being studied.
            [has_links] =&gt; 
        )

    [restricted system] =&gt; Array
        (
            [slug] =&gt; restricted system
            [name] =&gt; restricted system
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A method to {{reduce noise}} by limiting the functionality of a system. For example, in a social media application you might disable everything apart from music-related features if these are the focus of your study.
            [has_links] =&gt; 
        )

    [restricted tasks] =&gt; Array
        (
            [slug] =&gt; restricted tasks
            [name] =&gt; restricted tasks
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A method to {{reduce noise}} by limiting the activity of participants, while still maintaining a level of {{ecological validity}}; for example, giving them a scripted task to perform, but within their normal environment.
            [has_links] =&gt; 
        )

    [royal statistical society] =&gt; Array
        (
            [slug] =&gt; royal statistical society
            [name] =&gt; Royal Statistical Society
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; One of the main statistical professional bodies in the UK, and the oldest in the world.
            [has_links] =&gt; 1
        )

    [sample] =&gt; Array
        (
            [slug] =&gt; sample
            [name] =&gt; sample
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; You typically cannot measure or survey every person in the relevant {{population}} of interest: for example measure the heights of everyone in the country, or ask every possible user of your system to do a task. Instead you choose a small sample, one that you hope is large enough to give sufficient information about the full population, but is manageable.
            [has_links] =&gt; 
        )

    [sample composition] =&gt; Array
        (
            [slug] =&gt; sample composition
            [name] =&gt; sample composition
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Who or what is in your sample. When considering sample composition you need to think about whether it is a {{representative sample}} of the {{population}} you are interested in. In addition, you need to consider whether the members of your sample are {{independent}}: for example, if you do a {{snowball sample}}, people who know other people may share political views, support the same sports team or have similar musical taste.
            [has_links] =&gt; 
        )

    [sample size] =&gt; Array
        (
            [slug] =&gt; sample size
            [name] =&gt; sample size
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The number of people (or things) in a {{sample}}. Increasing the sample size typically reduces the {{variability}} of any form of {{average}} or similar measure and is one of the ways to increase {{statistical power}}. However, you typically have to increase the sample size a lot &ndash; for example to reduce the {{standard deviation}} by 10 you need a sample size 100 times larger. The {{noise-effect-number triangle}} suggests other factors that can be manipulated to increase {{statistical power}}. 
            [has_links] =&gt; 
        )

    [sample variance] =&gt; Array
        (
            [slug] =&gt; sample variance
            [name] =&gt; sample variance
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A measure of the {{variability}} of your sampled data. In the simplest case it is the average square of the difference between the data and the sample {{mean}}, but dividing by N&ndash;1 rather than N as you would for the {{population variance}} or the variance of a {{theoretical distribution}}. That is, if the data is x&lt;sub&gt;1&lt;/sub&gt;, x&lt;sub&gt;2&lt;/sub&gt;, ... x&lt;sub&gt;N&lt;/sub&gt;, the sample variance is ( &amp;Sigma;  (x&lt;sub&gt;i&lt;/sub&gt;&ndash;mean) &lt;sup&gt;2&lt;/sup&gt;) / N&ndash;1 &lt;br&gt;The reason for dividing by N&ndash;1 rather than N is that the mean is usually estimated from the sample data and so is slightly closer to the middle of the data than the real (theoretical or population) mean. Think about the extreme case with just two measured values, the sample mean is exactly in the middle, so their spread from that would give a {{biased estimate}} of the true population variance. This use of N&ndash;1 is an example where we can {{model the bias}} and hence correct it. &lt;br&gt;More generally the sample variance is the sum of the square of the {{residuals}} divided by the {{degrees of freedom}}, (N&ndash;n), where n is the number of things fitted by the statistical model. &lt;br&gt;The sample variance is often affected strongly by {{outliers}}, which then has an impact on the {{statistical power}}; one of the reason for (carefully!) removing outliers during data cleaning.
            [has_links] =&gt; 
        )

    [sample-based approaches] =&gt; Array
        (
            [slug] =&gt; sample-based approaches
            [name] =&gt; sample-based approaches
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Methods that focus on obtaining large numbers of subjects in a trial, or large numbers of systems that share the property being studied, in order to generalise results. Such an approach is typically impractical for studying properties of {{generative artefacts}} (things that make other things) because of the vast number of trials required. Alternatives include employing smaller amounts of {{rich data}} or very targeted {{evaluation}} combined with some form of {{justification}} (reasoned argument) as part of a holistic {{validation}}.
            [has_links] =&gt; 
        )

    [sampling bias] =&gt; Array
        (
            [slug] =&gt; sampling bias
            [name] =&gt; sampling bias
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This is when the way we choose a sample is not random, and becasue of this creates a {{bias}} in the eventual results.  For example, if we look through a crowd, we may be more likely to notice tall people.  This is a form of {{selection bias}}.
            [has_links] =&gt; 
        )

    [scale free distribution] =&gt; Array
        (
            [slug] =&gt; scale free distribution
            [name] =&gt; scale free distribution
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; An alternative name for a {{power-law distribution}}, which emphasises the {{scale free}} property. That is, if you imagine zooming in or out of the graph of a power law it looks the same shape at every scale. In contrast, if you have a {{Normal distribution}} (or indeed many kinds of distribution) and you zoom out, the distribution gets more narrow and the mean looks closer to zero. At a practical level Normally distributed data, such as people's heights, have a natural scale.
            [has_links] =&gt; 1
        )

    [scatter plots] =&gt; Array
        (
            [slug] =&gt; scatter plots
            [name] =&gt; scatter plots
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A simple graph that shows all the data as separate x,y points so that the nature of the spread can easily be seen, for example clumps, or extreme values. For a simple scatter plot, two data fields need to be selected for the x and y axes; however, more fields can be visualised by showing multiple plots with different pairs of fields and by colouring points differently depending on attributes.
            [has_links] =&gt; 
        )

    [schrodinger's cat] =&gt; Array
        (
            [slug] =&gt; schrodinger's cat
            [name] =&gt; Schrodinger's cat
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A quantum physics thought experiment in which a cat in a box may be both alive and dead at the same time.
            [has_links] =&gt; 
        )

    [sd] =&gt; Array
        (
            [slug] =&gt; sd
            [name] =&gt; s.d.
            [canonical] =&gt; 
            [see] =&gt; standard deviation
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [se] =&gt; Array
        (
            [slug] =&gt; se
            [name] =&gt; s.e.
            [canonical] =&gt; 
            [see] =&gt; standard error
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [seeded] =&gt; Array
        (
            [slug] =&gt; seeded
            [name] =&gt; seeded
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In a computer, 'random' numbers are usually obtained from a {{pseudo-random number generator}} &ndash; this follows a fixed deterministic algorithm to generate each number, but is so convoluted that the numbers are for most practical purposes random. However, because it is underneath a fixed algorithm, every time you restart a program you get the same series of 'random' numbers as the last time. To deal with this the random number generator is {{seeded}}, that is some data is fed in to start the generator off so that it creates  a new series of numbers. The 'seed' may include detailed system timings that depend on the momentary load of the machine, or other things that are hard to predict. However, for certain security applications that need large random numbers, this seeding process can be the weak point.
            [has_links] =&gt; 
        )

    [selection bias] =&gt; Array
        (
            [slug] =&gt; selection bias
            [name] =&gt; selection bias
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This is when the way the people who are considered in a study uninitentionally includes some kinds of people more than others and thus nduces a {{bias}} in the way resuts are interpretted.  This may occur due to {{sampling bias}} in the way people are initially selected, or other effects such as {{self-selection}} of participants in study, or participants dropping out.  Note that this affects both {{qualitative}} and {{quantitative}} analysis.
            [has_links] =&gt; 1
        )

    [selective publishing] =&gt; Array
        (
            [slug] =&gt; selective publishing
            [name] =&gt; selective publishing
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; When one only publishes particular (usually favourable!) results. This may be as a deliberate policy, but may also be due to the policy of journals only accepting {{positive results}}. The impact of {{selective publishing}} is that many results, in particular {{negative results}}, are never published, a phenomenon known as the {{file drawer effect}}. Because of this the number of {{false positives}} in the published literature may be far higher than one would expect given the quoted {{significance level}} of results, one of the causes of the {{statistical crisis}}.
            [has_links] =&gt; 
        )

    [sensitivity] =&gt; Array
        (
            [slug] =&gt; sensitivity
            [name] =&gt; sensitivity
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Experiments, data analysis and visualisation may differ in their &lt;strong&gt;sensitivity&lt;/strong&gt;, that is how easy it is for small differences to be apparent. For example, if you have histograms showing the heights of the 30 tallest mountains in the world, it may be hard to see the difference after the first few, as there are many that are close to 26,000  feet (7,000 m) tall. However, if you use a {{false baseline}} at 25,000 feet, the heights are very clearly different. In experimental design {{distractor tasks}} are also often useful as they expose mental load.
            [has_links] =&gt; 
        )

    [sensitivity analysis] =&gt; Array
        (
            [slug] =&gt; sensitivity analysis
            [name] =&gt; sensitivity analysis
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Sensitivity analysis is about trying to assess how much effect small changes in parameters or data values have on the ultimate result. If a small change in an uncertain or random value will create a large change in the result, then that will reduce one's confidence in the ultimate result. We may perform {{sensitivity analysis}} by making {{perturbations}} (small changes) in values, either mathematically calculating the impact or simply re-running the analysis with the perturbed values &ndash; if the small perturbation has made a big change, we know we need to be careful. This is sometimes also called {{perturbation analysis}}.&lt;br&gt;One example of this is in calculating different measures of {{variation}}. We know (mathematically) that {{outliers}} have a large impact on the {{standard deviation}}: it is very sensitive; in contrast the {{interquartile range}} is not affected at all by a single {{outlier}}: it is {{robust}}. Another example is found in {{Bayesian statistics}}, where the choice of {{prior distribution}} affects the {{posterior}}. One could therefore in principle re-work the analysis with slightly different priors ... although in practice this is rarely done.
            [has_links] =&gt; 
        )

    [significance] =&gt; Array
        (
            [slug] =&gt; significance
            [name] =&gt; significance
            [canonical] =&gt; statistically significant
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [significance level] =&gt; Array
        (
            [slug] =&gt; significance level
            [name] =&gt; significance level
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The threshold at which the decision is made to reject the {{null hypothesis}}. The smallest {{significance level}} that is normally regarded as reasonable evidence is 5%, which means that the probability of a {{false positive}} (or {{Type I error}}) is less than 1 in 20. There is now considerable debate on whether choosing such an arbitrary {{critical value}} is helpful. Instead it is suggested one should quote the exact {{p-value}}, allowing the reader to assess the strength of evidence this suggests.
            [has_links] =&gt; 
        )

    [significance levels] =&gt; Array
        (
            [slug] =&gt; significance levels
            [name] =&gt; significance levels
            [canonical] =&gt; significance level
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [significance test] =&gt; Array
        (
            [slug] =&gt; significance test
            [name] =&gt; significance test
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The generic term for various tests that are used to see if a {{hypothesis}} about data is {{statistically significant}} (that is, unlikely to happen by chance). Examples include the {{Student's t test}} used to see if the {{means}} of two sets of data are different, and the {{Pearson correlation coefficient}}, used to test whether there is some form of relation between two variables in {{linear regression}}. Typically the test will be based on some form of {{theoretical distribution}} leading to a {{parametric test}}, but there are also {{nonparametric tests}} based purely on the order of data items, and also tests based on {{simulation}} or {{empirical data}}.
            [has_links] =&gt; 
        )

    [significance testing] =&gt; Array
        (
            [slug] =&gt; significance testing
            [name] =&gt; significance testing
            [canonical] =&gt; significance test
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [significant] =&gt; Array
        (
            [slug] =&gt; significant
            [name] =&gt; significant
            [canonical] =&gt; statistically significant
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [significant difference] =&gt; Array
        (
            [slug] =&gt; significant difference
            [name] =&gt; significant difference
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; When you are comparing two sets of data, or two groups within a dataset, and the difference between them (usually the difference between their respective {{means}}) reaches a chosen {{significance level}}.
            [has_links] =&gt; 
        )

    [significant effect] =&gt; Array
        (
            [slug] =&gt; significant effect
            [name] =&gt; significant effect
            [canonical] =&gt; 
            [see] =&gt; significant result
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [significant range] =&gt; Array
        (
            [slug] =&gt; significant range
            [name] =&gt; significant range
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A range of results that reach a chosen {{significance level}}.
            [has_links] =&gt; 
        )

    [significant result] =&gt; Array
        (
            [slug] =&gt; significant result
            [name] =&gt; significant result
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; When the data you have obtained in an experiment or study reaches a chosen {{significance level}}.
            [has_links] =&gt; 
        )

    [significant results] =&gt; Array
        (
            [slug] =&gt; significant results
            [name] =&gt; significant results
            [canonical] =&gt; significant result
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [simpson's paradox] =&gt; Array
        (
            [slug] =&gt; simpson's paradox
            [name] =&gt; Simpson's paradox
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; When the trend in the overall data is in the opposite direction to the {{trend}} in every group within the data. This can lead to apparently conflicting arguments from the same  data. The original paradox was first noticed over 100 years ago, but one of the most widely cited examples was at Berkeley in the 1970s, when an analysis of admissions suggested there was bias against female admissions. It turned out that in fact in each individual course women were more likely to be admitted than men, but that women were more likely to apply for the more selective courses, leading to the apparent bias.
            [has_links] =&gt; 1
        )

    [simulation] =&gt; Array
        (
            [slug] =&gt; simulation
            [name] =&gt; simulation
            [canonical] =&gt; simulation methods
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [simulation methods] =&gt; Array
        (
            [slug] =&gt; simulation methods
            [name] =&gt; simulation methods
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Running large numbers of computer simulations based on mathematical probabilities or observed data, to discover the likelihood of results that have been observed in an experiment or study.
            [has_links] =&gt; 
        )

    [simulation techniques] =&gt; Array
        (
            [slug] =&gt; simulation techniques
            [name] =&gt; simulation techniques
            [canonical] =&gt; simulation methods
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [single cause] =&gt; Array
        (
            [slug] =&gt; single cause
            [name] =&gt; single cause
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; When there is a single causal factor that is creating measurable differences, usually in a {{controlled experiment}}. For example, if we take two large groups of people matched for all relevant background and skills and then give one group a standard interface (system A) and the other group a new interface (system B), the change in interface is a {{single cause}} for any differences in the average behaviour. &lt;br&gt; Having a {{single cause}} makes analysis easier, but is difficult to achieve in HCI. This is partly because we often want to perform {{in the wild}} research or other forms of more {{ecologically valid}} study where we cannot control the context. Even where we do have a relatively controlled situation, such as the system A vs B study, the complex nature of interface design means that there will be many differences between the two, making it difficult to pin down precisely which factors give rise to an observed effect.
            [has_links] =&gt; 
        )

    [size of an effect] =&gt; Array
        (
            [slug] =&gt; size of an effect
            [name] =&gt; size of an effect
            [canonical] =&gt; effect size
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [size of effect] =&gt; Array
        (
            [slug] =&gt; size of effect
            [name] =&gt; size of effect
            [canonical] =&gt; effect size
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [size of effects] =&gt; Array
        (
            [slug] =&gt; size of effects
            [name] =&gt; size of effects
            [canonical] =&gt; effect size
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [size of the `noise'] =&gt; Array
        (
            [slug] =&gt; size of the `noise'
            [name] =&gt; size of the `noise'
            [canonical] =&gt; size of the noise
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [size of the effect] =&gt; Array
        (
            [slug] =&gt; size of the effect
            [name] =&gt; size of the effect
            [canonical] =&gt; effect size
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [size of the noise] =&gt; Array
        (
            [slug] =&gt; size of the noise
            [name] =&gt; size of the noise
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The quantity of random or {{uncontrolled factors}} in a test. Part of the {{noise-effect-number triangle}}. Reducing the size of the noise increases {{statistical power}}.
            [has_links] =&gt; 
        )

    [size of the study] =&gt; Array
        (
            [slug] =&gt; size of the study
            [name] =&gt; size of the study
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The number of  trials or participants in a study. Part of the {{noise-effect-number triangle}}. Increasing the size of the study increases {{statistical power}}.
            [has_links] =&gt; 
        )

    [snowball sample] =&gt; Array
        (
            [slug] =&gt; snowball sample
            [name] =&gt; snowball sample
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; An initial set of contacts is used as the first set of subjects, and they are then asked to suggest any of their own contacts who might take part in the survey. A danger of this approach is that there may be {{internal non-independence}} if the contacts of each individual are likely to share the feature or ability being investigated &ndash; for example contacts may well share political views, or educational background.
            [has_links] =&gt; 1
        )

    [snowball samples] =&gt; Array
        (
            [slug] =&gt; snowball samples
            [name] =&gt; snowball samples
            [canonical] =&gt; snowball sample
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [social network connection data] =&gt; Array
        (
            [slug] =&gt; social network connection data
            [name] =&gt; social network connection data
            [canonical] =&gt; social network data
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [social network data] =&gt; Array
        (
            [slug] =&gt; social network data
            [name] =&gt; social network data
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Data arising from social networks, such as the number of friends or contacts. Often social network data follows a {{power-law distribution}} and so {{statistical tests}} designed for {{approximately Normal}} data &lt;em&gt;cannot&lt;/em&gt; be used.
            [has_links] =&gt; 
        )

    [social networks] =&gt; Array
        (
            [slug] =&gt; social networks
            [name] =&gt; social networks
            [canonical] =&gt; social network data
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [software design] =&gt; Array
        (
            [slug] =&gt; software design
            [name] =&gt; software design
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Software design involves many aspects, for example the choice of internal protocols between front-end and back-end services; however, from an HCI point of view, the crucial aspects are those relating to the {{user interface}}, {{user experience}} and the way in which it fits into broader {{socio-technical systems}}.
            [has_links] =&gt; 
        )

    [software development] =&gt; Array
        (
            [slug] =&gt; software development
            [name] =&gt; software development
            [canonical] =&gt; software development process
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [software development process] =&gt; Array
        (
            [slug] =&gt; software development process
            [name] =&gt; software development process
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; There are many processes by which software is designed and created that relate to the kind of software, application domain and organisation, but those parts that involve any form of interaction with the user typically follow an {{iterative process}}. Initial design activity leads to the making of a demonstrable artefact: from storyboards and sketches to wireframes, hi-res prototypes or a running system. The artefact is subjected to testing or evaluation: {{formative evaluation}} in the early stages and {{summative evaluation}} on a pre-release product.
            [has_links] =&gt; 
        )

    [sonic enhancement] =&gt; Array
        (
            [slug] =&gt; sonic enhancement
            [name] =&gt; sonic enhancement
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A way to make clicking an onscreen button more evident by adding sound to simulate a physical effect. More generally, many aspects of an interface can be enhanced by additional audio feedback, such as the crunching paper sound that some systems use when a file is thrown into the (desktop) trash bin. While sometimes provided for aesthetic reasons, this may also add feedback.
            [has_links] =&gt; 
        )

    [spss] =&gt; Array
        (
            [slug] =&gt; spss
            [name] =&gt; SPSS
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A common software package used for statistical calculations. It has been developed over many years and is now owned by IBM.
            [has_links] =&gt; 1
        )

    [standard deviation] =&gt; Array
        (
            [slug] =&gt; standard deviation
            [name] =&gt; standard deviation (s.d., &amp;sigma;)
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The square root of {{variance}} and probably the most commonly used measure of {{variation}}. The principal alternative is the {{interquartile range}}, which is more robust, but less mathematically tractable than the {{standard deviation}}.
            [has_links] =&gt; 1
        )

    [standard deviation of the population] =&gt; Array
        (
            [slug] =&gt; standard deviation of the population
            [name] =&gt; standard deviation of the population
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; When you have (or imagine you have) access to the entire {{population}} of interest, perhaps every person in a country or every user of a device, and can calculate the {{standard deviation}} of some facet of them all; that is &amp;radic; ( &amp;;Sigma (x&lt;sub&gt;i&lt;/sub&gt;&ndash;&amp;mu;)&lt;sup&gt;2&lt;/sup&gt;))/N &ndash; where x&lt;sub&gt;i&lt;/sub&gt; are the individual measurements of the facets, &amp;mu; is the arithmetic {{mean}} and N the number of people in the entire {{population}}.
            [has_links] =&gt; 
        )

    [standard deviation of the sample] =&gt; Array
        (
            [slug] =&gt; standard deviation of the sample
            [name] =&gt; standard deviation of the sample
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The calculation&nbsp;of&nbsp;an {{estimate}}&nbsp;of&nbsp;the {{standard deviation of the population}}&nbsp;that&nbsp;corrects for {{bias}} when using the data in a {{sample}}.  If you use the {{sample}} data to calculate the simple formula for the {{standard deviation}}, this slightly underestimates the true {{standard deviation of the population}}. This is because the {{mean}} also has to be calculated from the {{sample}}. It is possible to {{model the bias}} and then the {{bias}} can be corrected by dividing the sum of squares by N&ndash;1 instead of N. &lt;br&gt;{{standard deviation of the sample}} =  &amp;radic; ( &amp;;Sigma (x&lt;sub&gt;i&lt;/sub&gt;&ndash;&amp;mu;)&lt;sup&gt;2&lt;/sup&gt;))/N&ndash;1
            [has_links] =&gt; 
        )

    [standard deviations] =&gt; Array
        (
            [slug] =&gt; standard deviations
            [name] =&gt; standard deviations
            [canonical] =&gt; standard deviation
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [standard error] =&gt; Array
        (
            [slug] =&gt; standard error
            [name] =&gt; standard error (s.e.)
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The variability of the {{mean}} is called the {{standard error}} of the mean (or just s.e.), and is often drawn as little &lsquo;{{error bars}}&rsquo; on graphs or histograms, to give you some idea of the accuracy of the average measure. If the mean is averaged over a {{sample}} of N independent items, the standard error is &amp;sigma;/&amp;radic;N &ndash; where &amp;sigma; is the {{standard deviation of the sample}} or the {{standard deviation}} of a single item if that is already known.
            [has_links] =&gt; 
        )

    [starting hypothesis] =&gt; Array
        (
            [slug] =&gt; starting hypothesis
            [name] =&gt; starting hypothesis
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The original hypothesis that an experiment is designed to test. The gold standard statistical process is usually deemed to be that only {{hypotheses}} determined before the study should be admissible. &lt;br&gt;In practice researchers may notice new patterns in their data and then test a new {{hypothesis}} based on this, a process termed {{HARKing}} or {{Hypothesis After the Result is Known}}. This can be dangerous: it is a form of {{cherry picking}} and may lead to spurious, purely random results. One of the reasons for {{pre-registration}} is to discourage this practice. The ideal is that if you do notice such patterns you should perform a fresh study to investigate them, that is a fresh study where the observed pattern from the previous experiment becomes the {{original hypothesis}} of the new one. &lt;br&gt;However, sticking purely to the {{original hypothesis}} is a counsel of perfection that is not always appropriate or achievable, especially when dealing with data such as epidemiological data or historical data that it is impossible to recreate, or where the original data collection is very expensive or difficult to access. In such circumstances the crucial thing is to be aware that you are creating a new {{hypothesis}} and then, where possible, perform a suitable {{multiplicity adjustment}} such as a {{Bonferroni correction}}, and definitely be very clear in any published work that this was &lt;em&gt;not&lt;/em&gt; the {{original hypothesis}}.
            [has_links] =&gt; 
        )

    [statistical `crisis'] =&gt; Array
        (
            [slug] =&gt; statistical `crisis'
            [name] =&gt; statistical `crisis'
            [canonical] =&gt; statistical crisis
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [statistical analyses] =&gt; Array
        (
            [slug] =&gt; statistical analyses
            [name] =&gt; statistical analyses
            [canonical] =&gt; statistical analysis
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [statistical analysis] =&gt; Array
        (
            [slug] =&gt; statistical analysis
            [name] =&gt; statistical analysis
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Analysis of measured or sampled quantitative data, with the goal of obtaining a level of quantified understanding of something in the {{real world}} that is unknown. The three major kinds of statistical analysis are {{hypothesis testing}}, {{confidence intervals}}, and {{Bayesian statistics}}.
            [has_links] =&gt; 
        )

    [statistical crisis] =&gt; Array
        (
            [slug] =&gt; statistical crisis
            [name] =&gt; statistical crisis
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; There has been substantial publicity surrounding poor statistical practice in published work, casting doubt on how many published results that appear to be {{statistically significant}} are in fact true or {{reproducible}}. This is sometimes called the {{statistical crisis}}. It is treated as a recent phenomenon, focused predominantly on traditional {{hypothesis testing}}, but that largely ignores the fact that potential problems in traditional statistics have been well known for decades and are usually connected to poor use or interpretation, not intrinsic weaknesses in the statistical techniques themselves. Furthermore, the same problems are equally possible with alternatives such as {{Bayesian statistics}}. There is a positive side to this recent publicity in that it has focused many fields of science on improving statistical practice. However, the danger is that researchers either abandon statistical testing where it is in fact appropriate, or adopt alternatives whose weaknesses they understand even less.
            [has_links] =&gt; 
        )

    [statistical inference] =&gt; Array
        (
            [slug] =&gt; statistical inference
            [name] =&gt; statistical inference
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This is the formal name for what I have referred to as the {{job of statistics}}, that is attempting to make some forms of {{estimate}} or {{inference}} about the real world based on measured data. It contrasts with {{descriptive statistics}}, which is concerned with various metrics about the data itself. 
            [has_links] =&gt; 
        )

    [statistical method] =&gt; Array
        (
            [slug] =&gt; statistical method
            [name] =&gt; statistical method
            [canonical] =&gt; statistical analysis
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [statistical power] =&gt; Array
        (
            [slug] =&gt; statistical power
            [name] =&gt; statistical power
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The probability of having a {{false negative}, i.e. how likely it is that if there is some real effect it shows up in the statistical analysis of the data you have measured. If an experiment has high power the statistics have fewer {{false negatives}}, and real effects are more likely to be seen. The power is a combination of the experiment we have performed and choosing the correct statistical analysis technique for the data. As an example of the latter {{nonparametric statistics}} usually have lower power than {{parametric statistics}} as they make fewer assumptions, so if you know that your data is {{Binomial}} or {{approximately Normal}} it is better to use tests based on these distributions. The {{noise-effect-number triangle}} offers various tactics to increase the power of a study.
            [has_links] =&gt; 1
        )

    [statistical reasoning] =&gt; Array
        (
            [slug] =&gt; statistical reasoning
            [name] =&gt; statistical reasoning
            [canonical] =&gt; 
            [see] =&gt; statistical inference
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [statistical significance] =&gt; Array
        (
            [slug] =&gt; statistical significance
            [name] =&gt; statistical significance
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In some contexts this just means the {{significance level}} or {{p-value}}, but it can also mean the broader idea or quality of something that uses the process of {{significance testing}}, for example &quot;a test of {{statistical significance}}&quot;.
            [has_links] =&gt; 
        )

    [statistically significant] =&gt; Array
        (
            [slug] =&gt; statistically significant
            [name] =&gt; statistically significant
            [canonical] =&gt; statistical significance
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [statistically significant difference] =&gt; Array
        (
            [slug] =&gt; statistically significant difference
            [name] =&gt; statistically significant difference
            [canonical] =&gt; significant difference
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [statistically significant effect] =&gt; Array
        (
            [slug] =&gt; statistically significant effect
            [name] =&gt; statistically significant effect
            [canonical] =&gt; significant effect
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [statistically significant results] =&gt; Array
        (
            [slug] =&gt; statistically significant results
            [name] =&gt; statistically significant results
            [canonical] =&gt; significant result
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [steering task] =&gt; Array
        (
            [slug] =&gt; steering task
            [name] =&gt; steering task
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This can refer to literal steering of a tractor or car (or simulation of these), or may refer to a form of {{pointing tasks}} where the pointer has to be navigated along a path. In the former case, the steering may be used as a {{distractor task}} for another {{primary task}} that is the real purpose of an experiment.
            [has_links] =&gt; 
        )

    [student's t distribution] =&gt; Array
        (
            [slug] =&gt; student's t distribution
            [name] =&gt; Student's t distribution
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The {{theoretical distribution}} of the value calculated by the {{Student's t-test}}. Mathematically this is a {{Normal distribution}} divided by the square root of a {{Chi-squared distribution}}. This is because the {{sample variance}} follows the  {{Chi-squared distribution}}. 
            [has_links] =&gt; 
        )

    [student's t-test] =&gt; Array
        (
            [slug] =&gt; student's t-test
            [name] =&gt; Student's t-test
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Student's t-test is a {{statistical test}} that is usually used to compare the {{means}} of two or more groups to see if they are different. For example, comparing the average {{task completion time}} for a group of people using an old version of a system vs that of a second group using a new improved user interface. It is calculated by taking the difference of the means and dividing by an estimate of the {{standard deviation}} of the difference obtained from the {{standard deviation of the sample}} for each of the groups. &lt;br&gt;t = (&amp;mu;&lt;sub&gt;1&lt;/sub&gt; &ndash; &amp;mu;&lt;sub&gt;2&lt;/sub&gt;)  /  &amp;radic;(&amp;sigma;&lt;sub&gt;1&lt;/sub&gt;&lt;sup&gt;2&lt;/sup&gt;/N&lt;sub&gt;1&lt;/sub&gt; &ndash; &amp; sigma;&lt;sub&gt;2&lt;/sub&gt;&lt;sup&gt;2&lt;/sup&gt;/N&lt;sub&gt;2&lt;/sub&gt;) &lt;br&gt;The {{theoretical distribution}} of the value of this t-test follows the {{Student's t distribution}}. The t-test depends on the data being {{approximately Normal}}, but is fairly {{robust}} if this is not perfect. &lt;br&gt;The {{theoretical t distribution}} is a Normal variable divided by the square root of an independent variable with the {{Chi-squared distribution}}. This is because the {{sample variance}} follows the  {{Chi-squared distribution}}. 
            [has_links] =&gt; 
        )

    [subconscious reactions] =&gt; Array
        (
            [slug] =&gt; subconscious reactions
            [name] =&gt; subconscious reactions
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; One of two systems of thought identified in psychology: subconscious and conscious, sometimes also called System 1 and System 2 respectively. Subconscious reactions are based on semi probabilistic associations built up through experience. Sometimes subconscious reactions lead to an immediate action, such as jumping out of the way if we hear the sound of an approaching car; at other times they are more about a feeling or gut reaction &ndash; we are aware of the reaction, but not the reasons for it. Contrast with {{conscious thinking}}.
            [has_links] =&gt; 1
        )

    [subjects] =&gt; Array
        (
            [slug] =&gt; subjects
            [name] =&gt; subjects
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The humans involved in a study. Because the term may sound rather clinical or instrumental, some researchers prefer the terms &lsquo;user&rsquo;, or &lsquo;participant&rsquo; if there is a richer engagement. Both these alternatives have their own connotations, and 'subject' remains the most commonly used in statistics.
            [has_links] =&gt; 
        )

    [summative evaluation] =&gt; Array
        (
            [slug] =&gt; summative evaluation
            [name] =&gt; summative evaluation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Focused on checking that the system works and is good enough, summative evaluation is performed at the end of the {{software development process}} on a pre-release product. It may be related to contractual obligations (whether the system meets agreed specifications) or may compare the system with others. Less formally, summative evaluation may simply be an assessment that enough work has been done based on the cumulative evidence from earlier stages. Contrast with {{formative evaluation}}.
            [has_links] =&gt; 
        )

    [summative evaluations] =&gt; Array
        (
            [slug] =&gt; summative evaluations
            [name] =&gt; summative evaluations
            [canonical] =&gt; summative evaluation
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [survey data] =&gt; Array
        (
            [slug] =&gt; survey data
            [name] =&gt; survey data
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Data obtained by asking people questions, often utilising an electronic or paper form or through guided interviews. The volume of survey data can often be far higher than is provided by more intensive methods such as {{direct observation}}, {{usability tests}} or more in depth interviews. Closed questions, whose answers are chosen from a fixed set of options, are useful as they generate easy to analyse statistical data, but they often only tell you &lt;em&gt;what&lt;/em&gt; is happening. Typically you need more {{qualitative data}} to help you understand &lt;em&gt;why&lt;/em&gt; and &lt;em&gt;how&lt;/em&gt;, including the {{motivation}} and {{mechanism}}. Such data may be obtained through {{open questions}} in the survey, or through complementary methods.
            [has_links] =&gt; 
        )

    [systematic effects] =&gt; Array
        (
            [slug] =&gt; systematic effects
            [name] =&gt; systematic effects
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Aspects of data collection or analysis that are in some way connected to the thing that you are trying to measure, and hence interfere with analysis or lead to misleading effects or {{bias}}. There are many random or uncontrolled effects that influence the data you obtain, but so long as these are unrelated to the thing you are trying to measure ({{independent}}), they can be treated as {{random noise}}. Though potentially adding {{variability}}, they will eventually average out given a large enough sample. In contrast, systematic effects may cause {{bias}}. &lt;br&gt;For example, suppose you are using a laser to measure the heights of people coming and going from two different tourist attractions in London. As they will be wearing shoes and potentially hats, this will not be a true measure of height, but for the purposes of comparison between, say, Buckingham Palace and The Houses of Parliament, it is probably fine as there is no reason to suppose visitors at either wear substantially different attire. If, however, this method were used to compare the tourists at Buckingham Palace with the guards, since the guards would be likely to be wearing tall Bearskin hats, the size of hat would be a systematic effect, leading to {{bias}} and overestimating the height of guards compared with tourists. &lt;br&gt;Note that due to the nature of randomness apparently systematic effects sometimes happen by chance.
            [has_links] =&gt; 
        )

    [t-test] =&gt; Array
        (
            [slug] =&gt; t-test
            [name] =&gt; t-test
            [canonical] =&gt; 
            [see] =&gt; student's t-test
            [description] =&gt; 
            [has_links] =&gt; 1
        )

    [t-tests] =&gt; Array
        (
            [slug] =&gt; t-tests
            [name] =&gt; t-tests
            [canonical] =&gt; t-test
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [tail] =&gt; Array
        (
            [slug] =&gt; tail
            [name] =&gt; tail
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Tail has two meanings in statistics. One is the long spread of low frequency values in a {{long-tail distribution}}, which includes very extreme values at the end(s) of the scale. The other is the use of the term to indicate in which directions the data is of interest &ndash; one, or both; see {{one-tailed test}} and {{two-tailed test}}.
            [has_links] =&gt; 
        )

    [tail heavy] =&gt; Array
        (
            [slug] =&gt; tail heavy
            [name] =&gt; tail heavy
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; When the probability (or observed frequency) of data items far from the central value is large. This can be true of a {{symmetric distribution}} such as the {{Cauchy distribution}}, but is more often associated with asymmetric {{long-tail distributions}} especially the {{power-law distribution}} found in areas such as {{network data}}.
            [has_links] =&gt; 
        )

    [target size] =&gt; Array
        (
            [slug] =&gt; target size
            [name] =&gt; target size
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In a {{pointing task}} this is the size of the area you want to point at, such as a button or icon. It is used in {{Fitts' Law}} to predict the time it takes to locate the target.
            [has_links] =&gt; 
        )

    [targeted tasks] =&gt; Array
        (
            [slug] =&gt; targeted tasks
            [name] =&gt; targeted tasks
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Tasks designed to deliberately  expose the effects of an intervention by focusing on the aspects of the system that are being tested, minimising interaction with other elements. For example, if you have added a feature to help users keep track of what they are doing, a {{targeted task}} could include a lot of interruptions. Targeted tasks increase {{effect}} in the {{noise-effect-number triangle}} and so increase {{statistical power}}. However, targeting can be misleading if you are not careful, as the {{targeted task}} may not be representative of the full range of tasks that are likely to be carried out; that is {{generalisation}} can be difficult.
            [has_links] =&gt; 
        )

    [targeted user] =&gt; Array
        (
            [slug] =&gt; targeted user
            [name] =&gt; targeted user
            [canonical] =&gt; targeted user group
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [targeted user group] =&gt; Array
        (
            [slug] =&gt; targeted user group
            [name] =&gt; targeted user group
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A group for whom we believe we will see the maximum effect is deliberately chosen. For example, a new menu system would be tested more effectively for short-term memory requirements on an elderly user group who are more likely to have some short-term memory loss, rather than on a younger user group. Targeted user groups increase {{effect}} in the {{noise-effect-number triangle}} and so increase {{statistical power}}. However, targeting can be misleading if you are not careful as the {{targeted user group}} may not be representative of the full range of users who are likely to use the system, therefore {{generalisation}} can be difficult.
            [has_links] =&gt; 
        )

    [targeted user groups] =&gt; Array
        (
            [slug] =&gt; targeted user groups
            [name] =&gt; targeted user groups
            [canonical] =&gt; targeted user group
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [task completion time] =&gt; Array
        (
            [slug] =&gt; task completion time
            [name] =&gt; task completion time
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The time taken to complete a specified task is often used as a measure when comparing two systems. This may be a high-level task, such as finding a product on an online shopping site; or a low-level task, such as a {{pointing task}} modelled by  {{Fitts' Law}}. 
            [has_links] =&gt; 
        )

    [task completion times] =&gt; Array
        (
            [slug] =&gt; task completion times
            [name] =&gt; task completion times
            [canonical] =&gt; task completion time
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [tasks] =&gt; Array
        (
            [slug] =&gt; tasks
            [name] =&gt; tasks
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Experimental or study tasks are the actions that users are specifically asked to perform. Measurements such as {{task completion time}} can be observed and recorded. Tasks can be manipulated to gain {{statistical power}} using the {{noise-effect-number triangle}}.
            [has_links] =&gt; 
        )

    [temporal data] =&gt; Array
        (
            [slug] =&gt; temporal data
            [name] =&gt; temporal data
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Data recorded over a period of time, where the time at which events occur is a critical part of the data. This can include very long-term phenomena such as the level of usage of a web site, shorter term such as the trace of user interactions while performing a task, or very fine scale such as some biological measurements. There are special statistical techniques for modelling {{time series}} data.
            [has_links] =&gt; 
        )

    [the `job of statistics'] =&gt; Array
        (
            [slug] =&gt; the `job of statistics'
            [name] =&gt; the `job of statistics'
            [canonical] =&gt; the job of statistics
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [the `job of statistics`] =&gt; Array
        (
            [slug] =&gt; the `job of statistics`
            [name] =&gt; the `job of statistics`
            [canonical] =&gt; the job of statistics
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [the job of statistics] =&gt; Array
        (
            [slug] =&gt; the job of statistics
            [name] =&gt; the job of statistics
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; We may be interested in some facet of the {{real world}}, for example how much faster it is to use one system compared with another. However, we typically only have some sort of partial or noisy measured data. {{Probability theory}} can help us to assess the {{likelihood}} of the {{measurement}} given the state of the world. However, we do not know the real state of the world, only the measurement. The {{job of statistics}} is to reverse this, to create some sort of {{estimate}} of, or knowledge about, the {{real world}} from the {{measurement}}. Because the measured data is subject to {{random}} effects, the knowledge we obtain from statistics will always be {{uncertain}}.
            [has_links] =&gt; 
        )

    [theoretical distribution] =&gt; Array
        (
            [slug] =&gt; theoretical distribution
            [name] =&gt; theoretical distribution
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A mathematical calculation of the probability of an event in the real world; examples are the {{Normal distribution}} (for heights) and the {{Binomial distribution}} (for coin tosses). Contrast with {{empirical distribution}}.
            [has_links] =&gt; 
        )

    [theoretical distributions] =&gt; Array
        (
            [slug] =&gt; theoretical distributions
            [name] =&gt; theoretical distributions
            [canonical] =&gt; theoretical distribution
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [theoretical understanding] =&gt; Array
        (
            [slug] =&gt; theoretical understanding
            [name] =&gt; theoretical understanding
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; In analysis of {{qualitative data}}, theoretical understanding provides additional information; for example, cognitive and psychological understanding is required to build a model of how the user is behaving and why. Theoretical analysis may be based on past data and experiments, but also involves some level of analysis or reasoning either by the analyst, or drawing on previously published literature.
            [has_links] =&gt; 
        )

    [threshold effect] =&gt; Array
        (
            [slug] =&gt; threshold effect
            [name] =&gt; threshold effect
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A limiting condition that can cause {{nonlinearity}}; for example, plants growing near an electric fence may be electrocuted when they reach a certain height, leading to a chopped off {{Normal distribution}}.
            [has_links] =&gt; 
        )

    [tossing coins] =&gt; Array
        (
            [slug] =&gt; tossing coins
            [name] =&gt; tossing coins
            [canonical] =&gt; coin tossing
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [traditional methods] =&gt; Array
        (
            [slug] =&gt; traditional methods
            [name] =&gt; traditional methods
            [canonical] =&gt; traditional statistics
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [traditional statistics] =&gt; Array
        (
            [slug] =&gt; traditional statistics
            [name] =&gt; traditional statistics
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The statistical techniques of {{hypothesis testing}} and {{confidence intervals}}, which have the same underlying mathematics and share a common philosophical stance. Contrast with {{Bayesian statistics}}.

            [has_links] =&gt; 
        )

    [traditional stats] =&gt; Array
        (
            [slug] =&gt; traditional stats
            [name] =&gt; traditional stats
            [canonical] =&gt; traditional statistics
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [transformation] =&gt; Array
        (
            [slug] =&gt; transformation
            [name] =&gt; transformation
            [canonical] =&gt; transforming data
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [transforming data] =&gt; Array
        (
            [slug] =&gt; transforming data
            [name] =&gt; transforming data
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; When you apply some sort of calculation to data items in order to transform them into a form more suitable for analysis. For example, if you have a 5 point {{Likert scale}} you might transform this into {{binary data} by mapping the top two values (4 and 5) into 'satisfied' and the rest (1, 2, and 3) into 'not satisfied'. {{Temporal data}} is often asymmetric and furthermore larger items typically have greater {{variance}}. However, if you take the {{logarithm}} of the times, the resulting data is often {{approximately Normal}}. This last transformation is so common that data such as {{reaction times}} are termed {{log-Normal}} data.
            [has_links] =&gt; 
        )

    [transparent statistics] =&gt; Array
        (
            [slug] =&gt; transparent statistics
            [name] =&gt; Transparent Statistics
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; At the {{CHI conference}} a series of workshops on &quot;Transparent Statistics in Human&ndash;Computer Interaction&quot; is focused on creating guidance on best practice reporting of statistics.
            [has_links] =&gt; 1
        )

    [treemap] =&gt; Array
        (
            [slug] =&gt; treemap
            [name] =&gt; TreeMap
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The TreeMap is a {{constant area visualisation}} invented by Ben Shneiderman for hierarchical data with some form of numerical attribute, for example the sizes of files and folders on a computer disk. The most basic form starts with a rectangular area and then recursively divides the area alternatively horizontally using the total values for each level.
            [has_links] =&gt; 1
        )

    [two-tailed test] =&gt; Array
        (
            [slug] =&gt; two-tailed test
            [name] =&gt; two-tailed test
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A test where data is compared in two directions. For example, assume you are choosing which of two systems, A and B, to purchase for you company and want to check if either is measurably better than the other. The results matter in both directions: whether A is better than B or whether B is better than A. If comparing {{response times}} your test has to look both at when System A is significantly faster &lt;em&gt;and&lt;/em&gt; at when it is significantly slower. Contrast with {{one-tailed test}}.
            [has_links] =&gt; 
        )

    [two-tailed tests] =&gt; Array
        (
            [slug] =&gt; two-tailed tests
            [name] =&gt; two-tailed tests
            [canonical] =&gt; two-tailed test
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [type i error] =&gt; Array
        (
            [slug] =&gt; type i error
            [name] =&gt; Type I error
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This is a term from {{hypothesis testing}} where you wrongly believe there to be an effect when there is none; that is a {{false positive result}}. The {{significance level}} you choose helps to control for this. There is only a 1 in 20 chance that you hit a {{p-value}} of 5% by chance when there is no effect; that is, if you choose a significance level of 5%, the probability of a {{Type I error}} is 0.05 (1 in 20). Similarly if you choose a 1% {{significance level}} then the probability of the Type I error is only 0.01 (in in a hundred).
            [has_links] =&gt; 
        )

    [type ii error] =&gt; Array
        (
            [slug] =&gt; type ii error
            [name] =&gt; Type II error
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This is a term from {{hypothesis testing}} when you fail to detect a real effect. This typically happens because the {{statistical power}} of your experiment or analysis method is too weak. Often researchers and paper reviewers worry less about a {{Type II error}} than a {{Type I error}}. This is because if you think something is true wrongly and then base future science on the false belief everything you build on it will potentially also be fallacious. However, {{Type II errors}} are also critical. A certain proportion of {{statistically significant}} results will occur by chance, so if the {{statistical power}} of experiments is too weak and there are too many {{Type II errors}} there will be too few true positive results published compared with the {{false negatives}}.
            [has_links] =&gt; 
        )

    [typical user] =&gt; Array
        (
            [slug] =&gt; typical user
            [name] =&gt; typical user
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; There is, of course, no such thing as a typical user &ndash; everyone is different. However, one does have an idea of a central user, for instance a nineteen-year-old student for a University virtual learning system. Furthermore, when doing any form of statistics based on a user study the user you are trying to understand is not so much the specific people taking part in your study, or even the collection of all current users, but some kind of idealised conception of a potential user.
            [has_links] =&gt; 
        )

    [uk income distribution] =&gt; Array
        (
            [slug] =&gt; uk income distribution
            [name] =&gt; UK income distribution
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Graphs of income and wealth in most countries form a {{long-tail distribution}}, that is there are a large number of people with low or middle sized incomes and then a small number of people with very large incomes. The upper part of the graph typically follows closely to a {{power law}}, but at the lower end the effect of welfare benefits mean that there are few people with absolutely tiny incomes. The result for the UK is a highly {{asymmetric distribution}} that starts with counts close to zero (few people with no income), rises towards a {{mode}} at around &pound;300 per week, and then drops rapidly, with only about 5% of people earning more than a thousand pounds per week. As with most asymmetric distributions, the {{median}} income (~&pound;430 pw) is lower than the {{mean}} (~&pound;530 pw). Because of the distorting effect of the long tail on the arithmetic {{mean}}, when newspapers and government statistics quote the {{average}} income they are usually referring to the {{median}} income.
            [has_links] =&gt; 1
        )

    [unbiased] =&gt; Array
        (
            [slug] =&gt; unbiased
            [name] =&gt; unbiased
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Informally this refers to personal fairness, but within statistics it also has a more formal meaning. We say an {{estimate}} is {{unbiased}} if the long term or theoretical average of the estimate is the true value of the {{real world}} thing it is trying to measure. For example, if we want to know the average height of someone in Malaysia, we might randomly sample 1000 people from Malaysia and then take the arithmetic {{mean}} of the measured heights. It turns out that the {{sample mean}} is an {{unbiased estimator}} of the {{population mean}}. For any given sample of 1000 people this might be slightly more, or slightly less than the true national average, but if one were to take many such samples the average value of the sample average would be the true average!
            [has_links] =&gt; 
        )

    [unbounded] =&gt; Array
        (
            [slug] =&gt; unbounded
            [name] =&gt; unbounded
            [canonical] =&gt; unbounded data
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [unbounded above] =&gt; Array
        (
            [slug] =&gt; unbounded above
            [name] =&gt; unbounded above
            [canonical] =&gt; unbounded data
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [unbounded data] =&gt; Array
        (
            [slug] =&gt; unbounded data
            [name] =&gt; unbounded data
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Data can be potentially unbounded, if it is not {{finite data}}. Unbounded data may be unbounded in one or both directions. Examples of the former are: (coin tosses, {{discrete data}}) &ndash; the number of heads thrown before the first tail is thrown; (waiting for a bus, {{continuous data}}) &ndash; the length of time before the bus comes. These are both bounded below but unbounded above. An example of data that is unbounded in either direction is the difference between heights, which can be arbitrarily positive or negative. 
            [has_links] =&gt; 
        )

    [unbounded tail] =&gt; Array
        (
            [slug] =&gt; unbounded tail
            [name] =&gt; unbounded tail
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; When one {{tail}} of a distribution is {{unbounded}}, that is it has no theoretical maximum (or minimum). There are distributions where the tail can be quite long &ndash; there may be a small probability of very large or very small values, but it is still finite: for example, if you look at the number of double sixes when rolling two dice 100 times, the average is around 3, but there is a very small chance of there being a full 100 double sixes. However, there can never be more than 100. In the end the tail is {{bounded}}. In contrast, consider just starting tossing a coin until you reach the first tail and count how many heads you get before the first tail &ndash; that is a {{Negative Binomial distribution}}. Here the average value is just 1, but there is no maximum, you could, in principle, go on tossing heads for ever. That is the {{Negative Binomial distribution}} has an {{unbounded tail}}.
            [has_links] =&gt; 
        )

    [unbounded values] =&gt; Array
        (
            [slug] =&gt; unbounded values
            [name] =&gt; unbounded values
            [canonical] =&gt; 
            [see] =&gt; unbounded data
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [uncertain knowledge] =&gt; Array
        (
            [slug] =&gt; uncertain knowledge
            [name] =&gt; uncertain knowledge
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Although statistics works with actual measurements, it can only discover information about probable outcomes, never absolute certainty. See also {{likelihood}}.
            [has_links] =&gt; 
        )

    [uncontrolled factors] =&gt; Array
        (
            [slug] =&gt; uncontrolled factors
            [name] =&gt; uncontrolled factors
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Factors that are unrelated to the thing being measured, but that affect the result. For example, if people are asked to perform a website navigation task, and some have better web skills than others, or better spatial ability. Although such effects can sometimes be measured they must often be treated as effectively random.
            [has_links] =&gt; 
        )

    [uncontrolled variation] =&gt; Array
        (
            [slug] =&gt; uncontrolled variation
            [name] =&gt; uncontrolled variation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; These are various factors that we are unable to control, for example the weather on the day of an outdoor user study or the level of expertise of test subjects chosen at random. So long as there are no {{systematic effects}} these can be treated as part of {{random noise}}.
            [has_links] =&gt; 
        )

    [uniform distribution] =&gt; Array
        (
            [slug] =&gt; uniform distribution
            [name] =&gt; uniform distribution
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A {{uniform distribution}}  is where the {{probability}} of every outcome is the same. 

For {{discrete data}}, this is a probability of 1/N for each value, where N is the number of values.  In the case of data 1...N  the {{mean}} is (N+1)/2 and the {{variance}} is (N+1)(N-1)/12 

For {{bounded}} {{continuous data}} the probability is 1/W where W is the width of the data..  In the case of continuous data over [0,W] the {{mean}} is W/2 and the {{variance}} is W&lt;sup&gt;2&lt;/sup&gt;/12 
&lt;br&gt;&lt;br&gt;
Note that in the continuous case, the definition of 'uniform' depends on the form of the parameter.  For example, if dealing with sound waves a uniform distrubution over the wavelength is a different {{probability distribution}} to one that is uniform over the frequency.
            [has_links] =&gt; 
        )

    [uniform prior] =&gt; Array
        (
            [slug] =&gt; uniform prior
            [name] =&gt; uniform prior
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A {{prior distribution}} where the {{probability}} of every outcome is the same. For {{discrete data}}, this is a probability of 1/N for each value, where N is the number of values. If the data is {{unbounded}} there is no meaningful {{uniform prior}}, and the {{Cauchy distribution}} is often used as a best alternative, but for this you need to decide a midpoint and spread, either of which may introduce {{unintentional bias}}. You also have to be careful about the perspective from which data is measured. For example, with sound do you choose a prior that is uniform over power or decibels? Because the latter is the {{logarithm}} of the former, the uniform prior of one is not uniform for the other.  This is particularly important in {{Bayesian statistics}} as the choice of prior can make a big difference to the results.

See also {{uniform distribution}}.
            [has_links] =&gt; 
        )

    [uniformity fallacy] =&gt; Array
        (
            [slug] =&gt; uniformity fallacy
            [name] =&gt; uniformity fallacy
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The belief or gut feeling that, overall, results will even out, which leads people to reject clumped results as artificial or to suppose that a run of bad luck will be followed by a lucky break.
            [has_links] =&gt; 
        )

    [unintentional bias] =&gt; Array
        (
            [slug] =&gt; unintentional bias
            [name] =&gt; unintentional bias
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Sometimes an advert or a political message may deliberately (or intentionally) put some sort of spin on data and mislead those seeing it. As researchers and professionals, we would all, I hope,  try to avoid this deliberate bias, but we may still have {{unintentional bias}} due to the choice of {{sample}} or other aspects of our study methods. For example, if you interview people in an Apple store and ask them for their favourite phone model, the result may be somewhat skewed! Of course, unintentional bias is not always so obvious; for example, if you interview members of a particular social media network, it may include more younger people, or more well-off people, and this might have an impact on music taste or even favourite food.
            [has_links] =&gt; 
        )

    [unrepeatable event] =&gt; Array
        (
            [slug] =&gt; unrepeatable event
            [name] =&gt; unrepeatable event
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Sometimes we can repeat an identical event multiple times, for example repeatedly tossing a coin. However, some events are unrepeatable, for example, the weather on 3rd January 2020. Crucially once such an event has occurred it is, by definition, not {{ontologically random}} &ndash; its state is already determined. However, it may still be {{epistemologically random}}, unknown to you, and hence you may sometimes still treat it as having probabilistic or statistical properties.
            [has_links] =&gt; 
        )

    [upper bound] =&gt; Array
        (
            [slug] =&gt; upper bound
            [name] =&gt; upper bound
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The largest possible value that a variable or dataset can have. For example, when rolling a  die, the upper bound is 6. In contrast, some data is {{unbounded above}}, that is there is no meaningful absolute maximum.
            [has_links] =&gt; 
        )

    [upper quartile] =&gt; Array
        (
            [slug] =&gt; upper quartile
            [name] =&gt; upper quartile
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The value such that 75% (three quarters) of the data values are less and 25% more. It is one of three {{quartiles}}, the others being the {{median}} (50% point) and {{lower quartile}} (25% point).
            [has_links] =&gt; 
        )

    [usability evaluation] =&gt; Array
        (
            [slug] =&gt; usability evaluation
            [name] =&gt; usability evaluation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This is about evaluating how usable a system is; the term can be read quite broadly, or may be distinguished from evaluating {{user experience}}.There are many ways of evaluating usability: {{expert evaluation}} techniques are useful for early design or when users are difficult to contact, but typically the gold standard is expected to use real users. When evaluating with real users there are still many options from {{controlled laboratory experiments}} to {{in the wild}} studies.
            [has_links] =&gt; 
        )

    [user experience] =&gt; Array
        (
            [slug] =&gt; user experience
            [name] =&gt; user experience
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Aspects of a system that affect the user's emotional state; that is creating a (usually) positive experience of using the system.
            [has_links] =&gt; 
        )

    [user experience design] =&gt; Array
        (
            [slug] =&gt; user experience design
            [name] =&gt; user experience design
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This refers to the parts of the design process and the methods used to ensure a desired {{user experience}}.
            [has_links] =&gt; 
        )

    [user experience studies] =&gt; Array
        (
            [slug] =&gt; user experience studies
            [name] =&gt; user experience studies
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; These are studies that attempt to {{evaluate}} or otherwise understand the {{user experience}} of a study. It is possible to measure {{objective}} aspects such as heart rate, but the emotional and personal nature of {{user experience}} mean that more {{qualitative methods}} are often used, or various forms of subjective questionnaire. 
            [has_links] =&gt; 
        )

    [user experience testing] =&gt; Array
        (
            [slug] =&gt; user experience testing
            [name] =&gt; user experience testing
            [canonical] =&gt; user experience studies
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [user interface] =&gt; Array
        (
            [slug] =&gt; user interface
            [name] =&gt; user interface
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The part of a system that the user interacts with. Typically this is some form of screen-based interface with buttons, menus, etc., such as an app or desktop graphical user interface. However, interaction may be through alternatives such as speech, gesture or physical things in the environment that are sensed by the computer.
            [has_links] =&gt; 
        )

    [user interface development] =&gt; Array
        (
            [slug] =&gt; user interface development
            [name] =&gt; user interface development
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The process of creating the user-facing aspects of a system, from initial scoping and design through to coding and deployment. Typically {{iterative development}} is used where there are cycles where some form of {{prototype}} or {{design sketch}} is evaluated in a {{user study}} or {{expert evaluation}}. The design is then altered in the light of the {{formative evaluation}}, leading into another cycle.
            [has_links] =&gt; 
        )

    [user interface properties] =&gt; Array
        (
            [slug] =&gt; user interface properties
            [name] =&gt; user interface properties
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This usually refers to specific generic properties that influence the {{usability}} or {{user experience}} of a system in some way. For example, it is usually the case that interactions are more easily learnt if the interface is {{consistent}}. A {{user interface property}} can be a form of {{generative artefact}} when one deliberately tries to create a system that adheres to the principle, and as such can be hard to {{evaluate}}.
            [has_links] =&gt; 
        )

    [user interface testing] =&gt; Array
        (
            [slug] =&gt; user interface testing
            [name] =&gt; user interface testing
            [canonical] =&gt; 
            [see] =&gt; usability evaluation
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [user studies] =&gt; Array
        (
            [slug] =&gt; user studies
            [name] =&gt; user studies
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; When a system or context is investigated by using real users. {{User studies}} are often regarded as being the gold standard form of {{evaluation}}, but may not always be possible, for example, if the intended users are difficult to contact or a vulnerable group. 
            [has_links] =&gt; 
        )

    [user study] =&gt; Array
        (
            [slug] =&gt; user study
            [name] =&gt; user study
            [canonical] =&gt; user studies
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [user test] =&gt; Array
        (
            [slug] =&gt; user test
            [name] =&gt; user test
            [canonical] =&gt; user studies
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [user testing] =&gt; Array
        (
            [slug] =&gt; user testing
            [name] =&gt; user testing
            [canonical] =&gt; user studies
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [user trial] =&gt; Array
        (
            [slug] =&gt; user trial
            [name] =&gt; user trial
            [canonical] =&gt; user studies
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [user-testing] =&gt; Array
        (
            [slug] =&gt; user-testing
            [name] =&gt; user-testing
            [canonical] =&gt; user studies
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [validation] =&gt; Array
        (
            [slug] =&gt; validation
            [name] =&gt; validation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; The general process of establishing whether something (an interaction technique, system, theory, design principle, toolkit) is in some sense 'valid' &ndash; correct, true, usable. Within HCI there is often a focus on {{empirical evaluation}} of a system, theory or other artefact after it has been created. However, this is effectively impossible for {{generative artefacts}} (things that create other things) such as toolkits, methodologies or design principles, and then other methods, such as {{justification}} based on reasoning and past literature, become more important.
            [has_links] =&gt; 
        )

    [variability] =&gt; Array
        (
            [slug] =&gt; variability
            [name] =&gt; variability
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; As a statistical term, this is pretty close to its everyday meaning: how much data items vary from each other &ndash; are they all quite close, or do they vary over a wide range of values? There are various measures of {{variability}}, from the absolute range ({{maximum}}&ndash;{{minimum}}) to the {{interquartile range}}, or the ubiquitous {{standard deviation}}. For a {{theoretical distribution}} you can think of this as how widely spread the graph is, or how variable you would expect data that conforms to the distribution to be.
            [has_links] =&gt; 
        )

    [variance] =&gt; Array
        (
            [slug] =&gt; variance
            [name] =&gt; variance
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; A measure of the {{variation}} of {{empirical data}} or of a {{theoretical distribution}}. It is calculated by taking the average of the square of the difference between the data items and their {{mean}}, that is the average square of the {{residual}}. &lt;br&gt;{{variance}}  =   ( &amp;Sigma; (x&lt;sub&gt;i &lt;/sub&gt;&ndash; &amp;mu;)&lt;sup&gt;2&lt;/sup&gt; ) / N  &ndash; where x&lt;sub&gt;i &lt;/sub&gt; is the i-th data item and N is the number of items &lt;br&gt;Often the mean of {{empirical data}} is {{estimated}} based on the data in a {{sample}}; in such cases the simple formula slightly underestimates the real value of the {{variance}} and instead the {{sum of squares}} is divided by N&ndash;1. &lt;br&gt;The {{variance}} is a square measure, which is not easy to interpret (for example the variance of people's height is in square metres). So, more often the {{standard deviation}} is given, which is the square root of the variance. &lt;br&gt; The variance is also typically less {{robust}} than other measures of {{variation}} such as {{interquartile range}}. However, variances have nice mathematical properties &ndash; they 'add up' &ndash; and so are heavily used.
            [has_links] =&gt; 
        )

    [visualisation] =&gt; Array
        (
            [slug] =&gt; visualisation
            [name] =&gt; visualisation
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Methods and systems to make it easier to see trends or other patterns in data or derived results. This varies from simple {{scatter plots}} to more complex information visualisations such as the {{TreeMap}} or {{parallel coordinates}}. Where {{visualisation}} is combined with machine learning or other advanced data analysis techniques, it is referred to as {{visual analytics}}. Despite the 'visual' word in its name, {{visualisation}} is also sometimes used to refer to ways to communicate information through other senses such as auditory versions of graphs.
            [has_links] =&gt; 
        )

    [weird people] =&gt; Array
        (
            [slug] =&gt; weird people
            [name] =&gt; WEIRD people
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; Western, Educated, Industrialized, Rich, and Democratic &ndash; Henrich et al. pointed out that the vast majority of psychological studies use a very specific type of subject, not at all typical of the global population. IIndeed for many experiments the main subjects are psychology students!  In a {{meta-analysis}} where studies had also been performed in cultural settings which were not WEIRD, they found that cognitive and perceptual results that had been thought to be fundamental to human nature, for example the {{muller-lyer illusion}}, turn out to vary between cultures.
            [has_links] =&gt; 1
        )

    [within-groups studies] =&gt; Array
        (
            [slug] =&gt; within-groups studies
            [name] =&gt; within-groups studies
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This is similar to a {{within-subjects design}} except that a group of people need to work together in each condition, for example on some form of shared software or {{CSCW}} system. If anything, group dynamics make the differences between groups larger than {{individual differences}}, so there is even greater advantage in terms of {{statistical power}}. Unfortunately the interference effects between conditions are also greater and it is harder to match groups in the way that you can individually {{match subjects}}.
            [has_links] =&gt; 
        )

    [within-subjects design] =&gt; Array
        (
            [slug] =&gt; within-subjects design
            [name] =&gt; within-subjects design
            [canonical] =&gt; within-subjects experiment
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [within-subjects designs] =&gt; Array
        (
            [slug] =&gt; within-subjects designs
            [name] =&gt; within-subjects designs
            [canonical] =&gt; within-subjects experiment
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [within-subjects designs ] =&gt; Array
        (
            [slug] =&gt; within-subjects designs 
            [name] =&gt; within-subjects designs 
            [canonical] =&gt; within-subjects experiment
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [within-subjects experiment] =&gt; Array
        (
            [slug] =&gt; within-subjects experiment
            [name] =&gt; within-subjects experiment
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; When you ask the same person to take part in two or more conditions of an experiment, in contrast to {{between-subjects design}} where each person takes part in a single condition. For example, if you are comparing two user interfaces, you might ask each subject to perform tasks on both. The advantage of this is that if an individual is particularly skilled or unskilled at the tasks or has simply had a bad day, this cancels out between the two conditions &ndash; that is it cancels out {{individual differences}}. Crucially this increases the {{statistical power}} of your study by {{reducing noise}}. The downside is that there may be {{order effects}} or other forms of {{interference}} between the conditions; for example, a positive {{learning effect}} where the fact that someone has done a task with the first system makes it easier to perform a similar task on the second system. There are ways to ameliorate these effects, for example to {{randomise the order of conditions}}. If a {{within-subjects design}} is not possible and you need to use a {{between-subjects design}}, you might try {{matching subjects}}.
            [has_links] =&gt; 
        )

    [within-subjects experiments] =&gt; Array
        (
            [slug] =&gt; within-subjects experiments
            [name] =&gt; within-subjects experiments
            [canonical] =&gt; within-subjects experiment
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [within-subjects studies] =&gt; Array
        (
            [slug] =&gt; within-subjects studies
            [name] =&gt; within-subjects studies
            [canonical] =&gt; within-subjects experiment
            [see] =&gt; 
            [description] =&gt; 
            [has_links] =&gt; 
        )

    [wizard of oz prototyping] =&gt; Array
        (
            [slug] =&gt; wizard of oz prototyping
            [name] =&gt; Wizard of Oz prototyping
            [canonical] =&gt; 
            [see] =&gt; 
            [description] =&gt; This is a form of {{prototyping}} where some aspect of the interface that would be expected to be performed by a computer is instead performed by a human researcher, usually secretly. This can be used to try out ideas early before they are fully implemented, or even to explore technology that has not yet been fully developed.
            [has_links] =&gt; 
        )

)


  -->

<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
<html xmlns="http://www.w3.org/1999/xhtml">
<head>
<meta http-equiv="Content-Type" content="text/html; charset=UTF-8" />
<title>HCI Statistics</title>
<meta name="twitter:card" content="summary" />
<meta name="twitter:site" content="@HciStatistics" />
<meta name="twitter:creator" content="@alanjohndix" />
<meta property="og:url" content="https://alandix.com/glossary/hcistats" />
<meta property="og:title" content="HCI Statistics" />
<meta property="og:description" content="Terms from Statistics for HCI: Making Sense of Quantitative Data" />
<meta property="og:image" content="https://alandix.com/glossary/assets/images/cover-stats-for-HCI-square-128x128.jpg" />
<link rel="icon" href="https://alandix.com/glossary/assets/images/cover-stats-for-HCI-square-128x128.jpg" />
<script
  src="https://code.jquery.com/jquery-3.4.1.min.js"
  integrity="sha256-CSXorXvZcTkaix6Yvo6HppcZGetbYMGWSFlBw8HfCJo="
  crossorigin="anonymous"></script>
<link rel="stylesheet" href="https://code.jquery.com/ui/1.12.0/themes/smoothness/jquery-ui.css">
<link rel="stylesheet" href="https://alandix.com/glossary/assets/css/style.css">
<script src="https://code.jquery.com/ui/1.12.1/jquery-ui.js"></script>
<script src="https://alandix.com/glossary/assets/js/main.js"></script>
<script type="text/javascript" src="https://alandix.com/code/favicon/favicon.js"></script>
<script language="javascript">
$(function() {
  loadFavicons(document);
});
</script></style>
<script>
var path_info = "hcistats";
var glossaries = {"hcibook":{"slug":"hcibook","name":"HCI Book","description":"Terms related to <a href=\"https:\/\/hcibook.com\/e3\/\">Human&ndash;Computer Interaction<\/a>","list-message":"If the term you are after is not here, please also look at the <a href=\"https:\/\/hcibook.com\/e3\/search\/\">HCI Book search page<\/a>.","item-message":"","icon":"https:\/\/hcibook.com\/e3-images\/hand-sml.gif","twitter_site":"@hcibook","url":"https:\/\/alandix.com\/statistics\/book\/","options":{"json_entries":"\/var\/www\/vhosts\/alandix.com\/httpdocs\/glossary\/private\/data\/hcibook-info.json"},"menu":[{"name":"book","url":"https:\/\/hcibook.com\/e3\/"}]},"hcistats":{"slug":"hcistats","name":"HCI Statistics","description":"Terms from <a href=\"https:\/\/alandix.com\/statistics\/book\/\">Statistics for HCI: Making Sense of Quantitative Data<\/a>","list-message":"Terms in <strong>bold<\/strong> have additional details (short description, links, etc.).  <br>Those in <em>italics<\/em> are stubs at present, just pointing to relevant pages in the book.","item-message":"","icon":"cover-stats-for-HCI-square-128x128.jpg","twitter_site":"@HciStatistics","url":"https:\/\/alandix.com\/statistics\/book\/","menu":[{"name":"book","url":"https:\/\/alandix.com\/statistics\/book\/"}],"options":{"data_api":"https:\/\/datatodata.org\/HCIstats\/data.php\/"}},"aibook":{"slug":"aibook","name":"Artificial Intelligence","description":"Terms from <a href=\"https:\/\/alandix.com\/aibook\/\">Artificial Intelligence: humans at the heart of algorithms<\/a>","list-message":"Terms in <strong>bold<\/strong> have additional details (short description, links, etc.).  <br>Those in <em>italics<\/em> are stubs at present, just pointing to relevant pages in the book.","item-message":"","icon":"aibook-icon-128x128.png","url":"https:\/\/alandix.com\/aibook\/","options":{"json_entries":"\/var\/www\/vhosts\/alandix.com\/httpdocs\/glossary\/private\/data\/aibook-info.json"},"menu":[{"name":"book","url":"https:\/\/alandix.com\/aibook\/"}]}};
var glossary_info = {"slug":"hcistats","name":"HCI Statistics","description":"Terms from <a href=\"https:\/\/alandix.com\/statistics\/book\/\">Statistics for HCI: Making Sense of Quantitative Data<\/a>","list-message":"Terms in <strong>bold<\/strong> have additional details (short description, links, etc.).  <br>Those in <em>italics<\/em> are stubs at present, just pointing to relevant pages in the book.","item-message":"","icon":"cover-stats-for-HCI-square-128x128.jpg","twitter_site":"@HciStatistics","url":"https:\/\/alandix.com\/statistics\/book\/","menu":[{"name":"book","url":"https:\/\/alandix.com\/statistics\/book\/"}],"options":{"data_api":"https:\/\/datatodata.org\/HCIstats\/data.php\/"}};
var glossary = "hcistats";
var term_info = null;
var term = null;
</script>
<script>window.twttr = (function(d, s, id) {
  var js, fjs = d.getElementsByTagName(s)[0],
    t = window.twttr || {};
  if (d.getElementById(id)) return t;
  js = d.createElement(s);
  js.id = id;
  js.src = "https://platform.twitter.com/widgets.js";
  fjs.parentNode.insertBefore(js, fjs);

  t._e = [];
  t.ready = function(f) {
    t._e.push(f);
  };

  return t;
}(document, "script", "twitter-wjs"));</script></head>

<body class="glossary glossary-hcistats">
<div id="fb-root"></div>
<script async defer crossorigin="anonymous" src="https://connect.facebook.net/en_GB/sdk.js#xfbml=1&version=v6.0&appId=212157625472599"></script>
<header>
<img src="https://alandix.com/glossary/assets/images/cover-stats-for-HCI-square-128x128.jpg" class="icon"><h1>HCI Statistics</h1>
<h2><span class="description">Terms from <a href="https://alandix.com/statistics/book/">Statistics for HCI: Making Sense of Quantitative Data</a></span>
</h2></header>
<nav>
<ul class="topmenu">
<li><span class="inactive">glossary</span></li><li><a href="https://alandix.com/statistics/book/" class="active">book</a></li></ul>
<ul class="social">
<li class="twitter"><a class="twitter-share-button" href="https://twitter.com/intent/tweet?text=Glossary item">Tweet</a></li>
<li class="facebook"><div class="fb-share-button" data-href="https://alandix.com/glossary/hcistats" data-layout="button" data-size="small"><a target="_blank" href="https://www.facebook.com/sharer/sharer.php?u=https%3A%2F%2Fexample.com%2Ftest&amp;src=sdkpreparse" class="fb-xfbml-parse-ignore">Share</a></div></li>
</ul>
</nav>
<div class="content">
<p class="message">Terms in <strong>bold</strong> have additional details (short description, links, etc.).  <br>Those in <em>italics</em> are stubs at present, just pointing to relevant pages in the book.</p>
<ul class="glossary_list">
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/a-b test">A&ndash;B test</a></span><!--  Array
(
    [slug] =&gt; a-b test
    [name] =&gt; A&amp;ndash;B test
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A&amp;ndash;B test is a form of automated {{usability evaluation}}, used particularly on the web. It involves releasing two (or more) minor variants of an application, A and B, for example two different colour schemes or positions for a {{call-to-action button}}. Metrics are gathered automatically, such as {{click-through rate}} or {{time on page}}, and these are used to determine which of the variants is best.
    [has_links] =&gt; 
)
  --></li>
  <li ><span class="name">ACM CHI</span> - See <span class="has_content"><a href="https://alandix.com/glossary/hcistats/chi conference">CHI conference</a></span><!--  Array
(
    [slug] =&gt; acm chi
    [name] =&gt; ACM CHI
    [canonical] =&gt; 
    [see] =&gt; chi conference
    [description] =&gt; 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/acm conference">ACM conference</a></span><!--  Array
(
    [slug] =&gt; acm conference
    [name] =&gt; ACM conference
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; ACM is one of the major international organisations for computing. ACM conferences such as {{CHI}} are often among the most prestigious in their areas.
    [has_links] =&gt; 1
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/alternative hypothesis">alternative hypothesis</a></span><!--  Array
(
    [slug] =&gt; alternative hypothesis
    [name] =&gt; alternative hypothesis
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In {{hypothesis testing}} the alternative hypothesis, often written H&lt;sub&gt;1&lt;/sub&gt;, is usually the thing you would like to show is true. For example, that your updated interface design is better than the old one.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/alternative statistics">alternative statistics</a></span><!--  Array
(
    [slug] =&gt; alternative statistics
    [name] =&gt; alternative statistics
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; One of the responses to the {{statistical crisis}} has been to look towards different kinds of statistical methods including {{Bayesian statistics}} and {{simulation methods}}. Collectively these are sometimes called 'new statistics', 'alternative statistics', or even 'alt stats'. The term is also sometimes used to include lesser used aspects of more traditional statistics such as {{confidence intervals}} and measures of {{effect size}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/american statistical association">American Statistical Association (ASA)</a></span><!--  Array
(
    [slug] =&gt; american statistical association
    [name] =&gt; American Statistical Association (ASA)
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This is the world's largest statistical organisation and second oldest after the {{Royal Statistical Society}}.
    [has_links] =&gt; 1
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/anova">ANOVA (Analysis of Variance)</a></span><!--  Array
(
    [slug] =&gt; anova
    [name] =&gt; ANOVA (Analysis of Variance)
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This is a form of statistical analysis used for {{multivariate data}} that is {{Normally distributed}}. It is quite robust, in the sense that even if the data is not completely Normal, the results are fairly reliable. You might use this, for example, if you have performed an experiment measuring response times with several users, each performing several tasks with two alternative interactive systems; in this case you have three variables: user x task x system. You are probably interested in which system is better overall, but the different tasks may vary in difficulty and the users may differ in expertise. The ANOVA can disentangle these effects, giving you {{main effects}}: averages for each user, task and system; as well as {{interaction effects}} such as whether some tasks are better suited to one system or the other.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/applied mathematics">applied mathematics</a></span><!--  Array
(
    [slug] =&gt; applied mathematics
    [name] =&gt; applied mathematics
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The branch of mathematics that deals with real world problems.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/approximate distribution">approximate distribution</a></span><!--  Array
(
    [slug] =&gt; approximate distribution
    [name] =&gt; approximate distribution
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A {{distribution}} that is close enough to another distribution to be of use in calculations. The distribution being used for the approximation is usually in some ways more mathematically tractable or easier to deal with. For example, the {{Normal distribution}} is often used as an approximation for other distributions, even {{discrete distributions}} such as the {{Binomial distribution}}. 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/approximately normal">approximately Normal</a></span><!--  Array
(
    [slug] =&gt; approximately normal
    [name] =&gt; approximately Normal
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A distribution is &lt;strong&gt;approximately Normal&lt;/strong&gt; when the {{Normal distribution}} can be used as an {{approximate distribution}}. This is common when the number of samples or parts making up a distribution grows; for example, if you have 100 coin tosses the resulting {{Binomial distribution}} is, for most purposes, approximately Normal. This is often helpful as many statistical tests rely on the data being {{Normally distributed}}. The {{Central Limit Theorem}} explains why it is so common to see approximately Normal distributions.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/arithmetical data">arithmetical data</a></span><!--  Array
(
    [slug] =&gt; arithmetical data
    [name] =&gt; arithmetical data
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A form of {{discrete data}} where the values can be meaningfully added or averaged. Contrast with {{categorical data}} and {{ordinal data}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/asymmetric distribution">asymmetric distribution</a></span><!--  Array
(
    [slug] =&gt; asymmetric distribution
    [name] =&gt; asymmetric distribution
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Where a distribution curve is not symmetrical: an example would be a distribution that rises quite rapidly from zero, but then tails off more slowly.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/average">average</a></span><!--  Array
(
    [slug] =&gt; average
    [name] =&gt; average
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The term {{average}} most commonly refers to the arithmetic {{mean}}, but can also refer to the {{median}}, especially for {{asymmetric distributions}} where the {{median}} is typically a little smaller than the {{mean}}, which can be influenced by a few large {{outliers}}. For example, 'average income' usually refers to the median, not the {{mean}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/balance orders">balance orders</a></span><!--  Array
(
    [slug] =&gt; balance orders
    [name] =&gt; balance orders
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A way of mitigating {{order effects}} by, for example, having half the subjects in an experiment perform a task in the order A&ndash;B and half in the order B&ndash;A.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/base period">base period</a></span><!--  Array
(
    [slug] =&gt; base period
    [name] =&gt; base period
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; For temporal data, the starting point in time, also known as base year. This is a special case of a {{basepoint}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/base rate">base rate</a></span><!--  Array
(
    [slug] =&gt; base rate
    [name] =&gt; base rate
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The base rate is the underlying level of prevalence of some attribute of a population. For example, the base rate of a disease is the proportion of people who normally have the illness. Where such a base rate is known it can be used as {{prior probability}} in {{Bayesian methods}}. The base rate is important in assessing how likely the underlying attribute is for a particular individual. If you ignore the base rate you may make a poor assessment of the situation, sometimes called the base rate fallacy or base rate neglect. For example, imagine a test for a rare form of cancer that is 100% accurate in detecting if you the cancer and 99% accurate when you don't. You find you have tested positive. At first this seems very worrying; however, if the base rate is that only 1 in a million people have the cancer but the other 999,999 don't,  then for every person who has the cancer and tests positive there are 1% of 999,999, that is around 10,000, who test positive and are in the clear. It may be worth having further tests, but you are no more likely to have the disease than to be killed on the road that year.
    [has_links] =&gt; 1
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/baseline">baseline</a></span><!--  Array
(
    [slug] =&gt; baseline
    [name] =&gt; baseline
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The choice of where to place the bottom of a graph (y-axis): at zero, or at some other value, a {{false baseline}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/basepoint">basepoint</a></span><!--  Array
(
    [slug] =&gt; basepoint
    [name] =&gt; basepoint
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The choice of where to place the left-to-right start point of a graph (x-axis): at zero, or at some other value, a {{false basepoint}}.

    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/bayes factor">Bayes factor</a></span><!--  Array
(
    [slug] =&gt; bayes factor
    [name] =&gt; Bayes factor
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The &lt;strong&gt;Bayes factor&lt;/strong&gt; is used in {{Bayesian statistics}} to measure the relative probability of alternatives based on some form of observed data. For a simple H&lt;sub&gt;0&lt;/sub&gt; vs H&lt;sub&gt;1&lt;/sub&gt; alternative (e.g. human vs Martian), the Bayes factor is the ratio of the odds for the two alternatives: P(H&lt;sub&gt;1&lt;/sub&gt; | observed data) / P(H&lt;sub&gt;0&lt;/sub&gt; | observed data). However, if H&lt;sub&gt;0&lt;/sub&gt; vs H&lt;sub&gt;1&lt;/sub&gt; is more complex (e.g. mean = 0 vs any other value), then calculating the Bayes factor involves averaging over the alternative possibilities weighted by the {{prior probability distribution}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/bayes rule">Bayes rule</a></span><!--  Array
(
    [slug] =&gt; bayes rule
    [name] =&gt; Bayes rule
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Bayes rule is a mathematical equation that allows you to calculate {{conditional probabilities}} when you know conditional probabilities the 'other way round'. For example, if you know the probability of a patient exhibiting various symptoms if they have a particular illness, it allows you to calculate the probability that they have the illness given the symptoms they are showing. In the case of two mutually exclusive states A and B (e.g. A=&quot;has a cold&quot;, B=&quot;does not have a cold&quot;), and an observation C (e.g. &quot;has a runny nose&quot;), then Bayes rule is P(A|C) = P(C|A)P(A) / ( P(C|A)P(A) + P(C|A)P(B) ). Note that for this you need to know the {{prior probability}} of each occurrence, P(A) and P(B); when these are known from data they are also known as the {{base rate}}.
    [has_links] =&gt; 
)
  --></li>
  <li ><span class="name">Bayesian inference</span> - See <span class="has_content"><a href="https://alandix.com/glossary/hcistats/bayesian reasoning">Bayesian reasoning</a></span><!--  Array
(
    [slug] =&gt; bayesian inference
    [name] =&gt; Bayesian inference
    [canonical] =&gt; 
    [see] =&gt; bayesian reasoning
    [description] =&gt; 
    [has_links] =&gt; 
)
  --></li>
  <li ><span class="name">Bayesian methods</span> - See <span class="has_content"><a href="https://alandix.com/glossary/hcistats/bayesian reasoning">Bayesian reasoning</a></span><!--  Array
(
    [slug] =&gt; bayesian methods
    [name] =&gt; Bayesian methods
    [canonical] =&gt; 
    [see] =&gt; bayesian reasoning
    [description] =&gt; 
    [has_links] =&gt; 
)
  --></li>
  <li ><span class="name">Bayesian posterior probability</span> - See <span class="has_content"><a href="https://alandix.com/glossary/hcistats/posterior distribution">posterior distribution</a></span><!--  Array
(
    [slug] =&gt; bayesian posterior
    [name] =&gt; Bayesian posterior probability
    [canonical] =&gt; 
    [see] =&gt; posterior distribution
    [description] =&gt; 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/bayesian reasoning">Bayesian reasoning</a></span><!--  Array
(
    [slug] =&gt; bayesian reasoning
    [name] =&gt; Bayesian reasoning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Bayesian reasoning is the use of {{Bayes rule}} to allow inferences to be made about the probability of things based on evidence. It relies crucially on already knowing the {{prior probability}} of different alternatives before the evidence is available. Bayesian reasoning can be used for internal algorithms, for example to make adaptive or {{intelligent interfaces}}. It can also be used in a form of statistical reasoning, {{Bayesian statistics}}, which can be used as an alternative to traditional {{hypothesis testing}} or other statistical analyses.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/bayesian statistics">Bayesian statistics</a></span><!--  Array
(
    [slug] =&gt; bayesian statistics
    [name] =&gt; Bayesian statistics
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Bayesian statistics uses {{Bayesian reasoning}} as an alternative to {{traditional statistics}}, such as {{significance testing}} or {{confidence intervals}}. Crucially this demands that you make precise your {{prior belief}} about the outcomes {{encoded as a probability}}. In some cases your belief will be based on existing data or evidence; in these cases {{Bayes rule}} gives a strong mathematical foundation, yielding a precise {{posterior probability}} given the additional evidence. More commonly the prior is a guesstimate or subjective assessment and the {{posterior probability}} is not really a probability, but rather evidence-based belief. 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/bayesian test">Bayesian test</a></span><!--  Array
(
    [slug] =&gt; bayesian test
    [name] =&gt; Bayesian test
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A type of statistical technique using Bayesian methods to address similar problems to traditional statistical {{significance testing}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/between-subjects design">between-subjects design</a></span><!--  Array
(
    [slug] =&gt; between-subjects design
    [name] =&gt; between-subjects design
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; An experiment design where there are many subjects, but each only performs in a single condition. Contrast with a {{within-subjects design}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/bias">bias</a></span><!--  Array
(
    [slug] =&gt; bias
    [name] =&gt; bias
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
Bias is about systematic effects that skew your results in one way or another.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/big data">big data</a></span><!--  Array
(
    [slug] =&gt; big data
    [name] =&gt; big data
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Vast amounts of data that are now available and allow us to study phenomena in a way that would have been impossible only a few years ago.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/bimodal distribution">bimodal distribution</a></span><!--  Array
(
    [slug] =&gt; bimodal distribution
    [name] =&gt; bimodal distribution
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A bimodal distribution has two main peaks or {{modes}}. This often happens when there is a {{mixed population}} consisting of two distinct groups with different characteristics. Educators may be familiar with this in courses with more formal or mathematical materials; there is often a group that 'gets it' and has marks clustered towards the high grades, while the rest have marks clustered towards the lower end of the range.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/binary data">binary data</a></span><!--  Array
(
    [slug] =&gt; binary data
    [name] =&gt; binary data
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A special case of discrete data with only two states, such as yes/no answers, or present/not present.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/binomial distribution">Binomial distribution</a></span><!--  Array
(
    [slug] =&gt; binomial distribution
    [name] =&gt; Binomial distribution
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A distribution resulting from the sum of independent two-outcome events such as coin tosses.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/bonferroni correction">Bonferroni correction</a></span><!--  Array
(
    [slug] =&gt; bonferroni correction
    [name] =&gt; Bonferroni correction
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A method of dividing the target {{p-level}} by the number of tests, used when there have been multiple tests to reduce the chance of thinking that random effects are real. For example, if you run 10 tests there is an approximately 50:50 chance of seeing at least one come out at 5% significance level by chance. In order to have a real 5% level overall, you therefore test each one at 0.5% (5%/10). It is a form of {{multiplicity adjustment}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/borrowed methods">borrowed methods</a></span><!--  Array
(
    [slug] =&gt; borrowed methods
    [name] =&gt; borrowed methods
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Methods derived from  other disciplines, for example psychology, where controlled experiments typically have a single cause and are in highly controlled environments. Care must be taken, because interaction studies, even in lab settings, almost always have multiple causes and open situations.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/bounded below">bounded below</a></span><!--  Array
(
    [slug] =&gt; bounded below
    [name] =&gt; bounded below
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Data which is finite at the low end of the range but may be unbounded at the high end, e.g. when a coin is tossed, the number of heads until the first tail.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/bounded data">bounded data</a></span><!--  Array
(
    [slug] =&gt; bounded data
    [name] =&gt; bounded data
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Data which has a limited range of responses with either a fixed maximum or minimum. This may be {{discrete data}}: for example, the number of heads in 6 coin tosses can only be between 0 and 6. You cannot have -3 heads or 7 heads ({{bounded above}} and {{bounded below}}). Similarly, for {{continuous data}} the user {{response time}} cannot be negative ({{bounded below}}).
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/bounds">bounds</a></span><!--  Array
(
    [slug] =&gt; bounds
    [name] =&gt; bounds
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; These are the maximum {{upper bound}} and minimum {{lower bound}} values that it is possible for a value to take. For example the bounds for the roll of a die are 1 (lower bound) and 6 (upper bound). Not all variables have bounds: for example, when {{coin tossing}} the number of tails until you get a head is {{bounded below}} by 0 (you can't have a negative number of tails), but {{unbounded above}}, so it has a lower bound, but no upper bound.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/categorical data">categorical data</a></span><!--  Array
(
    [slug] =&gt; categorical data
    [name] =&gt; categorical data
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Data which has {{discrete values}} that cannot easily be compared, added, or averaged. Also known as nominal data. Contrast with {{arithmetical data}} and {{ordinal data}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/cauchy distribution">Cauchy distribution</a></span><!--  Array
(
    [slug] =&gt; cauchy distribution
    [name] =&gt; Cauchy distribution
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A bell-shaped distribution often used for unbounded value ranges in {{Bayesian statistics}}. Similar to a {{Normal distribution}} but with fatter edges (tails). It is a {{symmetric distribution}} with {{infinite variance}}. The formula for the simple case with {{mean}} 0 is 1/&amp;pi;(1+x&lt;sup&gt;2&lt;/sup&gt;).
    [has_links] =&gt; 1
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/causality">causality</a></span><!--  Array
(
    [slug] =&gt; causality
    [name] =&gt; causality
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; From data we may observe that two phenomena are related, even that one always precedes the other. However, it can sometimes be hard to establish causality - which is the cause of the other, the reason it happens or has the value it does. Sometimes two phenomena may be linked by a common cause rather than one being the cause of the other. For example, from data we might note that people who do a lot of exercise are also more healthy, and so think that exercise causes good health, but equally well it might be that people who are in good health are more likely to take exercise, or that there is some aspect of people's upbringing or genes that leads to both. Understanding {{mechanism}} can sometimes help establish causality.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/central limit theorem">Central Limit Theorem</a></span><!--  Array
(
    [slug] =&gt; central limit theorem
    [name] =&gt; Central Limit Theorem
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A mathematical result, which says that the average ({{arithmetic mean}}) of lots of items or the sum of lots of very small things behaves more and more like a {{Normal distribution}} as the number of small items gets larger.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/cherry picking">cherry picking</a></span><!--  Array
(
    [slug] =&gt; cherry picking
    [name] =&gt; cherry picking
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A common problem in statistics, when researchers deliberately or accidentally ignore results that are not to their liking and only report those that are advantageous. Cherry picked results may mislead people (including the researchers themselves) or make them falsely believe something is true.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/chi conference">CHI conference</a></span><!--  Array
(
    [slug] =&gt; chi conference
    [name] =&gt; CHI conference
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Annual ACM Conference on Human Factors in Computing Systems
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/citation count">citation count</a></span><!--  Array
(
    [slug] =&gt; citation count
    [name] =&gt; citation count
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The number of times a paper has been cited by others.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/citation data">citation data</a></span><!--  Array
(
    [slug] =&gt; citation data
    [name] =&gt; citation data
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This is the data arising from counting or recording the way academic papers reference each other. It is a form of {{network data}} where papers are linked in a directional way as one paper cites another. A simple metric of this is the {{citation count}}, the number of times a paper has been cited by others. However, it is important to note that, in common with other forms of network data, the citation count usually follows a {{power law}} with a {{long tail}} and is not suitable for tests and analysis that rely on the data being {{approximately Normal}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/claims analysis">claims analysis</a></span><!--  Array
(
    [slug] =&gt; claims analysis
    [name] =&gt; claims analysis
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A method developed by Carroll and Rosson, which seeks to better integrate activities directed at understanding with those directed at design. DOI: &lt;doi&gt;10.1145/146802.146834&lt;/doi&gt;
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/closed formula">closed formula</a></span><!--  Array
(
    [slug] =&gt; closed formula
    [name] =&gt; closed formula
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A tractable mathematical description.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/closed world">closed world</a></span><!--  Array
(
    [slug] =&gt; closed world
    [name] =&gt; closed world
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A world where we assume we know everything and there are no unknown external factors. {{Bayesian statistics}} effectively uses a closed world assumption as it requires all unknown parameters to be assigned a {{prior probability}}. In contrast, {{traditional statistics}} adopts an {{open world}} model where some parameters are treated as completely unknown.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/cognitive load">cognitive load</a></span><!--  Array
(
    [slug] =&gt; cognitive load
    [name] =&gt; cognitive load
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The amount of mental resources (e.g. memory, attention) needed for a task or activity.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/coin tossing">coin tossing</a></span><!--  Array
(
    [slug] =&gt; coin tossing
    [name] =&gt; coin tossing
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Simple coin tossing experiments can help you to understand random phenomena and statistical procedures. When doing statistics for real, you are having to disentangle some underlying effect (say the ease of use of an interface) from effectively random or interfering effects (such as the different levels of skill of your participants). This is difficult. With coin tossing you know the coins are unbiased, 50:50 heads:tails, and hence any odd effects you see are purely due to randomness. The number of heads or tails when you toss a fixed number of coins follow the {{Binomial distribution}}.
    [has_links] =&gt; 1
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/computer simulation">computer simulation</a></span><!--  Array
(
    [slug] =&gt; computer simulation
    [name] =&gt; computer simulation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; When you emulate some real world phenomenon in a computer. As used in statistics this will usually involve random values with repeated runs of the simulation in order to obtain average values or distributions of values. This can be used in {{simulation methods}} to calculate the {{likelihood}} of particular measurements even when there is no simple mathematical formula.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/conditional probability">conditional probability</a></span><!--  Array
(
    [slug] =&gt; conditional probability
    [name] =&gt; conditional probability
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This is the probability of one event (say A) &lt;emph&gt;given&lt;emp&gt; knowledge of another event (B). It is usually written as P(A|B), read as &quot;the probability of A given B&quot;. For example, while the probability that it will rain on a typical day might be 50%, on a cloudy day this might be 80%; that is P(rain|cloudy) = 0.8.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/confidence interval">confidence interval</a></span><!--  Array
(
    [slug] =&gt; confidence interval
    [name] =&gt; confidence interval
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Confidence intervals give a lower and upper bound for the true value of an effect. They allow you to be able to answer questions such as &quot;What is the maximum likely value of this?&quot;, or &quot;How close to no difference is there between these two conditions?&quot;. They use the same information and mathematics as are used to generate the {{p-values}} in a {{significance test}}. You can choose to set higher or lower confidence levels, for example the 95% confidence interval for a value might be [0.3,1.1] and the 99% confidence interval [-0.1,1.5]. In general the higher the level of confidence you ask for, the wider the range. Like all statistics these bounds are {{uncertain}}. However, if you do lots of studies and calculate 95% confidence intervals for them all, then over time the proportion of true values that lie within the 95% confidence intervals will be at least 95% ... although you can't tell which they are!
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/confirmation bias">confirmation bias</a></span><!--  Array
(
    [slug] =&gt; confirmation bias
    [name] =&gt; confirmation bias
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Confirmation bias is when we unintentionally only seek out evidence or take notice of results that reinforce our existing views. With {{traditional statistics}} this might lead us not to publish results that we think are wrong, perhaps even removing data points that we feel must have been errors of some kind. In {{Bayesian statistics}} confirmation bias can cause even more problems, as the choice of {{prior probability}} is effectively one's own beliefs {{encoded as a probability}} and hence the output of the process, the {{posterior probability}}, may simply reflect one's existing prejudices.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/confounding variables">confounding variables</a></span><!--  Array
(
    [slug] =&gt; confounding variables
    [name] =&gt; confounding variables
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; When there is some form of relationship or correlation between the various factors of which you are measuring aspects. For example, if you are comparing the speed of use of two systems, hoping that your new system is faster, a confounding variable might be user expertise: if the users of one system are more expert than those of the other, this may create a difference in {{response time}} that is not due to the system. One way to deal with such a situation would be to create a {{balanced design}}, where the expertise of the users of each system is the same. When this is not possible, one can attempt to {{model}} the effect of expertise and correct for it.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/conscious thinking">conscious thinking</a></span><!--  Array
(
    [slug] =&gt; conscious thinking
    [name] =&gt; conscious thinking
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; These are the thoughts that you are explicitly aware of having: what you would answer if asked &quot;What are you thinking of now?&quot;. The conscious mind is extremely powerful, allowing us to think rationally and logically, but it is also limited. In particular, your conscious system of thought likes to have a single model of the world and is really bad at probability. Contrast with {{subconscious reactions}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/constant area visualisation">constant area visualisation</a></span><!--  Array
(
    [slug] =&gt; constant area visualisation
    [name] =&gt; constant area visualisation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A visualisation for hierarchical data in which the area of each part reflects the number or size of the items it represents (e.g. a {{TreeMap}} or {{PieTree}})
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/context effects">context effects</a></span><!--  Array
(
    [slug] =&gt; context effects
    [name] =&gt; context effects
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Elements of the context in which an experiment is conducted that may affect the subjects' responses and therefore influence the results of the study, for example the mood of the researcher or even the weather. 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/continuous data">continuous data</a></span><!--  Array
(
    [slug] =&gt; continuous data
    [name] =&gt; continuous data
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A set of numerical data where the values can be anywhere in a range, including fractional values, for example task completion time, or height. Contrast with {{discrete data}} which can only take on specific values. 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/continuous distribution">continuous distribution</a></span><!--  Array
(
    [slug] =&gt; continuous distribution
    [name] =&gt; continuous distribution
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A probability distribution resulting from a set of {{continuous data}}, such as time or height values. The {{Normal distribution}} is an example of a continuous distribution. Because the continuous data can take on any of the infinite number of fractional values, the probability of any precise value is effectively zero. Instead, the probability of a range of values is calculated as the area under the curve of the distribution function over the range. 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/controlled environments">controlled environments</a></span><!--  Array
(
    [slug] =&gt; controlled environments
    [name] =&gt; controlled environments
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In physics and other sciences, scientists use very pure substances with very precise instruments to reduce the variability in a study. In HCI we can sometimes emulate this, controlling conditions to reduce the variability in our study ({{reduce noise}}). For example, in a user test we might ask people to use a mobile phone to perform the same tasks in the same location with similar levels of sound and distractions. The downside of this is that it can sometimes reduce {{ecological validity}}, making the resulting conditions less like the real world.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/controlled experiment">controlled experiment</a></span><!--  Array
(
    [slug] =&gt; controlled experiment
    [name] =&gt; controlled experiment
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Controlled experiments follow a strict protocol, a predefined number of trials, often in a laboratory setting. In HCI, controlled experiments may influence results. For example, experts may behave more like novices in the artificial  environment of a lab experiment, being more careful about their actions and monitoring the results. Typically, controlled experiments have to sacrifice {{ecological validity}}, being less like the equivalent real-world activity. Contrast with {{ethnographic methods}} and {{in-the-wild}} studies. 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/controlled tasks">controlled tasks</a></span><!--  Array
(
    [slug] =&gt; controlled tasks
    [name] =&gt; controlled tasks
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Restricting tasks to enable more valid study of individual variables: for example, participants are in a real situation, but instead of doing what they like, they are given a scripted task to perform. While some realism is lost in terms of the chosen task, this increases the level of {{ecological validity}} in the environment.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/coping mechanisms">coping mechanisms</a></span><!--  Array
(
    [slug] =&gt; coping mechanisms
    [name] =&gt; coping mechanisms
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The resources that people draw on when they find ways to manage a difficult task, whether consciously or unconsciously. While this is useful for getting a job done, for the designer it can mean that usability problems may be hidden. {{Distractor tasks}} are one way to deal with this.
    [has_links] =&gt; 
)
  --></li>
  <li ><span class="name">correlation coefficient</span> - See <span class="has_content"><a href="https://alandix.com/glossary/hcistats/pearson correlation coefficient">Pearson correlation coefficient</a></span><!--  Array
(
    [slug] =&gt; correlation coefficient
    [name] =&gt; correlation coefficient
    [canonical] =&gt; 
    [see] =&gt; pearson correlation coefficient
    [description] =&gt; 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/count data">count data</a></span><!--  Array
(
    [slug] =&gt; count data
    [name] =&gt; count data
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Data obtained by simply counting something, such as the number of people, errors, or clicks of a button. This is often found in data collected by surveys, such as opinion polls and questionnaires, or in discrete data in a user study, such as counting kinds of errors.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/counterfactual">counterfactual</a></span><!--  Array
(
    [slug] =&gt; counterfactual
    [name] =&gt; counterfactual
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A form of reasoning where one has to work with states of the world about which one is agnostic or which one even believes to be false.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/covariance">covariance</a></span><!--  Array
(
    [slug] =&gt; covariance
    [name] =&gt; covariance
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The covariance is the average product of two variables after their means have been subtracted: cov(x,y) = ( &amp;Sigma; (x&lt;sub&gt;i&lt;/sub&gt; - &amp;mu;&lt;sub&gt;x&lt;/sub&gt;) &amp;times (y&lt;sub&gt;i&lt;/sub&gt; - &amp;mu;&lt;sub&gt;y&lt;/sub&gt;) ) / N. It is a generalisation of the {{variance}} of a single variable. It measures how large is the common variability of the two variables; if they are unrelated it is near zero. For example, if you measured the heights and weights of people you would expect the covariance to be positive. The size of the covariance is also influenced by the level of variability of the two individual variables: for example, the covariance between length and weight of sparrows would be a lot smaller than the same calculation for whales. The {{correlation coefficient}} corrects for this by dividing by the product of the {{standard deviation}}&nbsp;of the two variables.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/critical values">critical values</a></span><!--  Array
(
    [slug] =&gt; critical values
    [name] =&gt; critical values
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In the days before computers, when all calculations were performed by hand, and even when computers were less powerful than they are today, you would check the {{significance level}} of a result by looking up the value in a table that listed the critical value of a given statistic for 5%, 1%, etc. For example, the critical value for the {{Student's t distribution}} with 10 {{degrees of freedom}} at a 5% significance level is 2.23. Suppose we performed a t test comparing the heights of 6 people from Cardiff and 6 from London (df = 6+6-2 = 10): if the t value were greater than this critical value of 2.23 we would conclude that London and Cardiff heights were different at the 5% significance level.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/data analyst">data analyst</a></span><!--  Array
(
    [slug] =&gt; data analyst
    [name] =&gt; data analyst
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The person who is performing the analysis of data, typically a professional such as a researcher or data scientist. The outcomes of the analysis may often be summarised or visualised to make them more comprehensible for the {{data consumer}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/data consumer">data consumer</a></span><!--  Array
(
    [slug] =&gt; data consumer
    [name] =&gt; data consumer
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The person using some form of data collected, or visualised: for example, a client, management, the general public, or an academic reading an illustrated article.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/data journals">data journals</a></span><!--  Array
(
    [slug] =&gt; data journals
    [name] =&gt; data journals
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; These are academic journals that primarily publish documented data. This is in contrast to traditional journals, which may publish data, but only as some form of addendum to a paper that is primarily about the analysis and interpretation of the data.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/day effects">day effects</a></span><!--  Array
(
    [slug] =&gt; day effects
    [name] =&gt; day effects
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A particular kind of {{context effect}} when you do the same experiment on two different days and get different effects. This may be due, for example, to the weather or the mood of the researcher on a particular day.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/default bayesian factors">default Bayesian factors</a></span><!--  Array
(
    [slug] =&gt; default bayesian factors
    [name] =&gt; default Bayesian factors
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Bayesian statistics requires some form of prior. Sometimes this is based on past data or is in the form of a researcher's beliefs {{encoded as a probability}}, but where there are no clear grounds for either of these, default Bayesian factors may be used. Typically this will be a {{uniform distribution}} over {{bounded values}}: for example, assigning a prior probability of 1/6 for each of the potential faces of a die. In the case of {{unbounded data}} the {{Cauchy distribution}} is a frequent default choice although this also requires a central value and spread parameter.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/degrees of freedom">degrees of freedom</a></span><!--  Array
(
    [slug] =&gt; degrees of freedom
    [name] =&gt; degrees of freedom
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; There are various quite complex (and mathematically precise) ways to think about degrees of freedom. However, for most purposes when analysing data it means the number of independent data items minus the number of values you have {{fitted}} to the data. For example, if you have 100 (x,y) data items and then perform a simple {{linear regression}} the software will fit a slope and intercept of the {{regression line}} &ndash; that is you have fitted two numbers, hence the remaining {{degrees of freedom}} is 100&ndash;2 = 98. &lt;br&gt;Informally (but very close to the vector space mathematics) you can think of this as having 100 pieces of randomness to start with; two of these are used up in the two fitted values, so you have 98 left.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/demonic intervention">demonic intervention</a></span><!--  Array
(
    [slug] =&gt; demonic intervention
    [name] =&gt; demonic intervention
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Creating extreme experimental situations to {{increase effect}} size.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/dependent variable">dependent variable</a></span><!--  Array
(
    [slug] =&gt; dependent variable
    [name] =&gt; dependent variable
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The thing that is being measured and out of the direct control of the experimenter, for example response time or error rate. Usually shown on the vertical (y) axis. Contrast with {{independent variable}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/design space">design space</a></span><!--  Array
(
    [slug] =&gt; design space
    [name] =&gt; design space
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The range of alternative design choices that can be combined in different ways, for example synchronous vs asynchronous, domain specific vs generic. The design space may be represented in a matrix of different semi-independent  choices, but also may be more complex.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/dichotomous readings of hypothesis testing">dichotomous readings of hypothesis testing</a></span><!--  Array
(
    [slug] =&gt; dichotomous readings of hypothesis testing
    [name] =&gt; dichotomous readings of hypothesis testing
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The simplistic binary distinction of {{significant}} vs {{non-significant}}, which stems from the days when computation was performed by hand. No longer necessary with automated computation since specific {{p-values}} are readily available.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/discrete data">discrete data</a></span><!--  Array
(
    [slug] =&gt; discrete data
    [name] =&gt; discrete data
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A set of numerical data where the values are at specific points rather than anywhere along a range, for example errors: a single user may have 12 or 13 errors but cannot have 12.5. Contrast with {{continuous data}}. 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/distractor task">distractor task</a></span><!--  Array
(
    [slug] =&gt; distractor task
    [name] =&gt; distractor task
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; An additional task given during an experimental situation, which aims to saturate some aspect of the user&rsquo;s cognitive abilities, so that differences in the {{cognitive load}} of the systems or conditions being studied become apparent; for example asking a subject to count backward as they perform the {{primary task}}.
    [has_links] =&gt; 
)
  --></li>
  <li ><span class="name">distribution</span> - See <span class="has_content"><a href="https://alandix.com/glossary/hcistats/probability distribution">probability distribution</a></span><!--  Array
(
    [slug] =&gt; distribution
    [name] =&gt; distribution
    [canonical] =&gt; 
    [see] =&gt; probability distribution
    [description] =&gt; 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/dna evidence">DNA evidence</a></span><!--  Array
(
    [slug] =&gt; dna evidence
    [name] =&gt; DNA evidence
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Forensic method to discover whether a person is connected to a crime. Each person&rsquo;s DNA is almost unique; however, DNA testing is imperfect and has the possibility of error.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/ecological validity">ecological validity</a></span><!--  Array
(
    [slug] =&gt; ecological validity
    [name] =&gt; ecological validity
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The ecological validity of a test, study or experiment is the extent to which it is like the real world. Performing a {{controlled experiment}} in a laboratory would usually have low ecological validity, whereas an {{ethnographic study}}, observing people in their own workplace, would have high ecological validity.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/effect size">effect size</a></span><!--  Array
(
    [slug] =&gt; effect size
    [name] =&gt; effect size
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Effect size is how large the effect is, as opposed to how likely it is to be valid or reproducible. For example, if we had a huge data set we might be able to assess that a really tiny effect is {{statistically significant}}. It is important to give some indication of the effect size so that you can assess the importance of the result. Note that some measures called 'effect size' are actually about the ratio between the raw size of the effect and other factors. Effect size is one of the three elements in the {{noise-effect-number triangle}} that can be manipulated to increase the {{statistical power}} of a study or experiment. 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/effective sample size">effective sample size</a></span><!--  Array
(
    [slug] =&gt; effective sample size
    [name] =&gt; effective sample size
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; If the elements of a sample are not independent, then the raw sample size may not be the best value to use when thinking about the data. The effective size of a sample when these factors are taken into account may be smaller than the raw sample size. For example, if we asked 100 random people attending a football match which team they supported, we could use this to estimate the popularity of the teams based on a sample size of 100. However, if we had instead obtained the sample by taking 50 people each from two coaches, then the effective sample size would probably be 2 rather than 100, as it is likely that everyone on a coach supports the same team.   
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/elderly users">elderly users</a></span><!--  Array
(
    [slug] =&gt; elderly users
    [name] =&gt; elderly users
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Older users often have different characteristics from younger ones: for example, on average poorer eyesight or problems with short-term memory. However, this said, there is also a wide variation between different users of any age group and, of course, similarities between ages; for example, many younger people have poor eyesight.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/emergent patterns">emergent patterns</a></span><!--  Array
(
    [slug] =&gt; emergent patterns
    [name] =&gt; emergent patterns
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Patterns that become apparent during a study and may warrant further investigation.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/empirical data">empirical data</a></span><!--  Array
(
    [slug] =&gt; empirical data
    [name] =&gt; empirical data
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Data obtained by some form of experiment or measurement of the real world.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/empirical distribution">empirical distribution</a></span><!--  Array
(
    [slug] =&gt; empirical distribution
    [name] =&gt; empirical distribution
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A distribution derived from actual data, for example showing the number of people in each of several categories. Contrast with a {{theoretical distribution}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/empirical evaluation">empirical evaluation</a></span><!--  Array
(
    [slug] =&gt; empirical evaluation
    [name] =&gt; empirical evaluation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This is when a system or aspect of a system is evaluated through actual data, whether gathered in a {{laboratory experiment}} or {{in the wild}}. Alternatives would be to use some form of {{expert evaluation}} or {{theoretical analysis}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/empirical methods">empirical methods</a></span><!--  Array
(
    [slug] =&gt; empirical methods
    [name] =&gt; empirical methods
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A way of working out statistical properties, primarily used when there is no standard test. Some form of computer simulation is run, based on mathematical probabilities or observed data. By running many simulations, we can observe the likelihood of the results we have observed. 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/encoding as probability">encoding (belief) as probability</a></span><!--  Array
(
    [slug] =&gt; encoding as probability
    [name] =&gt; encoding (belief) as probability
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Bayesian statistics requires a precise {{prior probability}}. Often this is obtained by quantifying a prior belief into an estimated number, which is then treated as if it is a real probability.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/epistemologically random">epistemologically random</a></span><!--  Array
(
    [slug] =&gt; epistemologically random
    [name] =&gt; epistemologically random
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Random in one's state of knowledge. For example, if you have tossed a coin, but not yet looked at it, you would still treat it as a 50:50 odds if asked to bet, but actually it is already in one state or the other, you just don't know which. Contrast with {{ontologically random}}. 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/error bars">error bars</a></span><!--  Array
(
    [slug] =&gt; error bars
    [name] =&gt; error bars
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; An indicator used on a graph or histogram to show the the variability of the average (the {{standard error}} of the mean). This gives some idea of the accuracy of the average measure. It is usually in the form of small 'T' shaped bars extending above and below the central point on the graph.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/error rate">error rate</a></span><!--  Array
(
    [slug] =&gt; error rate
    [name] =&gt; error rate
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The error rate for a task is the frequency or number of errors that a user encounters whilst performing the task. This is one of the metrics used to evaluate the usability of a system.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/estimate">estimate</a></span><!--  Array
(
    [slug] =&gt; estimate
    [name] =&gt; estimate
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In statistics, an estimate can have its ordinary day-to-day meaning: a value which is a best guess for some parameter or variable. However, it can also have a more precise meaning: an estimate calculated via some mathematical or computational method. For example, if we are interested in the average height of people in Wales, we might choose 100 people in the country at random and then work out the average of the sample. This sample average could then be used as an estimate of the overall average height of people in Wales.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/estimate of the variability">estimate of the variability</a></span><!--  Array
(
    [slug] =&gt; estimate of the variability
    [name] =&gt; estimate of the variability
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Random phenomena have {{variability}}. In some cases we know what this is based on existing literature, previous studies or theoretical calculations, but at other times we need to estimate the size of the variability, usually from data gathered in an experiment or study. For example, we might calculate the {{interquartile range}} of our measured data and use that as an estimate of the variability of the underlying phenomenon.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/estimated standard deviation">estimated standard deviation</a></span><!--  Array
(
    [slug] =&gt; estimated standard deviation
    [name] =&gt; estimated standard deviation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A special case of an {{estimate of the variability}} of data is the {{estimated standard deviation}}. For example, if we know that data follows the {{Normal distribution}} (say people's heights), we might work out the {{standard deviation of the population}} for a sample, and then use this as an estimate of the true {{standard deviation}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/ethnography">ethnography</a></span><!--  Array
(
    [slug] =&gt; ethnography
    [name] =&gt; ethnography
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A form of {{qualitative research}} based strongly on observation of real situations. A particular focus of {{ethnography}} is to understand a situation from the point of view of the people involved, to become encultured into their patterns of thought, rather than to impose one's own models, theories or values. Experienced ethnographers typically resist any attempt to claim {{generalisability}} of their work, emphasising the unique nature of each situation and context. However, in practice they are very good at bringing out aspects of the situation that inform similar situations elsewhere.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/experimental error">experimental error</a></span><!--  Array
(
    [slug] =&gt; experimental error
    [name] =&gt; experimental error
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; No interesting experiment gives a precise and unambiguous result; there is always a margin of error. For example, if you sample 100 people out of the workforce of a large company, by chance you might choose those who are unusually tall or short, or those who are unusually happy or unhappy with the systems they use. Furthermore, there may be limits to the accuracy of the measurement devices you use, although this is more of a problem in physical rather than human sciences. Where the error is due to random effects, and these are understood, you may be able to estimate the margin of error, for example the {{standard error}} of the mean.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/experimenter effects">experimenter effects</a></span><!--  Array
(
    [slug] =&gt; experimenter effects
    [name] =&gt; experimenter effects
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A particular type of {{context effect}} where something about the experimenter (for example their mood) affects the responses of the subjects to the tasks.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/expert evaluation">expert evaluation</a></span><!--  Array
(
    [slug] =&gt; expert evaluation
    [name] =&gt; expert evaluation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Evaluation carried out by a usability expert rather than an actual user. These are often used for early assessment of a system to avoid the costs of a full {{usability evaluation}}. Note that the term 'expert' here refers to usability or UX expertise, not expertise in the domain of the system. Examples of expert evaluation include {{cognitive walkthrough}} and {{heuristic evaluation}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/expert slip">expert slip</a></span><!--  Array
(
    [slug] =&gt; expert slip
    [name] =&gt; expert slip
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; An expert slip is a form of error that is more frequently performed by an expert. Whilst novices may make errors due to lack of knowledge or practice, expert slips are often about performing tasks quickly or with less attention. They are typically infrequent, but may be more damaging and may be noticed far later as experts do not usually need to check the results of their actions as often as novices.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/explanation">explanation</a></span><!--  Array
(
    [slug] =&gt; explanation
    [name] =&gt; explanation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; By analysing data, visualising it, calculating various measures or performing statistical tests, we can establish that a phenomenon occurs, &lt;emph&gt;what&lt;/emph&gt; is true. &lt;strong&gt;Explanation&lt;/strong&gt; tries to work out &lt;emph&gt;why&lt;/emph&gt; it happens and &lt;emph&gt;how&lt;/emph&gt; it works &amp;ndash; deep understanding.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/explanation stage">explanation stage</a></span><!--  Array
(
    [slug] =&gt; explanation stage
    [name] =&gt; explanation stage
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; One of three main uses of empirical work during research, which often relate to the stages or goals of a research project. The other two are {{exploration}} and {{validation}}. The explanation stage is about working out why an observation is true, and how it happens in detail. The method may be theory based or qualitative and helps to establish {{mechanism}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/exploration stage">exploration stage</a></span><!--  Array
(
    [slug] =&gt; exploration stage
    [name] =&gt; exploration stage
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; One of three main uses of empirical work during research, which often relate to the stages or goals of a research project. The other two are {{explanation}} and {{validation}}. Exploration is principally about identifying the questions you want to ask. Techniques for exploration are often open-ended. They may be {{qualitative}}: {{ethnography}}, {{in-depth interviews}}, or detailed {{observation of behaviour}} whether {{in the lab}} or {{in the wild}}. However, this is also a stage that might involve (relatively) {{big data}}, for example if you have deployed software with logging, or have conducted a large-scale but open-ended {{survey}}. Data analysis may then be used to uncover patterns, which in turn suggest research questions. Note, you may not need this as a stage of research if you are beginning with an existing hypothesis, perhaps from previous phases of your own research, questions arising from other published work, or based on your own experiences.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/external non-independence">external non-independence</a></span><!--  Array
(
    [slug] =&gt; external non-independence
    [name] =&gt; external non-independence
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; When the choice of subjects in a study is connected to the topic being studied: an obvious example might be surveying preferences for MacOS or Windows in an Apple Store; less obvious might be using a mobile app-based survey on a topic that is likely to be age related. Overlooking external non-independence may lead to unintentional bias in results.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/extrapolation">extrapolation</a></span><!--  Array
(
    [slug] =&gt; extrapolation
    [name] =&gt; extrapolation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Using results that are in a studied range to make predictions about potential effects of factors outside the range.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/extreme statistics">extreme statistics</a></span><!--  Array
(
    [slug] =&gt; extreme statistics
    [name] =&gt; extreme statistics
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Extreme measures such as {{maximum}} and {{minimum}} are affected greatly by a small number of values. One consequence of this is that they have relatively complex mathematics. More importantly, they are not {{robust}}, that is they tend to vary a lot and if you do the same study or experiment again you are likely to get very different results. In contrast, measures such as the {{mean}}, {{median}} or {{quartiles}} tend to be far more stable.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/extreme values">extreme values</a></span><!--  Array
(
    [slug] =&gt; extreme values
    [name] =&gt; extreme values
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This term is used in two senses in the book. &lt;br&gt;The first describes anomalous outcomes in your data. For example, if 10 coins are tossed there is about a 1 in 50 chance of getting only 1 head or 1 tail. Really extreme values in results are relatively unlikely. If they are unexpected they are called {{outliers}}, and may arise by chance or from faulty equipment; if they are valid data points but just happen to be extreme, they are part of the random nature of the phenomenon being studied. &lt;br&gt; The second usage refers to the various parameters in a distribution or model. Even if you expect real values to be more central, experimenting ... playing! ... with more extreme values can often give you a sense of the data. For example, experiments where you toss (virtual) coins with very high bias or coin-to-coin correlation.
    [has_links] =&gt; 1
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/factor effects">factor effects</a></span><!--  Array
(
    [slug] =&gt; factor effects
    [name] =&gt; factor effects
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; When there is some form of relationship or correlation between the various factors of which you are measuring aspects. Such factors are called {{confounding variables}}. For example, preferred social media is likely to be related to musical taste due to the different age groups using different applications.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/factorial">factorial</a></span><!--  Array
(
    [slug] =&gt; factorial
    [name] =&gt; factorial
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The product of all the numbers from 1 up to a given number. The notation is [given number]!, for example 6! (6 factorial) is 6x5x4x3x2x1.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/fair coin">fair coin</a></span><!--  Array
(
    [slug] =&gt; fair coin
    [name] =&gt; fair coin
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A coin where the probability of throwing a head or a tail is the same, 0.5. 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/fair estimate">fair estimate</a></span><!--  Array
(
    [slug] =&gt; fair estimate
    [name] =&gt; fair estimate
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This is the probability of one event (say A) &lt;emph&gt;given&lt;emp&gt; knowledge of another event (B). It is usually written as P(A|B), read as &quot;the probability of A given B&quot;. For example, while the probability that it will rain on a typical day might be 50%, on a cloudy day this might be 80%; that is P(rain|cloudy) = 0.8.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/false baseline">false baseline</a></span><!--  Array
(
    [slug] =&gt; false baseline
    [name] =&gt; false baseline
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; When the scale of a {{baseline}} starts at a value other than zero. This can be very useful where the range of values would make changes in a graph difficult to see, but without care it can also be misleading, making insignificant differences appear large and exaggerating effects.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/false basepoint">false basepoint</a></span><!--  Array
(
    [slug] =&gt; false basepoint
    [name] =&gt; false basepoint
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; When the x-axis of a scale of a {{basepoint}} starts at a value other than zero. This is common for time series &ndash; what time do you start from? It can be useful for highlighting changes, but may also be misleading sometimes.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/false negative">false negative</a></span><!--  Array
(
    [slug] =&gt; false negative
    [name] =&gt; false negative
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Where there may be a real effect, but an experiment or study is not sensitive enough to detect it. {{Statistical power}} is about avoiding false negatives. A false negative conclusion is known in statistics as a {{Type II error}}. Contrast with a {{false positive}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/false origin">false origin</a></span><!--  Array
(
    [slug] =&gt; false origin
    [name] =&gt; false origin
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Where the start point of a graph is not the &lsquo;true origin&rsquo; of the data (0,0), but starts at some other value on one or both axes. The {{baseline}} is the y coordinate of the false origin and the {{basepoint}} is the x coordinate. There are often good reasons for choosing a false origin, {{false baseline}} or {{false basepoint}} in order to explore or demonstrate a small effect, but without care this can also lead to misinterpretation or even be used for deliberate misinformation.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/false positive">false positive</a></span><!--  Array
(
    [slug] =&gt; false positive
    [name] =&gt; false positive
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This is where you wrongly believe something to be true. In traditional {{significance testing}} the choice of the {{significance level}} is precisely the probability of a false positive. For example, a p&lt;5% means that there is a 1 in 20 chance of getting a false positive result. Note, though, that this only applies to the raw results; if some results are not published (the {{file drawer effect}}) the false positive result of published results may be higher. A false positive conclusion is known in statistics as a {{Type I error}}. Contrast with a {{false negative}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/feedback effects">feedback effects</a></span><!--  Array
(
    [slug] =&gt; feedback effects
    [name] =&gt; feedback effects
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Where the outputs or past results of a process or phenomenon act as inputs to its future behaviour. These effects can be positive or negative. Snowflake growth is an example of {{positive feedback}} where any small growth grows further. In contrast, a negative feedback process smooths out changes, for example any protuberances on a pebble tend to be eroded more as it is turned over on the seashore, so that the pebble becomes rounded over time. Feedback effects may cause {{nonlinearity}, because some initial change can make huge differences.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/file drawer effect">file drawer effect</a></span><!--  Array
(
    [slug] =&gt; file drawer effect
    [name] =&gt; file drawer effect
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; For every {{positive result}} published there are many more {{negative results}} ({{non-significant}}) that are withheld. Such selective publishing may be a deliberate attempt to deceive, but is more often a combination of ignorance and {{bias}}. This is exacerbated by journal reviewing practices that typically make it hard or impossible to publish {{non-significant results}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/file tree visualisation">file tree visualisation</a></span><!--  Array
(
    [slug] =&gt; file tree visualisation
    [name] =&gt; file tree visualisation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This is an outline-style {{visualisation}} similar to that found in many operating system file-system views. A similar visualisation can be used for other forms of {{hierarchical data}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/fine positioning tasks">fine positioning tasks</a></span><!--  Array
(
    [slug] =&gt; fine positioning tasks
    [name] =&gt; fine positioning tasks
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A task requiring fine motor skills and direct gaze, for example selecting a small target with an on-screen pointer.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/finite data">finite data</a></span><!--  Array
(
    [slug] =&gt; finite data
    [name] =&gt; finite data
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Numerical data that is {{bounded}} at both ends of its range of values.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/finite variance">finite variance</a></span><!--  Array
(
    [slug] =&gt; finite variance
    [name] =&gt; finite variance
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; For any real set of data the {{variance}} is bound to be a finite number, but for some {{theoretical distributions}}, notably {{power-law distributions}}, if you mathematically calculate variance it is infinite. This does impact practical statistics. Normally if you take larger and larger samples the measured variances will nearly always shift closer and closer to some finite value. However, if the underlying distribution is not finite, for example the number of contacts in a social network, then the measured variances of larger and larger samples will tend to become larger and larger. 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/fishing expedition">fishing expedition</a></span><!--  Array
(
    [slug] =&gt; fishing expedition
    [name] =&gt; fishing expedition
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This is where, consciously or unconsciously, you look through data or results to find the ones that are particularly favourable.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/fitts' law">Fitts' Law</a></span><!--  Array
(
    [slug] =&gt; fitts' law
    [name] =&gt; Fitts' Law
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; {{Paul Fitts}} found that, for simple pointing tasks, {{task completion time}} is proportional to the {{Index of Difficulty (IoD)}}, which is essentially the logarithm of the distance to target divided by the target size: IoD = log2 ( distance to target/target size ). While Fitts' original research was on physical one-dimensional pointing alternating between two metal strips, Fitts&rsquo; Law has been found to hold for many kinds of pointing tasks, with a wide variety of devices, over multiple orders of magnitude. Although Fitts&rsquo; Law holds in many situations it is not universal; crucially it depends on situations where there is {{hand&ndash;eye coordination}} and where movement error is proportional to distance moved. 
    [has_links] =&gt; 1
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/five users">five users (the myth)</a></span><!--  Array
(
    [slug] =&gt; five users
    [name] =&gt; five users (the myth)
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Often it is said that five users are enough for {{usability testing}}. This had some truth in the particular original context of {{iterative software development}}, where only enough issues needed to be uncovered for the next cycle of re-development. However, it is certainly not sufficient for statistical purposes.
    [has_links] =&gt; 1
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/fixed effect">fixed effect</a></span><!--  Array
(
    [slug] =&gt; fixed effect
    [name] =&gt; fixed effect
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A fixed effect is one that you choose, which would stay the same if you repeated the study. It is typically something the researcher wants to find out about. Contrast with a {{random effect}}. 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/font size">font size</a></span><!--  Array
(
    [slug] =&gt; font size
    [name] =&gt; font size
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The size of characters in a {{font}} is usually measured in points, where a printer's point is 1/72 of an inch. So the tall characters in a &lt;span style=&quot;font-size:12pt&quot;&gt;12pt font&lt;/span&gt; are around 1/6 of an inch high, whereas a &lt;span style=&quot;font-size:14pt&quot;&gt;14pt font&lt;/span&gt; is about 1/5 of an inch and &lt;span style=&quot;font-size:7pt&quot;&gt;7pt font&lt;/span&gt; is about 1/10 of an inch. However, the exact size does vary between fonts. On screen the physical size of pixels (the {{screen resolution}}) varies. Old displays often used to have around 70 pixels per inch, so a point and a pixel were nearly the same, but now that varies a lot more, particularly for phone displays and retina displays. The device tries to choose a pixel-height that closely matches the paper dimensions of the desired font. If the precise screen size of characters matters, the font size may be measured in pixels.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/formalist">formalist</a></span><!--  Array
(
    [slug] =&gt; formalist
    [name] =&gt; formalist
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A pragmatic way of thinking about probability: it doesn&rsquo;t matter what probability &lsquo;really&rsquo; is, so long as it satisfies the right mathematical rules. In particular, {{Bayesian statistics}} encodes beliefs as values between 0 and 1, which satisfy the rules of probability ({{Kolomogrov laws}}). This {{belief encoded as probability}} is sometimes called {{plausibility}} or {{reasonable expectation}}). Contrast with {{frequentist}} and {{idealist}}. 

    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/formative evaluation">formative evaluation</a></span><!--  Array
(
    [slug] =&gt; formative evaluation
    [name] =&gt; formative evaluation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Focused on making the system better, formative evaluation is performed on the design artefacts (sketch, prototype, or experimental system) during the cycles of design&ndash;build&ndash;
test. Its form varies from {{expert evaluation}} to large-scale {{user testing}}, and its primary purpose is to uncover usability or experience problems for the next cycle. Contrast with {{summative evaluation}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/formative stages">formative stages</a></span><!--  Array
(
    [slug] =&gt; formative stages
    [name] =&gt; formative stages
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Stages of {{software development}} where {{formative evaluation}} is carried out.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/frequentist">frequentist</a></span><!--  Array
(
    [slug] =&gt; frequentist
    [name] =&gt; frequentist
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A down-to-earth way of looking at statistics: if a frequentist says the chance of a coin landing heads is 50:50, they mean that if they keep on tossing the coin again and again and again, on average, after many many tosses, the ratio of heads to tails will be about 50:50. Contrast with {{formalist}} and {{idealist}}.
    [has_links] =&gt; 
)
  --></li>
  <li ><span class="name">Gaussian distribution</span> - See <span class="has_content"><a href="https://alandix.com/glossary/hcistats/normal distribution">Normal distribution</a></span><!--  Array
(
    [slug] =&gt; gaussian distribution
    [name] =&gt; Gaussian distribution
    [canonical] =&gt; 
    [see] =&gt; normal distribution
    [description] =&gt; 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/generative artefact">generative artefact</a></span><!--  Array
(
    [slug] =&gt; generative artefact
    [name] =&gt; generative artefact
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A thing, such as a toolkit, design methodology or property, that is used to create or to generate other things. For example, with properties we may try to design a system that is synchronous, or one with consistency. Typically there are many systems that satisfy the property (or could be produced using the toolkit or methodology), and if we asked different practitioners, each one would take the same property and create a different final system. This makes it very hard to {{evaluate}} a generative artefact, but that does not mean it is impossible to {{validate}} them.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/good-enough solution">good-enough solution</a></span><!--  Array
(
    [slug] =&gt; good-enough solution
    [name] =&gt; good-enough solution
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The point in software development where costs start to outweigh any benefits gained by continuing to refine the system.
    [has_links] =&gt; 
)
  --></li>
  <li ><span class="name">H<sub>0</sub></span> - See <span class="has_content"><a href="https://alandix.com/glossary/hcistats/null hypothesis">null hypothesis</a></span><!--  Array
(
    [slug] =&gt; h_0
    [name] =&gt; H&lt;sub&gt;0&lt;/sub&gt;
    [canonical] =&gt; 
    [see] =&gt; null hypothesis
    [description] =&gt; 
    [has_links] =&gt; 
)
  --></li>
  <li ><span class="name">H<sub>1</sub></span> - See <span class="has_content"><a href="https://alandix.com/glossary/hcistats/alternative hypothesis">alternative hypothesis</a></span><!--  Array
(
    [slug] =&gt; h_1
    [name] =&gt; H&lt;sub&gt;1&lt;/sub&gt;
    [canonical] =&gt; 
    [see] =&gt; alternative hypothesis
    [description] =&gt; 
    [has_links] =&gt; 
)
  --></li>
  <li ><span class="name">HARK</span> - See <span class="has_content"><a href="https://alandix.com/glossary/hcistats/hypothesis after the results are known">hypothesis after the results are known</a></span><!--  Array
(
    [slug] =&gt; hark
    [name] =&gt; HARK
    [canonical] =&gt; 
    [see] =&gt; hypothesis after the results are known
    [description] =&gt; 
    [has_links] =&gt; 
)
  --></li>
  <li ><span class="name">HARKing</span> - See <span class="has_content"><a href="https://alandix.com/glossary/hcistats/hypothesis after the results are known">hypothesis after the results are known</a></span><!--  Array
(
    [slug] =&gt; harking
    [name] =&gt; HARKing
    [canonical] =&gt; 
    [see] =&gt; hypothesis after the results are known
    [description] =&gt; 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/hawthorne effect">Hawthorne effect</a></span><!--  Array
(
    [slug] =&gt; hawthorne effect
    [name] =&gt; Hawthorne effect
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This when subjects in a study or experiment behave differently because they know they are being observed.  For example in an educational or work setting, subjects may pay more attention to a task when they know they are being watched and recorded.  This may simply life or depress oerall performace, but is particlarly problematic when it casues {{bias}}, in particular, if subjects are aware that the pwerson organising the study or observing the study has an interest in a particular outcome.  In {{user studies}} this often occurs when new software is being trialed.  Subjects may try harder at the favoured system, or give better subjcetive assessments, just to be nice.
    [has_links] =&gt; 1
)
  --></li>
  <li ><span class="name">HCI</span> - See <span class="has_content"><a href="https://alandix.com/glossary/hcistats/human-computer interaction">Human&ndash;Computer Interaction</a></span><!--  Array
(
    [slug] =&gt; hci
    [name] =&gt; HCI
    [canonical] =&gt; 
    [see] =&gt; human-computer interaction
    [description] =&gt; 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/hci studies">HCI studies</a></span><!--  Array
(
    [slug] =&gt; hci studies
    [name] =&gt; HCI studies
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Research in {{HCI}} often involves some form of study with people: sometimes using a {{prototype}}, sometimes in some form of {{focus group}} or {{interview}}, sometimes in a more experimental situation closer to pure psychology. The methods used are often similar to the {{user studies}} or {{user experience studies}} used by {{usability practitioners}} during {{user interface development}}, but the aims are different.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/human-computer interaction">Human&ndash;Computer Interaction</a></span><!--  Array
(
    [slug] =&gt; human-computer interaction
    [name] =&gt; Human&amp;ndash;Computer Interaction
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This is the discipline focused on understanding of the way people use and are impacted by computer technology and the way that technology can be designed to better serve human needs. This includes single users interacting directly with devices, but also the impact of computers at the level of work groups, families, organisations and communities.
    [has_links] =&gt; 1
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/hypothesis after the results are known">hypothesis after the results are known</a></span><!--  Array
(
    [slug] =&gt; hypothesis after the results are known
    [name] =&gt; hypothesis after the results are known
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; When you notice something interesting in data it is tempting to do a statistical test of this. However, there is a clear {{selection bias}}, by definition one only notices patterns that are in the data and it is therefore not so surprising that these patterns emerge as (apparently) statistically significant.  One viewpoint is that one should *never* HARK, instead use the observation as a basis to frame a {{hypothesis}} for a fresh study or experiment. However, soemtimes this is not possible or practical, for example when analysing a pre-existing or very hard to replicate dataset.  In such cases one has to be very careful to assess how nay other patterns would have been be equally or more surprising and factor this into your results.. 
    [has_links] =&gt; 1
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/hypothesis testing">hypothesis testing</a></span><!--  Array
(
    [slug] =&gt; hypothesis testing
    [name] =&gt; hypothesis testing
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; You start with a number of predetermined options (hypotheses) and then use statistical techniques to determine which of them has more or less evidence for being true. There are forms of hypothesis testing used in {{Bayesian statistics}}, but the term is more usually associated with {{traditional statistics}}, and in particular with what is sometimes called {{null hypothesis significance testing}} or {{NHST}}. Here you start with a {{null hypothesis}} (usually what you would like to be false, such as your system not being any better) and have an {{alternative hypothesis}} (typically what you would like to be true, such as your system being better!). You then use a {{significance test}} to see if the data is consistent with the null hypothesis, and if not you 'reject' the null and instead conclude that the alternative is likely to be true.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/idealist">idealist</a></span><!--  Array
(
    [slug] =&gt; idealist
    [name] =&gt; idealist
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This way of thinking about statistics is based on the philosophy of Plato, who saw the actual events of the world as mere reflections of deeper ideals. The toss of the actual coin in some ways is &lsquo;just&rsquo; an example of an {{ideal coin toss}}. If an idealist tosses a coin five times in a row, even if it happens to come up heads each time, they probably still believe that it is &lsquo;really&rsquo; a 50:50 phenomenon. Contrast with {{formalist}} and {{frequentist}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/image recognition">image recognition</a></span><!--  Array
(
    [slug] =&gt; image recognition
    [name] =&gt; image recognition
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Software that uses algorithms to interpret image data, either static pictures or video. For example, it might be used to recognise faces in a picture so that they can be more easily labelled, or to recognise hand gestures using the computer's camera. Before {{GPS}} was common in phones, some applications would match the image on the phone with known landmarks to enable  location-specific services such as historical information or geo annotation. Similar techniques are also common in {{augmented reality}} systems that overlay digital imagery over a live feed of the real world. The accuracy of the image recognition can affect the {{user experience}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/in-the-wild">in-the-wild</a></span><!--  Array
(
    [slug] =&gt; in-the-wild
    [name] =&gt; in-the-wild
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; An {{ethnographic study}} is carried out in the normal context of the subjects of the study, for example observing people in their own workplace. It is a form of {{in the wild}} study; these have high {{ecological validity}} but hold their own specific challenges, such as {{replication}} and control of {{variability}}.
    [has_links] =&gt; 1
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/income distribution">income distribution</a></span><!--  Array
(
    [slug] =&gt; income distribution
    [name] =&gt; income distribution
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Income distribution data is an example of {{continuous data}} (weekly income may be a fraction of the base currency amount), an {{empirical distribution}} (it is derived from real data, not a theoretical model), {{bounded below}} (there are no negative incomes), {{unbounded above}} (although there is always some maximum income at any point in time, there is often no fundamental limit), an {{asymmetric distribution}} (the shape is very skewed with a large number of smaller incomes and small number of larger ones ), and a {{long-tail distribution}} (there are a small number of extremely large incomes). Furthermore, for practical purposes it effectively has {{infinite variance}} (larger and larger samples typically have larger and larger variances).
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/increase effect">increase effect</a></span><!--  Array
(
    [slug] =&gt; increase effect
    [name] =&gt; increase effect
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; There are several strategies to gain {{statistical power}} by manipulating the sensitivity of a study. One way is to attempt to increase the size of the measured effect using techniques such as a {{targeted user group}}, {{distractor tasks}}, {{targeted tasks}} and {{demonic interventions}}. However, care must be taken not to mislead with these techniques as they typically have low generalisability. See the {{noise-effect-number triangle}} for other ways to increase {{statistical power}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/increase number">increase number</a></span><!--  Array
(
    [slug] =&gt; increase number
    [name] =&gt; increase number
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The standard approach to gain {{statistical power}}. Used on its own this usually requires a very large increase in the number of subjects or trials. In a more nuanced approach researchers can also {{increase effects}} and {{reduce noise}}, the other two parts of the {{noise-effect-number triangle}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/independence">independence</a></span><!--  Array
(
    [slug] =&gt; independence
    [name] =&gt; independence
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Broadly, when one measurement or factor does not give information about another. There are several kinds of independence: {{independence of measurements}}, {{independence of factor effects}}, and {{independence of sample composition}}. Contrast with {{non-independence}}.

    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/independence condition">independence condition</a></span><!--  Array
(
    [slug] =&gt; independence condition
    [name] =&gt; independence condition
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; One of the conditions for the {{Central Limit Theorem}}, which determines whether the average of large numbers of small things ends up being {{approximately Normal}}. The things being averaged or added up need to be nearly independent, on order that one of them does not become overwhelming as part of the whole.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/independent evidence">independent evidence</a></span><!--  Array
(
    [slug] =&gt; independent evidence
    [name] =&gt; independent evidence
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This is where elements of evidence (for example data or measurements) are gathered from entirely distinct sources, or where they refer to aspects of the same data source that are not in any way related. For example, we might perform a user study on a new piece of educational software in two schools in completely different cities. It is far easier to reason about independent evidence than it is when there is some form of connection between the evidence. For example, if the new software was tested on two classes in the same school, the second class might have heard about it from their friends in the first class and so their satisfaction scores might merely reflect the views of the first class. &lt;br&gt;This can be a particular problem in {{Bayesian statistics}} if the {{prior probability}} is based on multiple evidence, some of which have been influenced by a common source. It is possible to attempt to model this common factor, but that can be complex.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/independent variable">independent variable</a></span><!--  Array
(
    [slug] =&gt; independent variable
    [name] =&gt; independent variable
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The thing that is being varied experimentally, for example distance to target, or is given by the world, for example a date. Usually shown on the horizontal (x) axis. Contrast with the {{dependent variable}}, the thing that is being measured. For example, in a {{pointing task}} experiment, the independent variables might be the size of a target and the distance to it, whereas the dependent variables would be the time to reach the target and the {{error rate}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/index of difficulty">index of difficulty (IoD)</a></span><!--  Array
(
    [slug] =&gt; index of difficulty
    [name] =&gt; index of difficulty (IoD)
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The logarithm of the distance to target divided by the target size: IoD = log&lt;sub&gt;2&lt;/sub&gt; ( distance to target/target size ). Used in {{Fitts' Law}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/individual differences">individual differences</a></span><!--  Array
(
    [slug] =&gt; individual differences
    [name] =&gt; individual differences
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Differences between subjects that may influence the effect being studied. For some studies these individual differences can mask the thing being examined (create {{noise}}). Ways to reduce the impact of individual differences in such situations include using a very {{narrow user group}}, having the same users perform on different systems (a {{within-subjects design}}), and choosing participants of the same age and experience to {{reduce noise}}. Each of these has implications for the {{validity}} of results. Within {{interaction design}} and {{user experience design}} it is crucial to remember that people are different. Sometimes you have to create a 'one size fits all' application, but if you can understand differences between users this can lead to better designs, whether different applications targeted at different user groups, {{customisation}} options within a single application, or {{intelligent interfaces}} that adapt themselves to the user.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/infinite variance">infinite variance</a></span><!--  Array
(
    [slug] =&gt; infinite variance
    [name] =&gt; infinite variance
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; For any real set of data the {{variance}} is bound to be a finite number, but for some {{theoretical distributions}}, notably {{power-law distributions}}, if you mathematically calculate variance, it is infinite. This does impact practical statistics. Normally if you take larger and larger samples the measured variances will nearly always shift closer and closer to some finite value. However, if the underlying distribution is not finite, for example the number of contacts in a social network, then the measured variances of larger and larger samples will tend to become larger and larger. 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/intelligent interfaces">intelligent interfaces</a></span><!--  Array
(
    [slug] =&gt; intelligent interfaces
    [name] =&gt; intelligent interfaces
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; These are systems that use some form of {{artificial intelligence}} or {{machine learning}} as part of {{user interaction}}. Many of us will see examples of this daily as web sites adapt content and especially adverts according to models based on our past behaviour. More advanced systems might try to help us to complete tasks or maybe automate repeated tasks based on examples.
    [has_links] =&gt; 1
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/inter-related factors">inter-related factors</a></span><!--  Array
(
    [slug] =&gt; inter-related factors
    [name] =&gt; inter-related factors
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Factors that influence the effect being studied, such as the experience of users being related to their age; also changes in a design which affect other parts of the design and make it difficult to study individual aspects.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/interference effect">interference effect</a></span><!--  Array
(
    [slug] =&gt; interference effect
    [name] =&gt; interference effect
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A type of {{order effect}} where a user does less well in a second test because they have got used to the detailed way things were done in an earlier test.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/internal non-independence">internal non-independence</a></span><!--  Array
(
    [slug] =&gt; internal non-independence
    [name] =&gt; internal non-independence
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; When subjects are likely to be similar to one another, but in no particular direction with regard to your question, for example in a survey of adjacent people in a football match queue. If they all support the same team it might simply mean that they all travelled on the same bus.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/interpolate">interpolate</a></span><!--  Array
(
    [slug] =&gt; interpolate
    [name] =&gt; interpolate
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Inferring information about values between the available measured values, for example an  assumption that the reading speed of an 11-point font falls between the reading speed data you have for the 10-point and 12-point variants. Contrast with {{extrapolation}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/interquartile range">interquartile range</a></span><!--  Array
(
    [slug] =&gt; interquartile range
    [name] =&gt; interquartile range
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This is a measure of {{variation}} calculated as the difference between the {{upper quartile}} and {{lower quartile}}. That is, half of the data items will fall within the interquartile range. It is typically a far more {{robust}} measure than the more common {{standard deviation}}, but less tractable for calculations.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/iterative development">iterative development</a></span><!--  Array
(
    [slug] =&gt; iterative development
    [name] =&gt; iterative development
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A software design cycle that includes repeated stages of evaluation of evolving designs, prototypes or systems. The two main kinds of evaluation are {{formative evaluation}} at each stage to improve the design and {{summative evaluation}} at the end to assess whether the final design is sufficiently good.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/iterative evaluation">iterative evaluation</a></span><!--  Array
(
    [slug] =&gt; iterative evaluation
    [name] =&gt; iterative evaluation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The {{formative evaluation}} in {{iterative development}}
    [has_links] =&gt; 
)
  --></li>
  <li ><span class="name">job of statistics</span> - See <span class="has_content"><a href="https://alandix.com/glossary/hcistats/the job of statistics">the job of statistics</a></span><!--  Array
(
    [slug] =&gt; job of statistics
    [name] =&gt; job of statistics
    [canonical] =&gt; 
    [see] =&gt; the job of statistics
    [description] =&gt; 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/kolomogrov laws">Kolomogrov laws</a></span><!--  Array
(
    [slug] =&gt; kolomogrov laws
    [name] =&gt; Kolomogrov laws
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A collection of three axioms developed by {{Andrey Kolmogorov}} in 1933, which summarise the laws of {{probability}}. The three axioms state that (i) the probability of any set of events is greater than or equal to zero (you can't have a negative chance of doing something); (ii) the probability of the set of all events is one (the probability that something will happen is one); and (iii) the probability of the union of disjoint sets of events is the sum of the probability of each set (probabilities add up when there is no overlap). For example, if we throw a (possibly loaded) die, the sets of events are subsets of {1,2,3,4,5,6}. Axiom (i) says, inter alia, that Prob({1,2}) &gt; 0 (meaning the probability of either a 1 or 2); axiom (ii) says that Prob({1,2,3,4,5,6})=1; and axiom (iii) that, inter alia, Prob( Union({1,2},{3,6}) ) =Prob( {1,2,3,6}) = Prob({1,2}) + Prob({3,6}).
    [has_links] =&gt; 1
)
  --></li>
  <li ><span class="name">larger samples</span> - See <span class="has_content"><a href="https://alandix.com/glossary/hcistats/sample size">sample size</a></span><!--  Array
(
    [slug] =&gt; larger samples
    [name] =&gt; larger samples
    [canonical] =&gt; 
    [see] =&gt; sample size
    [description] =&gt; 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/learning effects">learning effects</a></span><!--  Array
(
    [slug] =&gt; learning effects
    [name] =&gt; learning effects
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A type of {{order effect}} where a user is better at a second test because they have already got used to the general ideas of an application in the first test. 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/likelihood">likelihood</a></span><!--  Array
(
    [slug] =&gt; likelihood
    [name] =&gt; likelihood
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The {{conditional probability}} of a measurement given a hypothetical state of the world (or typically some specific parameters of the world). For example, the probability that six tosses of a coin will come out heads given the coin is fair is 1/64, so if this result occurs, the likelihood that the coin is fair is 1/64. This likelihood is used in slightly different ways by {{traditional statistics}} and {{Bayesian statistics}} to work back from the measurement to assess its evidence for or against the value of the underlying property or state of the world.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/likert scale">Likert scale</a></span><!--  Array
(
    [slug] =&gt; likert scale
    [name] =&gt; Likert scale
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A scale to measure degrees of agreement or satisfaction. The values are often coded as numbers and there is an order but otherwise no specific meaning to the values. This is {{ordinal data}}. &lt;br&gt; An example of survey question using a Likert scale might be:&lt;br&gt; The top level menu items were:  unintelligible &ndash; 1 &ndash; 2 &ndash; 3 &ndash; 4 &ndash; 5 &ndash; always easy to interpret.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/linear regression">linear regression</a></span><!--  Array
(
    [slug] =&gt; linear regression
    [name] =&gt; linear regression
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A statistical process that creates a best fit line through data points. The goodness of fit of the line is often measured using the {{Pearson correlation coefficient}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/linearity condition">linearity condition</a></span><!--  Array
(
    [slug] =&gt; linearity condition
    [name] =&gt; linearity condition
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; One of the conditions for the {{Central Limit Theorem}}. It says that the way in which lots of near-independent factors combine together must have a linear nature such as adding together lots of small factors, or averaging lots of larger ones.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/log-normal distribution">Log-Normal distribution</a></span><!--  Array
(
    [slug] =&gt; log-normal distribution
    [name] =&gt; Log-Normal distribution
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This distribution is common where there is some form of multiplicative effect: for example, with {{task completion times}} we would expect longer times to have greater variability. Tests designed for the {{Normal distribution}} cannot be used, but the data can be transformed by taking the logarithm of the raw times, then the resulting {{transformed data}} is often {{approximately Normal}} and can be analysed using off-the-shelf statistics.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/long-tail distribution">long-tail distribution</a></span><!--  Array
(
    [slug] =&gt; long-tail distribution
    [name] =&gt; long-tail distribution
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A {{distribution}} where there are very high (or very low) values but they occur with very low frequency. Examples include income distributions (small numbers of very rich individuals) and numbers of contacts in {{social network data}} (a small number of people with very large numbers of 'friends'). A specific example is a {{power-law distribution}}, which is often used as a {{model}} of {{network data}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/lower bound">lower bound</a></span><!--  Array
(
    [slug] =&gt; lower bound
    [name] =&gt; lower bound
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This is the smallest possible value that a variable or dataset can have. For example, when rolling a die, the lower bound is 1. In contrast, some data is {{unbounded below}}, that is there is no meaningful absolute minimum.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/lower quartile">lower quartile</a></span><!--  Array
(
    [slug] =&gt; lower quartile
    [name] =&gt; lower quartile
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The value such that 25% (one quarter) of the data values are less and 75% more. It is one of three {{quartiles}}, the others being the {{median}} (50% point) and {{upper quartile}} (75% point).
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/machine learning">machine learning</a></span><!--  Array
(
    [slug] =&gt; machine learning
    [name] =&gt; machine learning
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Various algorithms that allow a computer to learn from data in a way analogous to, but typically very different from, human or animal learning. This can be used as a research tool, or as part of some form of {{intelligent interface}}. Machine learning (ML) algorithms have existed for many years, but the vast amounts of data now available allow us to study phenomena in a way that would have been impossible only a few years ago. However, the analysis of such data using machine learning techniques can lead to unintentional {{cherry picking}}, as the ML will pick out the few patterns that work amongst a vast number that do not.
    [has_links] =&gt; 
)
  --></li>
  <li ><span class="name">manipulate sensitivity</span> - See <span class="has_content"><a href="https://alandix.com/glossary/hcistats/sensitivity">sensitivity</a></span><!--  Array
(
    [slug] =&gt; manipulate the sensitivity
    [name] =&gt; manipulate sensitivity
    [canonical] =&gt; 
    [see] =&gt; sensitivity
    [description] =&gt; 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/matching subjects">matching subjects</a></span><!--  Array
(
    [slug] =&gt; matching subjects
    [name] =&gt; matching subjects
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; When the same subject repeats a test under different conditions (e.g. evaluating two systems) there may be {{learning effects}}. One way around this is to have different subjects for each condition, but to have two or more subjects who are very similar (in gender, age, skills), and to allocate one from each pair (or matched group) to each condition. Matching subjects makes results more robust, requiring fewer assumptions, but typically has less {{statistical power}} than a {{within-subjects experiment}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/mathematics">mathematics</a></span><!--  Array
(
    [slug] =&gt; mathematics
    [name] =&gt; mathematics
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; There are many branches of mathematics, but within {{statistics}} the most important part is {{probability theory}}, which enables us to reason explicitly about probabilities.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/mean">mean (&mu;)</a></span><!--  Array
(
    [slug] =&gt; mean
    [name] =&gt; mean (&amp;mu;)
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; An average obtained by adding the numbers then dividing the sum by the number of numbers that were added. Contrast this with the {{median}} (the middle in the sense that 50% of values are below it) and the {{mode}}, the value with the largest frequency. The arithmetic mean is easy to deal with mathematically, but often the {{median}} is a more {{robust}} measure.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/mechanism">mechanism</a></span><!--  Array
(
    [slug] =&gt; mechanism
    [name] =&gt; mechanism
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Mechanism is about understanding how and why something occurs: the steps and processes, which buttons were pressed, what screens viewed, what information was examined and how this all creates a larger phenomenon. It is essential to understand mechanism in order to draw lessons and make predictions that {{generalise}} and  {{extrapolate}} beyond particular data and situations.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/median">median</a></span><!--  Array
(
    [slug] =&gt; median
    [name] =&gt; median
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A central measure that is the mid-point at which precisely half the values are below and half above. It is typically more {{robust} than the arithmetic {{mean}}, but the latter is easier for certain kinds of calculation. The term {{average}} most commonly refers to the arithmetic {{mean}}, but can also refer to the {{median}}, especially for {{asymmetric distributions}}, where the {{median}} is typically a little smaller than the {{mean}}, because the latter can be influenced by a few large {{outliers}}. For example, 'average income' usually refers to the median, not the {{mean}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/menu navigation">menu navigation</a></span><!--  Array
(
    [slug] =&gt; menu navigation
    [name] =&gt; menu navigation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Using menus as an interaction mechanism to find one's way around a system.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/meta-analysis">meta-analysis</a></span><!--  Array
(
    [slug] =&gt; meta-analysis
    [name] =&gt; meta-analysis
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A method of using multiple studies by different people, typically in the published literature, in order to cross-validate and find emerging patterns. To enable meta-analysis, methods and results of studies must be clearly reported. Making all study data and methodology available in the public domain is a good way to enable meta-analysis, with the proviso that ethical issues must be considered.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/mode">mode</a></span><!--  Array
(
    [slug] =&gt; mode
    [name] =&gt; mode
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The most likely value: in an {{empirical distribution}} the data value with the highest frequency (the hump!), or in the {{theoretical distribution}} the value with maximum probability. Sometimes a distribution has more than one mode, for example in university courses the more mathematical courses often have one group of students towards the higher end (those that 'get it'), then a gap and then another group with lower marks. Distributions with more than one mode sometimes happen because there is a mixed group (e.g. expert and novice users), or because of {{positive feedback}} effects, for example when students who understand some concepts learn further concepts, whereas those who misunderstand one thing are confused by everything that follows.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/model">model</a></span><!--  Array
(
    [slug] =&gt; model
    [name] =&gt; model
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This is some sort of formal or informal description, {{mechanism}} or mathematical formula that allows one to reason about or predict {{real-world}} phenomena.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/model the bias">model the bias</a></span><!--  Array
(
    [slug] =&gt; model the bias
    [name] =&gt; model the bias
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A calculation based on measurements may have {{bias}}, that is it may on average either over- or underestimate the {{real world effect}} that it is trying to estimate. Sometimes it is possible to work out how large this bias is and then correct the calculation to take that into account. For example, imagine you have a ladder with rungs 20cm apart. You throw a ball up towards the ladder, record which pair of rungs the ball falls between, then multiply the rung it goes over by 0.20 to give the ball height in metres. This will typically underestimate the height of the ball &ndash; it is a {biased estimate}}. However, we can model the bias: it is clear that the underestimation is about half the distance between rungs. Having modelled the bias, we can add 10cm to the original calculation and obtain a {{fair estimate}} of the ball height.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/monty hall problem">Monty Hall problem</a></span><!--  Array
(
    [slug] =&gt; monty hall problem
    [name] =&gt; Monty Hall problem
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A probability problem derived from a 1950s quiz challenge, where participants had to guess which of three doors concealed a prize, but were allowed to change their their initial choice after the quizmaster (Monty Hall) had opened one of the remaining doors. An example of a problem where common sense may not help and careful calculation is needed.
    [has_links] =&gt; 1
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/more subjects">more subjects</a></span><!--  Array
(
    [slug] =&gt; more subjects
    [name] =&gt; more subjects
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The most obvious way to increase the {{statistical power}} of an experiment is to have more subjects. This is one form of the {{increase number}} strategy when you want to {{gain power}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/muller-lyer illusion">M&uuml;ller&ndash;Lyer illusion</a></span><!--  Array
(
    [slug] =&gt; muller-lyer illusion
    [name] =&gt; M&amp;uuml;ller&amp;ndash;Lyer illusion
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A well-known optical illusion: there are two horizontal parallel lines of equal length, the top one bounded by outward-facing arrows and the lower one by inward-facing arrows. The lower line looks longer, though it is the same length. The illusion was described by Franz Carl M&uuml;ller-Lyer, a German sociologist, in 1889.
    [has_links] =&gt; 1
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/multiple causes">multiple causes</a></span><!--  Array
(
    [slug] =&gt; multiple causes
    [name] =&gt; multiple causes
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In an {{ecologically valid}} {{HCI study}}, {{in the wild}} or even in the lab, there are usually multiple causes for effects; that is different possible explanations for the effects you are seeing, several of which are often true. This happens because controls are difficult to implement and individuals and contexts always vary. {{Claims analysis}} is a method developed by Carroll and Rosson that specifically embraces multiple causes of the success or failure of systems.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/multiple statistics">multiple statistics</a></span><!--  Array
(
    [slug] =&gt; multiple statistics
    [name] =&gt; multiple statistics
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A form of {{cherry picking}} that tries different kinds of statistical technique till something comes out 'right'. For example, a non-parametric test, followed by a t-test, followed by a Bayesian test, and so on. Sometimes a change is justified, for example if it is clear that the wrong kind of test for the distribution was chosen initially, but the reasons for making such a change must be based on the data, not the results. 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/multiple tests">multiple tests</a></span><!--  Array
(
    [slug] =&gt; multiple tests
    [name] =&gt; multiple tests
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A form of {{cherry picking}} where multiple tests are performed on different aspects of the same data but only the successful ones are reported. For example, a few tests that show some effect may be selected and the rest ignored, or tests that show a desired effect may be selected while those that indicate the opposite are ignored.
    [has_links] =&gt; 
)
  --></li>
  <li ><span class="name">multiplicity adjustments</span> - See <span class="has_content"><a href="https://alandix.com/glossary/hcistats/multiplicity control">multiplicity control</a></span><!--  Array
(
    [slug] =&gt; multiplicity adjustments
    [name] =&gt; multiplicity adjustments
    [canonical] =&gt; 
    [see] =&gt; multiplicity control
    [description] =&gt; 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/multiplicity control">multiplicity control</a></span><!--  Array
(
    [slug] =&gt; multiplicity control
    [name] =&gt; multiplicity control
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Techniques for adjusting for the effect of running {{multiple tests}} on data. If you do ten tests you are ten times as likely to see an apparently positive result by pure chance; if you do 1000 tests then it starts to become highly likely that something will look promising even if there is no real effect. Some form of multiplicity adjustment or multiplicity control is therefore critical to prevent {{false positive}} results.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/narrow user group">narrow user group</a></span><!--  Array
(
    [slug] =&gt; narrow user group
    [name] =&gt; narrow user group
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Using a very narrow user group can {{reduce noise}} by reducing {{individual differences}} in knowledge and skill, thereby increasing {{statistical power}}. This makes it easier to see the effect of a novel interaction technique, but it also reduces the validity for generalisation beyond that group.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/negative binomial">negative binomial</a></span><!--  Array
(
    [slug] =&gt; negative binomial
    [name] =&gt; negative binomial
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In a test where only two outcomes are possible, the negative Binomial distribution gives you the probability of the number of times one outcome occurs before the other outcome occurs a specified number of times: for example, when a coin is tossed, the number of heads that fall before you get two tails. A special case of this is the {{Poisson distribution}}, which is when you are waiting for the very first outcome.
    [has_links] =&gt; 1
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/negative correlation">negative correlation</a></span><!--  Array
(
    [slug] =&gt; negative correlation
    [name] =&gt; negative correlation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Where two things are related and one is less likely to happen if the other happens, or, for two numerical variables, when one tends to decrease as the other increases. In the case of numbers, a perfect negative correlation occurs if one variable (say x) is linked to the other (say y) by an exact linear formula such as y = 10 &ndash; 2x. However, the effects can also be non-linear, such as y = 10 &ndash; 2x&lt;sup&gt;3&lt;/sup&gt;  and of course in statistics there is usually a substantial dose of {{noise}}! When performing {{linear regression}} the {{correlation coefficient}} will be negative. 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/negative results">negative results</a></span><!--  Array
(
    [slug] =&gt; negative results
    [name] =&gt; negative results
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; These are the times when you are not able to state anything definitively. If, given a hypothesis, you are able to say with a high degree of certainty either &quot;yes&quot; it is true or (paradoxically) &quot;no&quot; it is not, these are both regarded as {{positive results}}. If you are unable to say &quot;yes&quot; or &quot;no&quot; then that is a negative result. In traditional {{hypothesis testing}} a positive result is when you have rejected the {{null hypothesis}} and hence decided you have evidence for the {{alternative hypothesis}}; a negative result is when the test is {{non significant}}. In this case there is no definitive &quot;no&quot;, merely &quot;yes&quot; and &quot;not proven&quot;. A negative result may mean there is no effect, or that your study did not have sufficient {{statistical power}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/network data">network data</a></span><!--  Array
(
    [slug] =&gt; network data
    [name] =&gt; network data
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Data derived from social networks, such as friendship circles, or similar kinds of linked objects such as web pages. The ways in which these arise often involve {{positive feedback}}; for example, if a web page has many incoming links, people are more likely to find it, and hence more likely to add yet more links. For this kind of data the {{Normal distribution}} does not apply and the data is more likely to obey a {{power law}}. 
    [has_links] =&gt; 
)
  --></li>
  <li ><span class="name">new statistics</span> - See <span class="has_content"><a href="https://alandix.com/glossary/hcistats/alternative statistics">alternative statistics</a></span><!--  Array
(
    [slug] =&gt; new statistics
    [name] =&gt; new statistics
    [canonical] =&gt; 
    [see] =&gt; alternative statistics
    [description] =&gt; 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/noise-effect-number triangle">noise&ndash;effect&ndash;number triangle</a></span><!--  Array
(
    [slug] =&gt; noise-effect-number triangle
    [name] =&gt; noise&amp;ndash;effect&amp;ndash;number triangle
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Three factors that interact to influence the {{statistical power}} of an experiment or study: the amount of {{noise}}, the {{effect size}} and the number of {{subjects}} or {{trials}}. You are likely to get a {{positive result}} if the effect size is a lot greater than the size of the noise divided by the number of subjects/trials (effect &gt;&gt; noise/number). Hence a smaller noise, a larger effect or more subjects or trials all lead to increased power. Techniques for increasing power can target one or more of the elements of the noise-effect-number triangle.
    [has_links] =&gt; 
)
  --></li>
  <li ><span class="name">nominal data</span> - See <span class="has_content"><a href="https://alandix.com/glossary/hcistats/categorical data">categorical data</a></span><!--  Array
(
    [slug] =&gt; nominal data
    [name] =&gt; nominal data
    [canonical] =&gt; 
    [see] =&gt; categorical data
    [description] =&gt; 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/non-independence">non-independence</a></span><!--  Array
(
    [slug] =&gt; non-independence
    [name] =&gt; non-independence
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; When one measurement or factor influences or gives information about another. Non-independence may increase {{variability}}, lead to misattribution of effects, or suggest completely the wrong effect. Contrast with {{independence}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/non-significant result">non-significant result</a></span><!--  Array
(
    [slug] =&gt; non-significant result
    [name] =&gt; non-significant result
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; When the results from your experiment or study do not reach a pre-defined measure of {{significance}}, that is where they could simply have occurred randomly with a given probability. Often the {{p-value}} used for this is a standard one such as 5% or 1%, largely because in days past it was only possible to calculate these in books of statistical tables. It is very important to remember that non-significant does &lt;strong&gt;not&lt;/strong&gt; mean {{no effect}}, but simply not sufficient evidence to be confident that there is a real effect &ndash; {{not proven}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/nonlinearity">nonlinearity</a></span><!--  Array
(
    [slug] =&gt; nonlinearity
    [name] =&gt; nonlinearity
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This is when the relationship between things cannot be represented by a simple linear formula such as y=3+2x; that is where there is some sort of bend or curve in the graph of the data. In statistics this tends to be harder to see because there is usually a lot of noise in the data. Note particularly that when performing {{linear regression}}, two variables might have a large {{correlation coefficient}} but the relationship can still be non-linear.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/nonparametric statistics">nonparametric statistics</a></span><!--  Array
(
    [slug] =&gt; nonparametric statistics
    [name] =&gt; nonparametric statistics
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; One of two broad classes of statistical technique, where there is no known {{theoretical distribution}} or it is too complex, and universal properties such as the ordering of values must be relied upon instead. Contrast with {{parametric statistics}}.
    [has_links] =&gt; 1
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/normal approximation">Normal approximation</a></span><!--  Array
(
    [slug] =&gt; normal approximation
    [name] =&gt; Normal approximation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; When we approximate some other distribution using the {{Normal distribution}}. For example, if we have tossed 100 coins and counted the number of heads, this is strictly a {{Binomial distribution}}; however, we might approximate this by a Normal distribution with {{mean}} of 50 and {{standard deviation}} of 0.05.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/normal distribution">Normal distribution</a></span><!--  Array
(
    [slug] =&gt; normal distribution
    [name] =&gt; Normal distribution
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This is the ubiquitous bell-shaped curve. Some phenomena such as heights seem to naturally follow this distribution, whilst others, including coin-toss scores (the {{Binomial distribution}}), end up looking {{approximately Normal}} for sufficiently many tosses. This {{theoretical distribution}} is also known as a {{Gaussian distribution}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/not proven">not proven</a></span><!--  Array
(
    [slug] =&gt; not proven
    [name] =&gt; not proven
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The state of not being able to make a strong statement either for or against the effect, system, or theory being tested. If results are non-significant, it can never be reasoned that a hypothesis is false, only that it is not statistically proven. This is why {{statistical power}} is so important, to increase the likelihood that an experiment or study is likely to reveal an effect if it is present. &lt;br&gt;In Scottish law, courts can return three possible verdicts: &quot;guilty&quot;, &quot;not guilty&quot; and &quot;not proven&quot;, where &quot;not guilty&quot; is a strong statement of innocence, whereas &quot;not proven&quot; is when the court does not feel there is sufficient evidence to make a definitive judgement in either direction. In contrast, English courts only have the options &quot;guilty&quot; or &quot;not guilty&quot;, where the latter encompasses both strong evidence of innocence and lack of evidence. Traditional {{hypothesis testing}} is more like the English court, whereas using {{confidence intervals}} or certain kinds of {{Bayesian statistics}} (where applicable) may allow a strong negative &quot;no difference&quot; to be distinguished from the weak &quot;don't know&quot;.


    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/null hypothesis">null hypothesis</a></span><!--  Array
(
    [slug] =&gt; null hypothesis
    [name] =&gt; null hypothesis
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In {{hypothesis testing}} the null hypothesis, often written H&lt;sub&gt;0&lt;/sub&gt;, is usually the thing you would like to show is false. For example, that your updated interface design has not made any improvement.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/numeric data">numeric data</a></span><!--  Array
(
    [slug] =&gt; numeric data
    [name] =&gt; numeric data
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Data that can be counted or measured as a number, such as completion time in an experiment, or number of heads in coin tosses.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/observed outcome">observed outcome</a></span><!--  Array
(
    [slug] =&gt; observed outcome
    [name] =&gt; observed outcome
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This is a term for the actual data or evidence seen, as opposed to a predicted value or probabilistic range of possibilities.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/odds ratio">odds ratio</a></span><!--  Array
(
    [slug] =&gt; odds ratio
    [name] =&gt; odds ratio
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The odds ratio is the general term for the ratio of a particular outcome given two different conditions. For example how much more likely a person is to get type 2 diabetes if they are obese, compared with if they are of a lower weight. The {{Bayes factor}} is a particular example of an odds ratio, where it is used as a measure of the strength of evidence between hypotheses.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/one-tailed test">one-tailed test</a></span><!--  Array
(
    [slug] =&gt; one-tailed test
    [name] =&gt; one-tailed test
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A test where data is only compared in one direction. For example, when comparing two systems you may only care about seeing if the new system is &lt;em&gt;better&lt;/em&gt;. If they are the same or if the new system is worse, these are equally failures of the new design. So, you care that the new system has a  &lt;em&gt;faster&lt;/em&gt; {{response time}} than the old one (one direction), or if it has a &lt;em&gt;lower&lt;/em&gt; {{error rate}} (one direction). Contrast with {{two-tailed test}}. 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/ontologically random">ontologically random</a></span><!--  Array
(
    [slug] =&gt; ontologically random
    [name] =&gt; ontologically random
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; (An effect) that is random in its actual state of being, such as the radioactive decay of an atom, contrasted with (an effect) that is {{epistemologically random}} &mdash; random in your state of knowledge. For example, if you are about to toss a coin, the future result of the toss is ontologically random; however, once you have tossed the coin but not yet looked at it, it is epistemologically random: you would still treat it as a 50:50 odds if asked to bet, but actually it is already on one state or the other, you just don't know which.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/open scholarship">open scholarship</a></span><!--  Array
(
    [slug] =&gt; open scholarship
    [name] =&gt; open scholarship
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The placement of research into the public domain: instructions, software, raw study data, and derived data. This enables {{replication}} and {{meta-analysis}}, as well as new research that tests alternative theories, or mines the data for emergent patterns. Consideration must be given to ethical issues such as privacy and transparency, practical issues such as documentation, and cultural issues around the use of data in academic disciplines.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/open situations">open situations</a></span><!--  Array
(
    [slug] =&gt; open situations
    [name] =&gt; open situations
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Where we do not have control of the context within which a system under study is being used. It is a feature of {{in the wild}} research or studies of deployed systems in actual use.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/opinion poll">opinion poll</a></span><!--  Array
(
    [slug] =&gt; opinion poll
    [name] =&gt; opinion poll
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A form of survey or {{count data}} where members of the public are asked for their judgments or intentions, for example voting intentions before an election, or opinion of a government policy.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/order effects">order effects</a></span><!--  Array
(
    [slug] =&gt; order effects
    [name] =&gt; order effects
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; One of a number of ways in which measurements may be related to one another, which can occur where, for example, a test compares a user using a particular interface followed by the same user using a different interface. Types of order effects include: {{positive learning effects}}, where the user is better at the second interface because they have already become familiar with the application while using the first; and {{interference effects}}, where the user does less well in the second interface because they have got used to the features of the first.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/ordinal data">ordinal data</a></span><!--  Array
(
    [slug] =&gt; ordinal data
    [name] =&gt; ordinal data
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Data that has an order, but otherwise no specific meaning to the values; for example the degrees of agreement or satisfaction in a {{Likert scale}}. Sometimes this can be converted to {{binary data}} for ease of analysis, though this loses some information. Contrast with {{arithmetical data}} and {{categorical data}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/outliers">outliers</a></span><!--  Array
(
    [slug] =&gt; outliers
    [name] =&gt; outliers
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Unexpectedly extreme values in data, which may be caused by a fault in equipment, or other irrelevant effect, or may occur by chance. While they may be valid data points within the random nature of the phenomenon being studied, even a few outliers can have a disproportionate effect on statistics such as arithmetic mean and variance. For some purposes, better results are obtained by removing the most extreme outliers, but this can introduce the potential for {{cherry picking}}, so, where possible, a strategy for dealing with outliers should be determined prior to any analysis.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/p-curve">p-curve</a></span><!--  Array
(
    [slug] =&gt; p-curve
    [name] =&gt; p-curve
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; If we conduct an experiment where there is no real effect, we would expect, by random chance, that we would get, around 5% of the time, a {{p-value}} of 0.05 or lower and, about 1% of the time, a {{p-value}} less than 0.01 &ndash; that is precisely what these {{significance levels}} mean: on average, a flat curve for p-values vs number of experiments with that value. In contrast, where there is a real effect we would hope to see a considerably higher proportion of lower {{p-values}}. Because of the {{file drawer effect}} not all the results of experiments are published, so typically we are unlikely to see the {{negative results}}. However, it has been suggested by Simonsohn et al. that the shape of the {{p-curve}}, that is the graph of results with each {{p-value}}, can give a sense of the health of a discipline and the impact of the {{file drawer effect}}. If this curve is flat, it suggests that there are few real results and most are pure chance; if it is highly curved, with less than a 5:1 ratio of 5% or more to 1% or more results, then it is likely that the discipline is using studies with sufficient {{statistical power}} and that most or many of the published results do in fact correspond to real effects.
    [has_links] =&gt; 1
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/p-hacking">p-hacking</a></span><!--  Array
(
    [slug] =&gt; p-hacking
    [name] =&gt; p-hacking
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Where a researcher manipulates their results in some way to obtain a {{p-value}} that is deemed {{significant}} and hence able to be published. This may be deliberate, or simply the result of poor practice or poor training. One example of {{p-hacking}} is to try out many, many different experiments and data analysis methods, and only report those that are {{significant}}, meaning that all the {{non-significant}} experiments are ignored; a practice called the {{file drawer effect}}. The impact of {{p-hacking}} is that there may be published results that are pure chance and do not correspond to a real effect; this increases the number of {{false positives}} in the literature, hence reducing the validity of the entire discipline.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/p-value">p-value</a></span><!--  Array
(
    [slug] =&gt; p-value
    [name] =&gt; p-value
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The {{likelihood}} that the data we have measured, or the result we have obtained, could have arisen by chance given the {{null hypothesis}}. Note that because this is related to the probability of making an error, big p-values are bad and small p-values are good. &lt;br&gt;The null hypothesis is often some form of {{no effect}}, such as there being no improvement due to a new system being used. P-values are often quoted at specific levels such as 5%, 1% or 0.1% (p&lt;0.05, p&lt;0.01, or p&lt;0.001 respectively), but this is largely because they were difficult to calculate in the past and hence one had to look them up in a book of statistical tables where they had been (laboriously) precalculated. Recent advice from the {{Royal Statistical Society}} is to quote the exact p-value as a measure of evidence, rather than rely on magic values. &lt;br&gt;Note also that even if there is no effect, a p-value of 5% is likely to occur one time in twenty and a p-value of 1% one time in a hundred, that is one may occasionally get a {{false positive}} result, or {{Type I error}}. A low p-value means that a {{false positive}} is less likely.
    [has_links] =&gt; 1
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/pair subjects">pair subjects</a></span><!--  Array
(
    [slug] =&gt; pair subjects
    [name] =&gt; pair subjects
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A special, but common, case of {{matching subjects}} where you match pairs of subjects with similar characteristics and make sure that one from each pair is assigned to two different conditions.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/parameters">parameters</a></span><!--  Array
(
    [slug] =&gt; parameters
    [name] =&gt; parameters
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; These are variables in {{models}} or {{theoretical distributions}} that typically need to be fitted to data or provided with values based on other knowledge. For example, the {{Binomial distribution}} has two parameters &ndash; Binomial(N.p). One parameter, N, is the number of trials and the other parameter, p, the probability of success of each ({{independent}}) trial. If we use this for 10 tosses of a {{fair coin}}, we can fill in these parameters based on our knowledge: N=10 and p=0.5. However, if we are uncertain whether the coin is biased, we may toss it a hundred times (N=100), but use the fact that it is a binomial distribution to fit p, that is to estimate the {{bias}} of the coin. &lt;br&gt;In a more {{HCI}}-specific example, {{Fitts' Law}} says that time to hit a target is A + B x IoD (where IoD is the {{index of difficulty}}), but the {{parameters}} A and B vary on the characteristics of the device used (e.g. mouse vs trackpad vs finger pointing).
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/parametric statistics">parametric statistics</a></span><!--  Array
(
    [slug] =&gt; parametric statistics
    [name] =&gt; parametric statistics
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A statistical technique where there is a known theoretical distribution, and only a few unknown {{parameters}} need to be estimated. For example, people's heights, where we may know that heights have a {{Normal distribution}}, but need to estimate average height and its {{standard deviation}}. Statisticians prefer to use parametric statistics where possible, because they are less likely to miss a real effect; the statistics have greater {{power}}. Contrast with {{nonparametric statistics}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/pearson correlation coefficient">Pearson correlation coefficient</a></span><!--  Array
(
    [slug] =&gt; pearson correlation coefficient
    [name] =&gt; Pearson correlation coefficient
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This measures the level of linear relationship between two variables. It lies in the range [-1,+1] where +1 denotes a perfect positive relationship and -1 a perfect negative relationship. It is calculated by working out the {{covariance}} (average product) of the two variables and dividing by the product of their individual {{standard deviation}}:  ( cov(x,y) / ( &amp;sigma;&lt;sub&gt;x&lt;/sub&gt; &amp;times; &amp;sigma;&lt;sub&gt;y&lt;/sub&gt;).
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/peripheral vision">peripheral vision</a></span><!--  Array
(
    [slug] =&gt; peripheral vision
    [name] =&gt; peripheral vision
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Where a person is not looking directly at something, but it is still visible out of their central vision. Human peripheral vision is not accurate enough for fine {{positioning tasks}}.
    [has_links] =&gt; 
)
  --></li>
  <li ><span class="name">perturbation analysis </span> - See <span class="has_content"><a href="https://alandix.com/glossary/hcistats/sensitivity analysis">sensitivity analysis</a></span><!--  Array
(
    [slug] =&gt; perturbation analysis 
    [name] =&gt; perturbation analysis 
    [canonical] =&gt; 
    [see] =&gt; sensitivity analysis
    [description] =&gt; 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/pietree visualisation">PieTree</a></span><!--  Array
(
    [slug] =&gt; pietree visualisation
    [name] =&gt; PieTree
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The PieTree is an interactive {{constant area visualisation}} for hierarchical data with some form of numerical attribute including values at the nodes &ndash; for example in a website where there are visits to both the main page of a section and also to sub-pages. The visualisation starts off like a standard pie chart where each segment represents a branch of the tree (e.g. section of the site), except there is also a small circle in the middle representing the value at the root (e.g home page). If any of the segments is selected it breaks into smaller segments for each sub-branch, and so on. 
    [has_links] =&gt; 1
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/plausibility">plausibility</a></span><!--  Array
(
    [slug] =&gt; plausibility
    [name] =&gt; plausibility
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In {{Bayesian statistics}} the 'probabilities' used for the {{prior distribution}} and hence the resulting {{posterior distribution}} are {{belief encoded as probability}} &ndash; that is using a number between 0 and 1 to represent a level of belief. To distinguish these from true probabilities they are called {{plausibility}} or {{reasonable expectation}}. For the purposes of mathematical calculation, {{plausibility}} satisfies the {{Kolomogrov Laws}} of probability and so can be treated mathematically in the same way even though its meaning is subtly different. 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/pointing task">pointing task</a></span><!--  Array
(
    [slug] =&gt; pointing task
    [name] =&gt; pointing task
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A class of low-level device interaction task where a mouse, trackpad, stylus or finger is used  to move towards a target or follow some sort of path. This is typically to target an element onscreen in order to act on it, for example to drag it or select it. The time for positioning tasks often satisfies some variant of {{Fitts' Law}}. 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/poisson distribution">Poisson distribution</a></span><!--  Array
(
    [slug] =&gt; poisson distribution
    [name] =&gt; Poisson distribution
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This {{probability distribution}} is used when you have some sort of {{binary data}} (such as the toss of a coin, or whether a die rolls a 6), a fixed number of trials (say tossing the coin 10 times) and each trial is {{independent}}. The Poisson distribution gives the probability of getting a certain number of 'successes' (e.g. head in a coin toss). The Poisson distribution has two parameters, N (the number of trials) and p (the probability that a single trial is a 'success'). The probability of n successes is then Comb(N,n)  p&lt;sup&gt;n&lt;/sup&gt; (1-p)&lt;sup&gt;N-n&lt;/sup&gt;, where Comb(N.n) is the number of combinations of n things out of N = N! / ( n! (N-n)! ). When N becomes large, the central portion of the Poisson distribution is {{approximately Normal}}.
    [has_links] =&gt; 1
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/population">population</a></span><!--  Array
(
    [slug] =&gt; population
    [name] =&gt; population
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A complete group of people one wants to know about, for example all the employees in a company, or all users of a product. By gathering data about a small {{sample}} from the population an estimate may be made about the population as a whole.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/positive correlation">positive correlation</a></span><!--  Array
(
    [slug] =&gt; positive correlation
    [name] =&gt; positive correlation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Where two things are related and one is more likely to happen if the other happens or if one thing tends to become greater when the other does. Contrast with {{negative correlation}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/positive feedback">positive feedback</a></span><!--  Array
(
    [slug] =&gt; positive feedback
    [name] =&gt; positive feedback
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Where effects accumulate and increase the overall effect. Examples are: snowflake growth where ice forms more easily on sharp tips, so any small growth grows further; earthquakes, where a very small release of tension in a fault can lead to a larger release in adjacent areas and culminate in an earthquake; citation rates, where the more citations a paper gains, the more it is seen and the more it is cited further. 
    [has_links] =&gt; 
)
  --></li>
  <li ><span class="name">positive learning</span> - See <span class="has_content"><a href="https://alandix.com/glossary/hcistats/learning effects">learning effects</a></span><!--  Array
(
    [slug] =&gt; positive learning
    [name] =&gt; positive learning
    [canonical] =&gt; 
    [see] =&gt; learning effects
    [description] =&gt; 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/positive results">positive results</a></span><!--  Array
(
    [slug] =&gt; positive results
    [name] =&gt; positive results
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Results that are {{significant}} and either prove or disprove a hypothesis. There is a cultural tendency for only positive results to be published &ndash; the {{file drawer effect}}. Contrast with a {{negative result}} where you are unable to make a strong statement other than {{not proven}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/post-hoc hypothesis">post-hoc hypothesis</a></span><!--  Array
(
    [slug] =&gt; post-hoc hypothesis
    [name] =&gt; post-hoc hypothesis
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A form of {{cherry picking}} where the initial hypothesis is replaced by one suggested by a pattern in the data. Although this may be less than ideal it is pragmatically necessary in some contexts. However, where a post-hoc hypothesis is used its use should be openly acknowledged and any potential impact on results should be assessed.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/posterior distribution">posterior distribution</a></span><!--  Array
(
    [slug] =&gt; posterior distribution
    [name] =&gt; posterior distribution
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In {{Bayesian methods}} this is the probability calculated after combining fresh evidence with the {{prior distribution}}. It is given by {{Bayes rule}}: Posterior(A&lt;sub&gt;f&lt;/sub&gt;|E)  = P(E|A&lt;sub&gt;f&lt;/sub&gt;)xPrior(A&lt;sub&gt;f&lt;/sub&gt;)/(&amp;Sigma;&lt;sub&gt;all i&lt;/sub&gt; P(E|A&lt;sub&gt;i&lt;/sub&gt;)xPrior(A&lt;sub&gt;i&lt;/sub&gt;)), where E is the evidence, and A&lt;sub&gt;i&lt;/sub&gt; ranges over all alternatives including the focus one A&lt;sub&gt;f&lt;/sub&gt;. Note that in {{Bayesian statistics}} the {{posterior distribution}} is typically not a real probability but a {{quantified belief}} or {{ plausibility}}.
    [has_links] =&gt; 
)
  --></li>
  <li ><span class="name">power (statistical)</span> - See <span class="has_content"><a href="https://alandix.com/glossary/hcistats/statistical power">statistical power</a></span><!--  Array
(
    [slug] =&gt; power (statistical)
    [name] =&gt; power (statistical)
    [canonical] =&gt; 
    [see] =&gt; statistical power
    [description] =&gt; 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/power-law distribution">power-law distribution</a></span><!--  Array
(
    [slug] =&gt; power-law distribution
    [name] =&gt; power-law distribution
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A form of {{long-tail distribution}} where the frequency is inversely proportional to some power: that is freq(n) = 1/n&lt;sup&gt;&amp;lambda;&lt;/sup&gt;. These occur in many forms of {{network data}} such as number of web page links, number of contacts in {{social networks}}, or number of citations of papers, and in other areas such as city sizes, wealth of individuals or companies, and size of earthquakes. They are often the result of {{positive feedback}} effects; for example, papers that are highly cited are more widely seen and hence attract more citations; or people with lots of money have more opportunities to make more. The power {{parameter}} &amp;lambda; varies between different kinds of data, but is often a little under 2 for web data. This is critical, because if the value is less than 2, the distribution does not have {{finite variance}}, one of the prerequisites for the {{Central Limit Theorem}}. The practical impact is that averages based on the data are not {{approximately Normal}} and hence you &lt;em&gt;must not&lt;/em&gt; use tests designed for the {{Normal distribution}}, even when the {{sample size}} is very large.
    [has_links] =&gt; 1
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/pre-registration">pre-registration</a></span><!--  Array
(
    [slug] =&gt; pre-registration
    [name] =&gt; pre-registration
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A way to avoid the {{file drawer effect}} by recording a statement before starting an experiment to specify the hypothesis being tested and which statistical technique will be used. This forces justification of the reasons for any {{post-hoc hypothesis}}. It is standard practice in many areas of science, and becoming so in {{HCI}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/prior belief">prior belief</a></span><!--  Array
(
    [slug] =&gt; prior belief
    [name] =&gt; prior belief
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The assumptions that are held to be true before starting an experiment. {{Bayesian reasoning}} tries to quantify these beliefs.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/prior distribution">prior distribution</a></span><!--  Array
(
    [slug] =&gt; prior distribution
    [name] =&gt; prior distribution
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; When applying {{Bayes rule}} this is the probability one would assign to various parameters &lt;em&gt;before&lt;/em&gt; considering additional evidence. It is then combined with the evidence from the data to obtain the {{posterior probability}}. In {{Bayesian statistics}}, the values for this prior may not be a true probability but rather {{prior belief}} {{encoded as probability}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/probabilistic phenomena">probabilistic phenomena</a></span><!--  Array
(
    [slug] =&gt; probabilistic phenomena
    [name] =&gt; probabilistic phenomena
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Events that have a likelihood of happening but are not predetermined.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/probability distribution">probability distribution</a></span><!--  Array
(
    [slug] =&gt; probability distribution
    [name] =&gt; probability distribution
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The function or histogram that for any data value or potential outcome of a measurement tells you its probability. This can be an {{empirical distribution}} based on measurements of real data, such as an {{income distribution}} that tells you what proportion of people earn any specific amount. Alternatively, it can be a {{theoretical distribution}} based on mathematical analysis, for example the {{Binomial distribution}} that tells you the probability of getting exactly n heads out of 20 coin tosses.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/proof by contradiction">proof by contradiction</a></span><!--  Array
(
    [slug] =&gt; proof by contradiction
    [name] =&gt; proof by contradiction
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; An argument form in logic or mathematics, where what is believed to be false is asserted as if it were true, in order to reason from that to something that is clearly contradictory, and then use the contradiction to argue that what was first asserted must indeed be false. Compare with the {{null hypothesis}} in statistics.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/public domain">public domain</a></span><!--  Array
(
    [slug] =&gt; public domain
    [name] =&gt; public domain
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This is when the various materials used or produced in research are made available to everyone in some publicly accessible form. These materials include instructions, software (if applicable), and of course the raw study data, whether this is survey reports, video, or keystroke logging, as well as derived data, all the way to the data that lies behind the graphs in your published papers.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/public sector borrowing">public sector borrowing</a></span><!--  Array
(
    [slug] =&gt; public sector borrowing
    [name] =&gt; public sector borrowing
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Borrowing by government or government bodies as opposed to borrowing by companies and individuals (private sector).
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/publication bias">publication bias</a></span><!--  Array
(
    [slug] =&gt; publication bias
    [name] =&gt; publication bias
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; If only {{positive results}} are publshied, then it may appaer that they are more {{significant}} (in all senses of the word) than they really are.  This is becasue of you do enough experiments or studies, then by raodm chance one of them will come up 5% significant about 1 time in 20. Publication bias occurs artly because researchers self-select to report things that are most impressive, but also because journal and conference reviewers may not accept papers with {{negative results}}. In {{qualitative}} studies, there can be similar effects, if you look at enough people, or enough events you will see some interesting things, and may tend to report these leaing to a sense that they are common, whereas they may not be at all typical.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/publish or perish">publish or perish</a></span><!--  Array
(
    [slug] =&gt; publish or perish
    [name] =&gt; publish or perish
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A cultural definition of academic success that requires peer-reviewed publication of research results.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/qualitative methods">qualitative methods</a></span><!--  Array
(
    [slug] =&gt; qualitative methods
    [name] =&gt; qualitative methods
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Research methods such as rich observations, ethnography, or post-experiment interviews, which are suited to exploratory research, revealing the processes and mechanisms by which something occurs and answering how and why questions. Contrast with {{quantitative methods}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/quantified belief">quantified belief</a></span><!--  Array
(
    [slug] =&gt; quantified belief
    [name] =&gt; quantified belief
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In {{Bayesian statistics}} you have to represent your level of belief in terms of a {{probability}}. This not a real probability, but it does obey the {{Kolomogrov laws}}, so behaves like probability in many circumstances. Crucially, when a posterior 'probability' is calculated this too is a measure of belief informed by additional evidence.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/quantitative data">quantitative data</a></span><!--  Array
(
    [slug] =&gt; quantitative data
    [name] =&gt; quantitative data
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Any data based on numbers or measurable values. This may include objectively measurable values such as {{task completion times}} or more subjective data such as {{opinion polls}}. Contrast with {{qualitative methods}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/quartile">quartile</a></span><!--  Array
(
    [slug] =&gt; quartile
    [name] =&gt; quartile
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The result of dividing a range into four sections with an equal number of items in each section when sorted. For example, if your (sorted) data values are 1.3, 1.7, 2.2, 2.5, 2.6, 2.6, 2.7, 2.9, 3.0, 3.1, 3.3, 3.4, then you divide these into four equal sections:  1.3, 1.7, 2.2  |  2.5, 2.6, 2.6 |  2.7, 2.9, 3.0 | 3.1, 3.3, 3.4; and the quartiles are therefore 2.35, 2.65, and 3.05. The middle of these is called the {{median}} and the other two are the lower and upper quartile respectively. Of the data, 25% of values are smaller than the {{lower quartile}}, 50% smaller than the {{median}} and 75% smaller than the {{upper quartile}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/r">R</a></span><!--  Array
(
    [slug] =&gt; r
    [name] =&gt; R
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; R is a programming language and environment designed for statistical computation. It is an open source project and free for anyone to use. Its user base has grown rapidly and it challenges {{SPSS}} as the &lt;emph&gt;de facto&lt;/emph&gt; statistical tool.
    [has_links] =&gt; 1
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/random effect">random effect</a></span><!--  Array
(
    [slug] =&gt; random effect
    [name] =&gt; random effect
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A random effect is one that might vary if you repeated the same study; for example, the skill level of participants, or weather on the day of an experiment. Random effects need to be treated as an additional source of noise. Contrast with a {{fixed effect}}. 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/random noise">random noise</a></span><!--  Array
(
    [slug] =&gt; random noise
    [name] =&gt; random noise
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Random effects that are irrelevant for whatever you are trying to measure or study, but add {{noise}}, making the results harder to interpret. For example, if you are measuring someone's weight, they may sway slightly, meaning the scales register slightly more or less than their actual weight; or if you ask someone whether they like a new interface feature their response may depend on how good their day has been. Part of the {{job of statistics}} is to see back, past this noise, to the underlying facts about the {{real world}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/randomise orders">randomise orders</a></span><!--  Array
(
    [slug] =&gt; randomise orders
    [name] =&gt; randomise orders
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A way to mitigate {{order effects}} by performing tests in a random order: for example some subjects try system A followed by system B,  and other subjects do the opposite.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/real world">real world</a></span><!--  Array
(
    [slug] =&gt; real world
    [name] =&gt; real world
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; At a practical level, actual things and happenings as opposed to more theoretical or mathematical abstractions. However, what is 'real' in statistics has a variety of philosophical meanings, from concrete events to more idealised notions such as the 'typical user' of a system.
    [has_links] =&gt; 
)
  --></li>
  <li ><span class="name">reasonable expectation</span> - See <span class="has_content"><a href="https://alandix.com/glossary/hcistats/plausibility">plausibility</a></span><!--  Array
(
    [slug] =&gt; reasonable expectation
    [name] =&gt; reasonable expectation
    [canonical] =&gt; 
    [see] =&gt; plausibility
    [description] =&gt; 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/reduce noise">reduce noise</a></span><!--  Array
(
    [slug] =&gt; reduce noise
    [name] =&gt; reduce noise
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A way to increase {{statistical power}}. Variability due to factors that are not controlled or not known about is called noise. Noise can be reduced by (1) controlling conditions in a study to limit variability and (2) obtaining information about factors that may affect results and modelling these effects.
    [has_links] =&gt; 
)
  --></li>
  <li ><span class="name">regression analysis</span> - See <span class="has_content"><a href="https://alandix.com/glossary/hcistats/linear regression">linear regression</a></span><!--  Array
(
    [slug] =&gt; regression analysis
    [name] =&gt; regression analysis
    [canonical] =&gt; 
    [see] =&gt; linear regression
    [description] =&gt; 
    [has_links] =&gt; 
)
  --></li>
  <li ><span class="name">removing outliers</span> - See <span class="has_content"><a href="https://alandix.com/glossary/hcistats/outliers">outliers</a></span><!--  Array
(
    [slug] =&gt; removing outliers
    [name] =&gt; removing outliers
    [canonical] =&gt; 
    [see] =&gt; outliers
    [description] =&gt; 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/repeatability">repeatability</a></span><!--  Array
(
    [slug] =&gt; repeatability
    [name] =&gt; repeatability
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The extent to which the results of an experiment or study can be recreated by others. The language used for this varies, but the ASA distinguishes between {{reproducibility}} and {{replication}}. If others can use your data, etc., to obtain the same published result, that is {{reproducibility}}. For {{replication}}, the question is: if others follow your procedures with fresh subjects, etc., they will come to a similar result? Furthermore, some writers use the term 'repeatability' in a very specific sense: does the same scientist in the same laboratory with the same equipment get identical results? Some of these distinctions are far less clear in {{HCI}} and human sciences; for example, due to {{learning effects}} if the same subject repeats a task we do not expect a similar outcome.
    [has_links] =&gt; 1
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/replication">replication</a></span><!--  Array
(
    [slug] =&gt; replication
    [name] =&gt; replication
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The extent to which a study or experiment can be replicated by others with the same results. True replication is difficult in HCI, where differences in subjects and contexts mean that even the most controlled experimental studies cannot be precisely repeated. A form of {{repeatability}}; contrast with {{reproducibility}}. 
    [has_links] =&gt; 1
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/replication crisis">replication crisis</a></span><!--  Array
(
    [slug] =&gt; replication crisis
    [name] =&gt; replication crisis
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The situation where many attempts to repeat scientific studies have failed to reproduce the original outcomes.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/replichi">RepliCHI</a></span><!--  Array
(
    [slug] =&gt; replichi
    [name] =&gt; RepliCHI
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; An initiative of ACM CHI to increase {{repeatability}}, which led to the addition of a &ldquo;validation and refutation&rdquo; category in some subsequent conferences, and to long-term changes in the ACM {{CHI Conference}} publication guidelines.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/representative sample">representative sample</a></span><!--  Array
(
    [slug] =&gt; representative sample
    [name] =&gt; representative sample
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A small group who exhibit the typical characteristics of a larger group. Results from the study of a representative sample can be extrapolated to the larger group. Sometimes this sample may be chosen randomly, or may be deliberately chosen so as to ensure that important features such as gender, age or experience match those of the relevant {{population}} as a whole.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/reproducibility">reproducibility</a></span><!--  Array
(
    [slug] =&gt; reproducibility
    [name] =&gt; reproducibility
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; 
    [has_links] =&gt; 1
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/residuals">residuals</a></span><!--  Array
(
    [slug] =&gt; residuals
    [name] =&gt; residuals
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Many statistical procedures have some sort of underlying {{model}} for which you fit {{parameters}} based on your data. If there were no {{noise}} this might fit your data precisely, but in practice there is a difference between the value predicted by the model and the actual data points; this difference is called the residual. &lt;br&gt;For example, suppose you are performing a user study and have pre-tested the skill level of users with a value between 0 and 10. You measure their {{task completion time}} for a particular task and then calculate a {{linear regression}} to obtain a best fit line: time = 50 &ndash; 2*skill_level. Although this is a best fit line, it does not fit every user perfectly. Perhaps Aled has a task completion time of 43 seconds and a skill level of 8. The model would predict a time of 50 &ndash; 2*8 = 34 seconds; the difference of 9 (43&ndash;34) is the residual. 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/response time">response time</a></span><!--  Array
(
    [slug] =&gt; response time
    [name] =&gt; response time
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The time it takes for someone to respond to some sort of event, notification, alarm or cue. This may be affected by the form in which the event is presented (e.g. a loud sound vs a subtle change at the edge of the screen), and by the cognitive state of the user (e.g. waiting attentively or highly focused on another task). In the case when some sort of decision is also needed, then other issues will also be important, such as the number of choices and preparedness or implicit biases of the user.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/restricted device">restricted device</a></span><!--  Array
(
    [slug] =&gt; restricted device
    [name] =&gt; restricted device
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A method to {{reduce noise}} by limiting the functionality of a device; for example, locking a mobile phone so that it can only use the test app, thereby ensuring that activity is directed toward the aspects being studied.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/restricted system">restricted system</a></span><!--  Array
(
    [slug] =&gt; restricted system
    [name] =&gt; restricted system
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A method to {{reduce noise}} by limiting the functionality of a system. For example, in a social media application you might disable everything apart from music-related features if these are the focus of your study.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/restricted tasks">restricted tasks</a></span><!--  Array
(
    [slug] =&gt; restricted tasks
    [name] =&gt; restricted tasks
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A method to {{reduce noise}} by limiting the activity of participants, while still maintaining a level of {{ecological validity}}; for example, giving them a scripted task to perform, but within their normal environment.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/royal statistical society">Royal Statistical Society</a></span><!--  Array
(
    [slug] =&gt; royal statistical society
    [name] =&gt; Royal Statistical Society
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; One of the main statistical professional bodies in the UK, and the oldest in the world.
    [has_links] =&gt; 1
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/sample">sample</a></span><!--  Array
(
    [slug] =&gt; sample
    [name] =&gt; sample
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; You typically cannot measure or survey every person in the relevant {{population}} of interest: for example measure the heights of everyone in the country, or ask every possible user of your system to do a task. Instead you choose a small sample, one that you hope is large enough to give sufficient information about the full population, but is manageable.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/sample composition">sample composition</a></span><!--  Array
(
    [slug] =&gt; sample composition
    [name] =&gt; sample composition
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Who or what is in your sample. When considering sample composition you need to think about whether it is a {{representative sample}} of the {{population}} you are interested in. In addition, you need to consider whether the members of your sample are {{independent}}: for example, if you do a {{snowball sample}}, people who know other people may share political views, support the same sports team or have similar musical taste.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/sample size">sample size</a></span><!--  Array
(
    [slug] =&gt; sample size
    [name] =&gt; sample size
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The number of people (or things) in a {{sample}}. Increasing the sample size typically reduces the {{variability}} of any form of {{average}} or similar measure and is one of the ways to increase {{statistical power}}. However, you typically have to increase the sample size a lot &ndash; for example to reduce the {{standard deviation}} by 10 you need a sample size 100 times larger. The {{noise-effect-number triangle}} suggests other factors that can be manipulated to increase {{statistical power}}. 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/sample variance">sample variance</a></span><!--  Array
(
    [slug] =&gt; sample variance
    [name] =&gt; sample variance
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A measure of the {{variability}} of your sampled data. In the simplest case it is the average square of the difference between the data and the sample {{mean}}, but dividing by N&ndash;1 rather than N as you would for the {{population variance}} or the variance of a {{theoretical distribution}}. That is, if the data is x&lt;sub&gt;1&lt;/sub&gt;, x&lt;sub&gt;2&lt;/sub&gt;, ... x&lt;sub&gt;N&lt;/sub&gt;, the sample variance is ( &amp;Sigma;  (x&lt;sub&gt;i&lt;/sub&gt;&ndash;mean) &lt;sup&gt;2&lt;/sup&gt;) / N&ndash;1 &lt;br&gt;The reason for dividing by N&ndash;1 rather than N is that the mean is usually estimated from the sample data and so is slightly closer to the middle of the data than the real (theoretical or population) mean. Think about the extreme case with just two measured values, the sample mean is exactly in the middle, so their spread from that would give a {{biased estimate}} of the true population variance. This use of N&ndash;1 is an example where we can {{model the bias}} and hence correct it. &lt;br&gt;More generally the sample variance is the sum of the square of the {{residuals}} divided by the {{degrees of freedom}}, (N&ndash;n), where n is the number of things fitted by the statistical model. &lt;br&gt;The sample variance is often affected strongly by {{outliers}}, which then has an impact on the {{statistical power}}; one of the reason for (carefully!) removing outliers during data cleaning.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/sample-based approaches">sample-based approaches</a></span><!--  Array
(
    [slug] =&gt; sample-based approaches
    [name] =&gt; sample-based approaches
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Methods that focus on obtaining large numbers of subjects in a trial, or large numbers of systems that share the property being studied, in order to generalise results. Such an approach is typically impractical for studying properties of {{generative artefacts}} (things that make other things) because of the vast number of trials required. Alternatives include employing smaller amounts of {{rich data}} or very targeted {{evaluation}} combined with some form of {{justification}} (reasoned argument) as part of a holistic {{validation}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/sampling bias">sampling bias</a></span><!--  Array
(
    [slug] =&gt; sampling bias
    [name] =&gt; sampling bias
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This is when the way we choose a sample is not random, and becasue of this creates a {{bias}} in the eventual results.  For example, if we look through a crowd, we may be more likely to notice tall people.  This is a form of {{selection bias}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/scale free distribution">scale free distribution</a></span><!--  Array
(
    [slug] =&gt; scale free distribution
    [name] =&gt; scale free distribution
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; An alternative name for a {{power-law distribution}}, which emphasises the {{scale free}} property. That is, if you imagine zooming in or out of the graph of a power law it looks the same shape at every scale. In contrast, if you have a {{Normal distribution}} (or indeed many kinds of distribution) and you zoom out, the distribution gets more narrow and the mean looks closer to zero. At a practical level Normally distributed data, such as people's heights, have a natural scale.
    [has_links] =&gt; 1
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/scatter plots">scatter plots</a></span><!--  Array
(
    [slug] =&gt; scatter plots
    [name] =&gt; scatter plots
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A simple graph that shows all the data as separate x,y points so that the nature of the spread can easily be seen, for example clumps, or extreme values. For a simple scatter plot, two data fields need to be selected for the x and y axes; however, more fields can be visualised by showing multiple plots with different pairs of fields and by colouring points differently depending on attributes.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/schrodinger's cat">Schrodinger's cat</a></span><!--  Array
(
    [slug] =&gt; schrodinger's cat
    [name] =&gt; Schrodinger's cat
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A quantum physics thought experiment in which a cat in a box may be both alive and dead at the same time.
    [has_links] =&gt; 
)
  --></li>
  <li ><span class="name">s.d.</span> - See <span class="has_content"><a href="https://alandix.com/glossary/hcistats/standard deviation">standard deviation (s.d., &sigma;)</a></span><!--  Array
(
    [slug] =&gt; sd
    [name] =&gt; s.d.
    [canonical] =&gt; 
    [see] =&gt; standard deviation
    [description] =&gt; 
    [has_links] =&gt; 
)
  --></li>
  <li ><span class="name">s.e.</span> - See <span class="has_content"><a href="https://alandix.com/glossary/hcistats/standard error">standard error (s.e.)</a></span><!--  Array
(
    [slug] =&gt; se
    [name] =&gt; s.e.
    [canonical] =&gt; 
    [see] =&gt; standard error
    [description] =&gt; 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/seeded">seeded</a></span><!--  Array
(
    [slug] =&gt; seeded
    [name] =&gt; seeded
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In a computer, 'random' numbers are usually obtained from a {{pseudo-random number generator}} &ndash; this follows a fixed deterministic algorithm to generate each number, but is so convoluted that the numbers are for most practical purposes random. However, because it is underneath a fixed algorithm, every time you restart a program you get the same series of 'random' numbers as the last time. To deal with this the random number generator is {{seeded}}, that is some data is fed in to start the generator off so that it creates  a new series of numbers. The 'seed' may include detailed system timings that depend on the momentary load of the machine, or other things that are hard to predict. However, for certain security applications that need large random numbers, this seeding process can be the weak point.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/selection bias">selection bias</a></span><!--  Array
(
    [slug] =&gt; selection bias
    [name] =&gt; selection bias
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This is when the way the people who are considered in a study uninitentionally includes some kinds of people more than others and thus nduces a {{bias}} in the way resuts are interpretted.  This may occur due to {{sampling bias}} in the way people are initially selected, or other effects such as {{self-selection}} of participants in study, or participants dropping out.  Note that this affects both {{qualitative}} and {{quantitative}} analysis.
    [has_links] =&gt; 1
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/selective publishing">selective publishing</a></span><!--  Array
(
    [slug] =&gt; selective publishing
    [name] =&gt; selective publishing
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; When one only publishes particular (usually favourable!) results. This may be as a deliberate policy, but may also be due to the policy of journals only accepting {{positive results}}. The impact of {{selective publishing}} is that many results, in particular {{negative results}}, are never published, a phenomenon known as the {{file drawer effect}}. Because of this the number of {{false positives}} in the published literature may be far higher than one would expect given the quoted {{significance level}} of results, one of the causes of the {{statistical crisis}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/sensitivity">sensitivity</a></span><!--  Array
(
    [slug] =&gt; sensitivity
    [name] =&gt; sensitivity
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Experiments, data analysis and visualisation may differ in their &lt;strong&gt;sensitivity&lt;/strong&gt;, that is how easy it is for small differences to be apparent. For example, if you have histograms showing the heights of the 30 tallest mountains in the world, it may be hard to see the difference after the first few, as there are many that are close to 26,000  feet (7,000 m) tall. However, if you use a {{false baseline}} at 25,000 feet, the heights are very clearly different. In experimental design {{distractor tasks}} are also often useful as they expose mental load.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/sensitivity analysis">sensitivity analysis</a></span><!--  Array
(
    [slug] =&gt; sensitivity analysis
    [name] =&gt; sensitivity analysis
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Sensitivity analysis is about trying to assess how much effect small changes in parameters or data values have on the ultimate result. If a small change in an uncertain or random value will create a large change in the result, then that will reduce one's confidence in the ultimate result. We may perform {{sensitivity analysis}} by making {{perturbations}} (small changes) in values, either mathematically calculating the impact or simply re-running the analysis with the perturbed values &ndash; if the small perturbation has made a big change, we know we need to be careful. This is sometimes also called {{perturbation analysis}}.&lt;br&gt;One example of this is in calculating different measures of {{variation}}. We know (mathematically) that {{outliers}} have a large impact on the {{standard deviation}}: it is very sensitive; in contrast the {{interquartile range}} is not affected at all by a single {{outlier}}: it is {{robust}}. Another example is found in {{Bayesian statistics}}, where the choice of {{prior distribution}} affects the {{posterior}}. One could therefore in principle re-work the analysis with slightly different priors ... although in practice this is rarely done.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/significance level">significance level</a></span><!--  Array
(
    [slug] =&gt; significance level
    [name] =&gt; significance level
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The threshold at which the decision is made to reject the {{null hypothesis}}. The smallest {{significance level}} that is normally regarded as reasonable evidence is 5%, which means that the probability of a {{false positive}} (or {{Type I error}}) is less than 1 in 20. There is now considerable debate on whether choosing such an arbitrary {{critical value}} is helpful. Instead it is suggested one should quote the exact {{p-value}}, allowing the reader to assess the strength of evidence this suggests.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/significance test">significance test</a></span><!--  Array
(
    [slug] =&gt; significance test
    [name] =&gt; significance test
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The generic term for various tests that are used to see if a {{hypothesis}} about data is {{statistically significant}} (that is, unlikely to happen by chance). Examples include the {{Student's t test}} used to see if the {{means}} of two sets of data are different, and the {{Pearson correlation coefficient}}, used to test whether there is some form of relation between two variables in {{linear regression}}. Typically the test will be based on some form of {{theoretical distribution}} leading to a {{parametric test}}, but there are also {{nonparametric tests}} based purely on the order of data items, and also tests based on {{simulation}} or {{empirical data}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/significant difference">significant difference</a></span><!--  Array
(
    [slug] =&gt; significant difference
    [name] =&gt; significant difference
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; When you are comparing two sets of data, or two groups within a dataset, and the difference between them (usually the difference between their respective {{means}}) reaches a chosen {{significance level}}.
    [has_links] =&gt; 
)
  --></li>
  <li ><span class="name">significant effect</span> - See <span class="has_content"><a href="https://alandix.com/glossary/hcistats/significant result">significant result</a></span><!--  Array
(
    [slug] =&gt; significant effect
    [name] =&gt; significant effect
    [canonical] =&gt; 
    [see] =&gt; significant result
    [description] =&gt; 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/significant range">significant range</a></span><!--  Array
(
    [slug] =&gt; significant range
    [name] =&gt; significant range
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A range of results that reach a chosen {{significance level}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/significant result">significant result</a></span><!--  Array
(
    [slug] =&gt; significant result
    [name] =&gt; significant result
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; When the data you have obtained in an experiment or study reaches a chosen {{significance level}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/simpson's paradox">Simpson's paradox</a></span><!--  Array
(
    [slug] =&gt; simpson's paradox
    [name] =&gt; Simpson's paradox
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; When the trend in the overall data is in the opposite direction to the {{trend}} in every group within the data. This can lead to apparently conflicting arguments from the same  data. The original paradox was first noticed over 100 years ago, but one of the most widely cited examples was at Berkeley in the 1970s, when an analysis of admissions suggested there was bias against female admissions. It turned out that in fact in each individual course women were more likely to be admitted than men, but that women were more likely to apply for the more selective courses, leading to the apparent bias.
    [has_links] =&gt; 1
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/simulation methods">simulation methods</a></span><!--  Array
(
    [slug] =&gt; simulation methods
    [name] =&gt; simulation methods
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Running large numbers of computer simulations based on mathematical probabilities or observed data, to discover the likelihood of results that have been observed in an experiment or study.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/single cause">single cause</a></span><!--  Array
(
    [slug] =&gt; single cause
    [name] =&gt; single cause
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; When there is a single causal factor that is creating measurable differences, usually in a {{controlled experiment}}. For example, if we take two large groups of people matched for all relevant background and skills and then give one group a standard interface (system A) and the other group a new interface (system B), the change in interface is a {{single cause}} for any differences in the average behaviour. &lt;br&gt; Having a {{single cause}} makes analysis easier, but is difficult to achieve in HCI. This is partly because we often want to perform {{in the wild}} research or other forms of more {{ecologically valid}} study where we cannot control the context. Even where we do have a relatively controlled situation, such as the system A vs B study, the complex nature of interface design means that there will be many differences between the two, making it difficult to pin down precisely which factors give rise to an observed effect.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/size of the noise">size of the noise</a></span><!--  Array
(
    [slug] =&gt; size of the noise
    [name] =&gt; size of the noise
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The quantity of random or {{uncontrolled factors}} in a test. Part of the {{noise-effect-number triangle}}. Reducing the size of the noise increases {{statistical power}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/size of the study">size of the study</a></span><!--  Array
(
    [slug] =&gt; size of the study
    [name] =&gt; size of the study
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The number of  trials or participants in a study. Part of the {{noise-effect-number triangle}}. Increasing the size of the study increases {{statistical power}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/snowball sample">snowball sample</a></span><!--  Array
(
    [slug] =&gt; snowball sample
    [name] =&gt; snowball sample
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; An initial set of contacts is used as the first set of subjects, and they are then asked to suggest any of their own contacts who might take part in the survey. A danger of this approach is that there may be {{internal non-independence}} if the contacts of each individual are likely to share the feature or ability being investigated &ndash; for example contacts may well share political views, or educational background.
    [has_links] =&gt; 1
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/social network data">social network data</a></span><!--  Array
(
    [slug] =&gt; social network data
    [name] =&gt; social network data
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Data arising from social networks, such as the number of friends or contacts. Often social network data follows a {{power-law distribution}} and so {{statistical tests}} designed for {{approximately Normal}} data &lt;em&gt;cannot&lt;/em&gt; be used.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/software design">software design</a></span><!--  Array
(
    [slug] =&gt; software design
    [name] =&gt; software design
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Software design involves many aspects, for example the choice of internal protocols between front-end and back-end services; however, from an HCI point of view, the crucial aspects are those relating to the {{user interface}}, {{user experience}} and the way in which it fits into broader {{socio-technical systems}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/software development process">software development process</a></span><!--  Array
(
    [slug] =&gt; software development process
    [name] =&gt; software development process
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; There are many processes by which software is designed and created that relate to the kind of software, application domain and organisation, but those parts that involve any form of interaction with the user typically follow an {{iterative process}}. Initial design activity leads to the making of a demonstrable artefact: from storyboards and sketches to wireframes, hi-res prototypes or a running system. The artefact is subjected to testing or evaluation: {{formative evaluation}} in the early stages and {{summative evaluation}} on a pre-release product.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/sonic enhancement">sonic enhancement</a></span><!--  Array
(
    [slug] =&gt; sonic enhancement
    [name] =&gt; sonic enhancement
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A way to make clicking an onscreen button more evident by adding sound to simulate a physical effect. More generally, many aspects of an interface can be enhanced by additional audio feedback, such as the crunching paper sound that some systems use when a file is thrown into the (desktop) trash bin. While sometimes provided for aesthetic reasons, this may also add feedback.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/spss">SPSS</a></span><!--  Array
(
    [slug] =&gt; spss
    [name] =&gt; SPSS
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A common software package used for statistical calculations. It has been developed over many years and is now owned by IBM.
    [has_links] =&gt; 1
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/standard deviation">standard deviation (s.d., &sigma;)</a></span><!--  Array
(
    [slug] =&gt; standard deviation
    [name] =&gt; standard deviation (s.d., &amp;sigma;)
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The square root of {{variance}} and probably the most commonly used measure of {{variation}}. The principal alternative is the {{interquartile range}}, which is more robust, but less mathematically tractable than the {{standard deviation}}.
    [has_links] =&gt; 1
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/standard deviation of the population">standard deviation of the population</a></span><!--  Array
(
    [slug] =&gt; standard deviation of the population
    [name] =&gt; standard deviation of the population
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; When you have (or imagine you have) access to the entire {{population}} of interest, perhaps every person in a country or every user of a device, and can calculate the {{standard deviation}} of some facet of them all; that is &amp;radic; ( &amp;;Sigma (x&lt;sub&gt;i&lt;/sub&gt;&ndash;&amp;mu;)&lt;sup&gt;2&lt;/sup&gt;))/N &ndash; where x&lt;sub&gt;i&lt;/sub&gt; are the individual measurements of the facets, &amp;mu; is the arithmetic {{mean}} and N the number of people in the entire {{population}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/standard deviation of the sample">standard deviation of the sample</a></span><!--  Array
(
    [slug] =&gt; standard deviation of the sample
    [name] =&gt; standard deviation of the sample
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The calculation&nbsp;of&nbsp;an {{estimate}}&nbsp;of&nbsp;the {{standard deviation of the population}}&nbsp;that&nbsp;corrects for {{bias}} when using the data in a {{sample}}.  If you use the {{sample}} data to calculate the simple formula for the {{standard deviation}}, this slightly underestimates the true {{standard deviation of the population}}. This is because the {{mean}} also has to be calculated from the {{sample}}. It is possible to {{model the bias}} and then the {{bias}} can be corrected by dividing the sum of squares by N&ndash;1 instead of N. &lt;br&gt;{{standard deviation of the sample}} =  &amp;radic; ( &amp;;Sigma (x&lt;sub&gt;i&lt;/sub&gt;&ndash;&amp;mu;)&lt;sup&gt;2&lt;/sup&gt;))/N&ndash;1
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/standard error">standard error (s.e.)</a></span><!--  Array
(
    [slug] =&gt; standard error
    [name] =&gt; standard error (s.e.)
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The variability of the {{mean}} is called the {{standard error}} of the mean (or just s.e.), and is often drawn as little &lsquo;{{error bars}}&rsquo; on graphs or histograms, to give you some idea of the accuracy of the average measure. If the mean is averaged over a {{sample}} of N independent items, the standard error is &amp;sigma;/&amp;radic;N &ndash; where &amp;sigma; is the {{standard deviation of the sample}} or the {{standard deviation}} of a single item if that is already known.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/starting hypothesis">starting hypothesis</a></span><!--  Array
(
    [slug] =&gt; starting hypothesis
    [name] =&gt; starting hypothesis
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The original hypothesis that an experiment is designed to test. The gold standard statistical process is usually deemed to be that only {{hypotheses}} determined before the study should be admissible. &lt;br&gt;In practice researchers may notice new patterns in their data and then test a new {{hypothesis}} based on this, a process termed {{HARKing}} or {{Hypothesis After the Result is Known}}. This can be dangerous: it is a form of {{cherry picking}} and may lead to spurious, purely random results. One of the reasons for {{pre-registration}} is to discourage this practice. The ideal is that if you do notice such patterns you should perform a fresh study to investigate them, that is a fresh study where the observed pattern from the previous experiment becomes the {{original hypothesis}} of the new one. &lt;br&gt;However, sticking purely to the {{original hypothesis}} is a counsel of perfection that is not always appropriate or achievable, especially when dealing with data such as epidemiological data or historical data that it is impossible to recreate, or where the original data collection is very expensive or difficult to access. In such circumstances the crucial thing is to be aware that you are creating a new {{hypothesis}} and then, where possible, perform a suitable {{multiplicity adjustment}} such as a {{Bonferroni correction}}, and definitely be very clear in any published work that this was &lt;em&gt;not&lt;/em&gt; the {{original hypothesis}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/statistical analysis">statistical analysis</a></span><!--  Array
(
    [slug] =&gt; statistical analysis
    [name] =&gt; statistical analysis
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Analysis of measured or sampled quantitative data, with the goal of obtaining a level of quantified understanding of something in the {{real world}} that is unknown. The three major kinds of statistical analysis are {{hypothesis testing}}, {{confidence intervals}}, and {{Bayesian statistics}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/statistical crisis">statistical crisis</a></span><!--  Array
(
    [slug] =&gt; statistical crisis
    [name] =&gt; statistical crisis
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; There has been substantial publicity surrounding poor statistical practice in published work, casting doubt on how many published results that appear to be {{statistically significant}} are in fact true or {{reproducible}}. This is sometimes called the {{statistical crisis}}. It is treated as a recent phenomenon, focused predominantly on traditional {{hypothesis testing}}, but that largely ignores the fact that potential problems in traditional statistics have been well known for decades and are usually connected to poor use or interpretation, not intrinsic weaknesses in the statistical techniques themselves. Furthermore, the same problems are equally possible with alternatives such as {{Bayesian statistics}}. There is a positive side to this recent publicity in that it has focused many fields of science on improving statistical practice. However, the danger is that researchers either abandon statistical testing where it is in fact appropriate, or adopt alternatives whose weaknesses they understand even less.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/statistical inference">statistical inference</a></span><!--  Array
(
    [slug] =&gt; statistical inference
    [name] =&gt; statistical inference
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This is the formal name for what I have referred to as the {{job of statistics}}, that is attempting to make some forms of {{estimate}} or {{inference}} about the real world based on measured data. It contrasts with {{descriptive statistics}}, which is concerned with various metrics about the data itself. 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/statistical power">statistical power</a></span><!--  Array
(
    [slug] =&gt; statistical power
    [name] =&gt; statistical power
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The probability of having a {{false negative}, i.e. how likely it is that if there is some real effect it shows up in the statistical analysis of the data you have measured. If an experiment has high power the statistics have fewer {{false negatives}}, and real effects are more likely to be seen. The power is a combination of the experiment we have performed and choosing the correct statistical analysis technique for the data. As an example of the latter {{nonparametric statistics}} usually have lower power than {{parametric statistics}} as they make fewer assumptions, so if you know that your data is {{Binomial}} or {{approximately Normal}} it is better to use tests based on these distributions. The {{noise-effect-number triangle}} offers various tactics to increase the power of a study.
    [has_links] =&gt; 1
)
  --></li>
  <li ><span class="name">statistical reasoning</span> - See <span class="has_content"><a href="https://alandix.com/glossary/hcistats/statistical inference">statistical inference</a></span><!--  Array
(
    [slug] =&gt; statistical reasoning
    [name] =&gt; statistical reasoning
    [canonical] =&gt; 
    [see] =&gt; statistical inference
    [description] =&gt; 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/statistical significance">statistical significance</a></span><!--  Array
(
    [slug] =&gt; statistical significance
    [name] =&gt; statistical significance
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In some contexts this just means the {{significance level}} or {{p-value}}, but it can also mean the broader idea or quality of something that uses the process of {{significance testing}}, for example &quot;a test of {{statistical significance}}&quot;.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/steering task">steering task</a></span><!--  Array
(
    [slug] =&gt; steering task
    [name] =&gt; steering task
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This can refer to literal steering of a tractor or car (or simulation of these), or may refer to a form of {{pointing tasks}} where the pointer has to be navigated along a path. In the former case, the steering may be used as a {{distractor task}} for another {{primary task}} that is the real purpose of an experiment.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/student's t distribution">Student's t distribution</a></span><!--  Array
(
    [slug] =&gt; student's t distribution
    [name] =&gt; Student's t distribution
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The {{theoretical distribution}} of the value calculated by the {{Student's t-test}}. Mathematically this is a {{Normal distribution}} divided by the square root of a {{Chi-squared distribution}}. This is because the {{sample variance}} follows the  {{Chi-squared distribution}}. 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/student's t-test">Student's t-test</a></span><!--  Array
(
    [slug] =&gt; student's t-test
    [name] =&gt; Student's t-test
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Student's t-test is a {{statistical test}} that is usually used to compare the {{means}} of two or more groups to see if they are different. For example, comparing the average {{task completion time}} for a group of people using an old version of a system vs that of a second group using a new improved user interface. It is calculated by taking the difference of the means and dividing by an estimate of the {{standard deviation}} of the difference obtained from the {{standard deviation of the sample}} for each of the groups. &lt;br&gt;t = (&amp;mu;&lt;sub&gt;1&lt;/sub&gt; &ndash; &amp;mu;&lt;sub&gt;2&lt;/sub&gt;)  /  &amp;radic;(&amp;sigma;&lt;sub&gt;1&lt;/sub&gt;&lt;sup&gt;2&lt;/sup&gt;/N&lt;sub&gt;1&lt;/sub&gt; &ndash; &amp; sigma;&lt;sub&gt;2&lt;/sub&gt;&lt;sup&gt;2&lt;/sup&gt;/N&lt;sub&gt;2&lt;/sub&gt;) &lt;br&gt;The {{theoretical distribution}} of the value of this t-test follows the {{Student's t distribution}}. The t-test depends on the data being {{approximately Normal}}, but is fairly {{robust}} if this is not perfect. &lt;br&gt;The {{theoretical t distribution}} is a Normal variable divided by the square root of an independent variable with the {{Chi-squared distribution}}. This is because the {{sample variance}} follows the  {{Chi-squared distribution}}. 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/subconscious reactions">subconscious reactions</a></span><!--  Array
(
    [slug] =&gt; subconscious reactions
    [name] =&gt; subconscious reactions
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; One of two systems of thought identified in psychology: subconscious and conscious, sometimes also called System 1 and System 2 respectively. Subconscious reactions are based on semi probabilistic associations built up through experience. Sometimes subconscious reactions lead to an immediate action, such as jumping out of the way if we hear the sound of an approaching car; at other times they are more about a feeling or gut reaction &ndash; we are aware of the reaction, but not the reasons for it. Contrast with {{conscious thinking}}.
    [has_links] =&gt; 1
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/subjects">subjects</a></span><!--  Array
(
    [slug] =&gt; subjects
    [name] =&gt; subjects
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The humans involved in a study. Because the term may sound rather clinical or instrumental, some researchers prefer the terms &lsquo;user&rsquo;, or &lsquo;participant&rsquo; if there is a richer engagement. Both these alternatives have their own connotations, and 'subject' remains the most commonly used in statistics.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/summative evaluation">summative evaluation</a></span><!--  Array
(
    [slug] =&gt; summative evaluation
    [name] =&gt; summative evaluation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Focused on checking that the system works and is good enough, summative evaluation is performed at the end of the {{software development process}} on a pre-release product. It may be related to contractual obligations (whether the system meets agreed specifications) or may compare the system with others. Less formally, summative evaluation may simply be an assessment that enough work has been done based on the cumulative evidence from earlier stages. Contrast with {{formative evaluation}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/survey data">survey data</a></span><!--  Array
(
    [slug] =&gt; survey data
    [name] =&gt; survey data
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Data obtained by asking people questions, often utilising an electronic or paper form or through guided interviews. The volume of survey data can often be far higher than is provided by more intensive methods such as {{direct observation}}, {{usability tests}} or more in depth interviews. Closed questions, whose answers are chosen from a fixed set of options, are useful as they generate easy to analyse statistical data, but they often only tell you &lt;em&gt;what&lt;/em&gt; is happening. Typically you need more {{qualitative data}} to help you understand &lt;em&gt;why&lt;/em&gt; and &lt;em&gt;how&lt;/em&gt;, including the {{motivation}} and {{mechanism}}. Such data may be obtained through {{open questions}} in the survey, or through complementary methods.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/systematic effects">systematic effects</a></span><!--  Array
(
    [slug] =&gt; systematic effects
    [name] =&gt; systematic effects
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Aspects of data collection or analysis that are in some way connected to the thing that you are trying to measure, and hence interfere with analysis or lead to misleading effects or {{bias}}. There are many random or uncontrolled effects that influence the data you obtain, but so long as these are unrelated to the thing you are trying to measure ({{independent}}), they can be treated as {{random noise}}. Though potentially adding {{variability}}, they will eventually average out given a large enough sample. In contrast, systematic effects may cause {{bias}}. &lt;br&gt;For example, suppose you are using a laser to measure the heights of people coming and going from two different tourist attractions in London. As they will be wearing shoes and potentially hats, this will not be a true measure of height, but for the purposes of comparison between, say, Buckingham Palace and The Houses of Parliament, it is probably fine as there is no reason to suppose visitors at either wear substantially different attire. If, however, this method were used to compare the tourists at Buckingham Palace with the guards, since the guards would be likely to be wearing tall Bearskin hats, the size of hat would be a systematic effect, leading to {{bias}} and overestimating the height of guards compared with tourists. &lt;br&gt;Note that due to the nature of randomness apparently systematic effects sometimes happen by chance.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name">t-test</span> - See <span class="has_content"><a href="https://alandix.com/glossary/hcistats/student's t-test">Student's t-test</a></span><!--  Array
(
    [slug] =&gt; t-test
    [name] =&gt; t-test
    [canonical] =&gt; 
    [see] =&gt; student's t-test
    [description] =&gt; 
    [has_links] =&gt; 1
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/tail">tail</a></span><!--  Array
(
    [slug] =&gt; tail
    [name] =&gt; tail
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Tail has two meanings in statistics. One is the long spread of low frequency values in a {{long-tail distribution}}, which includes very extreme values at the end(s) of the scale. The other is the use of the term to indicate in which directions the data is of interest &ndash; one, or both; see {{one-tailed test}} and {{two-tailed test}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/tail heavy">tail heavy</a></span><!--  Array
(
    [slug] =&gt; tail heavy
    [name] =&gt; tail heavy
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; When the probability (or observed frequency) of data items far from the central value is large. This can be true of a {{symmetric distribution}} such as the {{Cauchy distribution}}, but is more often associated with asymmetric {{long-tail distributions}} especially the {{power-law distribution}} found in areas such as {{network data}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/target size">target size</a></span><!--  Array
(
    [slug] =&gt; target size
    [name] =&gt; target size
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In a {{pointing task}} this is the size of the area you want to point at, such as a button or icon. It is used in {{Fitts' Law}} to predict the time it takes to locate the target.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/targeted tasks">targeted tasks</a></span><!--  Array
(
    [slug] =&gt; targeted tasks
    [name] =&gt; targeted tasks
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Tasks designed to deliberately  expose the effects of an intervention by focusing on the aspects of the system that are being tested, minimising interaction with other elements. For example, if you have added a feature to help users keep track of what they are doing, a {{targeted task}} could include a lot of interruptions. Targeted tasks increase {{effect}} in the {{noise-effect-number triangle}} and so increase {{statistical power}}. However, targeting can be misleading if you are not careful, as the {{targeted task}} may not be representative of the full range of tasks that are likely to be carried out; that is {{generalisation}} can be difficult.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/targeted user group">targeted user group</a></span><!--  Array
(
    [slug] =&gt; targeted user group
    [name] =&gt; targeted user group
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A group for whom we believe we will see the maximum effect is deliberately chosen. For example, a new menu system would be tested more effectively for short-term memory requirements on an elderly user group who are more likely to have some short-term memory loss, rather than on a younger user group. Targeted user groups increase {{effect}} in the {{noise-effect-number triangle}} and so increase {{statistical power}}. However, targeting can be misleading if you are not careful as the {{targeted user group}} may not be representative of the full range of users who are likely to use the system, therefore {{generalisation}} can be difficult.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/task completion time">task completion time</a></span><!--  Array
(
    [slug] =&gt; task completion time
    [name] =&gt; task completion time
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The time taken to complete a specified task is often used as a measure when comparing two systems. This may be a high-level task, such as finding a product on an online shopping site; or a low-level task, such as a {{pointing task}} modelled by  {{Fitts' Law}}. 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/tasks">tasks</a></span><!--  Array
(
    [slug] =&gt; tasks
    [name] =&gt; tasks
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Experimental or study tasks are the actions that users are specifically asked to perform. Measurements such as {{task completion time}} can be observed and recorded. Tasks can be manipulated to gain {{statistical power}} using the {{noise-effect-number triangle}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/temporal data">temporal data</a></span><!--  Array
(
    [slug] =&gt; temporal data
    [name] =&gt; temporal data
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Data recorded over a period of time, where the time at which events occur is a critical part of the data. This can include very long-term phenomena such as the level of usage of a web site, shorter term such as the trace of user interactions while performing a task, or very fine scale such as some biological measurements. There are special statistical techniques for modelling {{time series}} data.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/the job of statistics">the job of statistics</a></span><!--  Array
(
    [slug] =&gt; the job of statistics
    [name] =&gt; the job of statistics
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; We may be interested in some facet of the {{real world}}, for example how much faster it is to use one system compared with another. However, we typically only have some sort of partial or noisy measured data. {{Probability theory}} can help us to assess the {{likelihood}} of the {{measurement}} given the state of the world. However, we do not know the real state of the world, only the measurement. The {{job of statistics}} is to reverse this, to create some sort of {{estimate}} of, or knowledge about, the {{real world}} from the {{measurement}}. Because the measured data is subject to {{random}} effects, the knowledge we obtain from statistics will always be {{uncertain}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/theoretical distribution">theoretical distribution</a></span><!--  Array
(
    [slug] =&gt; theoretical distribution
    [name] =&gt; theoretical distribution
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A mathematical calculation of the probability of an event in the real world; examples are the {{Normal distribution}} (for heights) and the {{Binomial distribution}} (for coin tosses). Contrast with {{empirical distribution}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/theoretical understanding">theoretical understanding</a></span><!--  Array
(
    [slug] =&gt; theoretical understanding
    [name] =&gt; theoretical understanding
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; In analysis of {{qualitative data}}, theoretical understanding provides additional information; for example, cognitive and psychological understanding is required to build a model of how the user is behaving and why. Theoretical analysis may be based on past data and experiments, but also involves some level of analysis or reasoning either by the analyst, or drawing on previously published literature.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/threshold effect">threshold effect</a></span><!--  Array
(
    [slug] =&gt; threshold effect
    [name] =&gt; threshold effect
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A limiting condition that can cause {{nonlinearity}}; for example, plants growing near an electric fence may be electrocuted when they reach a certain height, leading to a chopped off {{Normal distribution}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/traditional statistics">traditional statistics</a></span><!--  Array
(
    [slug] =&gt; traditional statistics
    [name] =&gt; traditional statistics
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The statistical techniques of {{hypothesis testing}} and {{confidence intervals}}, which have the same underlying mathematics and share a common philosophical stance. Contrast with {{Bayesian statistics}}.

    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/transforming data">transforming data</a></span><!--  Array
(
    [slug] =&gt; transforming data
    [name] =&gt; transforming data
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; When you apply some sort of calculation to data items in order to transform them into a form more suitable for analysis. For example, if you have a 5 point {{Likert scale}} you might transform this into {{binary data} by mapping the top two values (4 and 5) into 'satisfied' and the rest (1, 2, and 3) into 'not satisfied'. {{Temporal data}} is often asymmetric and furthermore larger items typically have greater {{variance}}. However, if you take the {{logarithm}} of the times, the resulting data is often {{approximately Normal}}. This last transformation is so common that data such as {{reaction times}} are termed {{log-Normal}} data.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/transparent statistics">Transparent Statistics</a></span><!--  Array
(
    [slug] =&gt; transparent statistics
    [name] =&gt; Transparent Statistics
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; At the {{CHI conference}} a series of workshops on &quot;Transparent Statistics in Human&ndash;Computer Interaction&quot; is focused on creating guidance on best practice reporting of statistics.
    [has_links] =&gt; 1
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/treemap">TreeMap</a></span><!--  Array
(
    [slug] =&gt; treemap
    [name] =&gt; TreeMap
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The TreeMap is a {{constant area visualisation}} invented by Ben Shneiderman for hierarchical data with some form of numerical attribute, for example the sizes of files and folders on a computer disk. The most basic form starts with a rectangular area and then recursively divides the area alternatively horizontally using the total values for each level.
    [has_links] =&gt; 1
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/two-tailed test">two-tailed test</a></span><!--  Array
(
    [slug] =&gt; two-tailed test
    [name] =&gt; two-tailed test
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A test where data is compared in two directions. For example, assume you are choosing which of two systems, A and B, to purchase for you company and want to check if either is measurably better than the other. The results matter in both directions: whether A is better than B or whether B is better than A. If comparing {{response times}} your test has to look both at when System A is significantly faster &lt;em&gt;and&lt;/em&gt; at when it is significantly slower. Contrast with {{one-tailed test}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/type i error">Type I error</a></span><!--  Array
(
    [slug] =&gt; type i error
    [name] =&gt; Type I error
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This is a term from {{hypothesis testing}} where you wrongly believe there to be an effect when there is none; that is a {{false positive result}}. The {{significance level}} you choose helps to control for this. There is only a 1 in 20 chance that you hit a {{p-value}} of 5% by chance when there is no effect; that is, if you choose a significance level of 5%, the probability of a {{Type I error}} is 0.05 (1 in 20). Similarly if you choose a 1% {{significance level}} then the probability of the Type I error is only 0.01 (in in a hundred).
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/type ii error">Type II error</a></span><!--  Array
(
    [slug] =&gt; type ii error
    [name] =&gt; Type II error
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This is a term from {{hypothesis testing}} when you fail to detect a real effect. This typically happens because the {{statistical power}} of your experiment or analysis method is too weak. Often researchers and paper reviewers worry less about a {{Type II error}} than a {{Type I error}}. This is because if you think something is true wrongly and then base future science on the false belief everything you build on it will potentially also be fallacious. However, {{Type II errors}} are also critical. A certain proportion of {{statistically significant}} results will occur by chance, so if the {{statistical power}} of experiments is too weak and there are too many {{Type II errors}} there will be too few true positive results published compared with the {{false negatives}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/typical user">typical user</a></span><!--  Array
(
    [slug] =&gt; typical user
    [name] =&gt; typical user
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; There is, of course, no such thing as a typical user &ndash; everyone is different. However, one does have an idea of a central user, for instance a nineteen-year-old student for a University virtual learning system. Furthermore, when doing any form of statistics based on a user study the user you are trying to understand is not so much the specific people taking part in your study, or even the collection of all current users, but some kind of idealised conception of a potential user.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/uk income distribution">UK income distribution</a></span><!--  Array
(
    [slug] =&gt; uk income distribution
    [name] =&gt; UK income distribution
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Graphs of income and wealth in most countries form a {{long-tail distribution}}, that is there are a large number of people with low or middle sized incomes and then a small number of people with very large incomes. The upper part of the graph typically follows closely to a {{power law}}, but at the lower end the effect of welfare benefits mean that there are few people with absolutely tiny incomes. The result for the UK is a highly {{asymmetric distribution}} that starts with counts close to zero (few people with no income), rises towards a {{mode}} at around &pound;300 per week, and then drops rapidly, with only about 5% of people earning more than a thousand pounds per week. As with most asymmetric distributions, the {{median}} income (~&pound;430 pw) is lower than the {{mean}} (~&pound;530 pw). Because of the distorting effect of the long tail on the arithmetic {{mean}}, when newspapers and government statistics quote the {{average}} income they are usually referring to the {{median}} income.
    [has_links] =&gt; 1
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/unbiased">unbiased</a></span><!--  Array
(
    [slug] =&gt; unbiased
    [name] =&gt; unbiased
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Informally this refers to personal fairness, but within statistics it also has a more formal meaning. We say an {{estimate}} is {{unbiased}} if the long term or theoretical average of the estimate is the true value of the {{real world}} thing it is trying to measure. For example, if we want to know the average height of someone in Malaysia, we might randomly sample 1000 people from Malaysia and then take the arithmetic {{mean}} of the measured heights. It turns out that the {{sample mean}} is an {{unbiased estimator}} of the {{population mean}}. For any given sample of 1000 people this might be slightly more, or slightly less than the true national average, but if one were to take many such samples the average value of the sample average would be the true average!
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/unbounded data">unbounded data</a></span><!--  Array
(
    [slug] =&gt; unbounded data
    [name] =&gt; unbounded data
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Data can be potentially unbounded, if it is not {{finite data}}. Unbounded data may be unbounded in one or both directions. Examples of the former are: (coin tosses, {{discrete data}}) &ndash; the number of heads thrown before the first tail is thrown; (waiting for a bus, {{continuous data}}) &ndash; the length of time before the bus comes. These are both bounded below but unbounded above. An example of data that is unbounded in either direction is the difference between heights, which can be arbitrarily positive or negative. 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/unbounded tail">unbounded tail</a></span><!--  Array
(
    [slug] =&gt; unbounded tail
    [name] =&gt; unbounded tail
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; When one {{tail}} of a distribution is {{unbounded}}, that is it has no theoretical maximum (or minimum). There are distributions where the tail can be quite long &ndash; there may be a small probability of very large or very small values, but it is still finite: for example, if you look at the number of double sixes when rolling two dice 100 times, the average is around 3, but there is a very small chance of there being a full 100 double sixes. However, there can never be more than 100. In the end the tail is {{bounded}}. In contrast, consider just starting tossing a coin until you reach the first tail and count how many heads you get before the first tail &ndash; that is a {{Negative Binomial distribution}}. Here the average value is just 1, but there is no maximum, you could, in principle, go on tossing heads for ever. That is the {{Negative Binomial distribution}} has an {{unbounded tail}}.
    [has_links] =&gt; 
)
  --></li>
  <li ><span class="name">unbounded values</span> - See <span class="has_content"><a href="https://alandix.com/glossary/hcistats/unbounded data">unbounded data</a></span><!--  Array
(
    [slug] =&gt; unbounded values
    [name] =&gt; unbounded values
    [canonical] =&gt; 
    [see] =&gt; unbounded data
    [description] =&gt; 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/uncertain knowledge">uncertain knowledge</a></span><!--  Array
(
    [slug] =&gt; uncertain knowledge
    [name] =&gt; uncertain knowledge
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Although statistics works with actual measurements, it can only discover information about probable outcomes, never absolute certainty. See also {{likelihood}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/uncontrolled factors">uncontrolled factors</a></span><!--  Array
(
    [slug] =&gt; uncontrolled factors
    [name] =&gt; uncontrolled factors
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Factors that are unrelated to the thing being measured, but that affect the result. For example, if people are asked to perform a website navigation task, and some have better web skills than others, or better spatial ability. Although such effects can sometimes be measured they must often be treated as effectively random.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/uncontrolled variation">uncontrolled variation</a></span><!--  Array
(
    [slug] =&gt; uncontrolled variation
    [name] =&gt; uncontrolled variation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; These are various factors that we are unable to control, for example the weather on the day of an outdoor user study or the level of expertise of test subjects chosen at random. So long as there are no {{systematic effects}} these can be treated as part of {{random noise}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/uniform distribution">uniform distribution</a></span><!--  Array
(
    [slug] =&gt; uniform distribution
    [name] =&gt; uniform distribution
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A {{uniform distribution}}  is where the {{probability}} of every outcome is the same. 

For {{discrete data}}, this is a probability of 1/N for each value, where N is the number of values.  In the case of data 1...N  the {{mean}} is (N+1)/2 and the {{variance}} is (N+1)(N-1)/12 

For {{bounded}} {{continuous data}} the probability is 1/W where W is the width of the data..  In the case of continuous data over [0,W] the {{mean}} is W/2 and the {{variance}} is W&lt;sup&gt;2&lt;/sup&gt;/12 
&lt;br&gt;&lt;br&gt;
Note that in the continuous case, the definition of 'uniform' depends on the form of the parameter.  For example, if dealing with sound waves a uniform distrubution over the wavelength is a different {{probability distribution}} to one that is uniform over the frequency.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/uniform prior">uniform prior</a></span><!--  Array
(
    [slug] =&gt; uniform prior
    [name] =&gt; uniform prior
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A {{prior distribution}} where the {{probability}} of every outcome is the same. For {{discrete data}}, this is a probability of 1/N for each value, where N is the number of values. If the data is {{unbounded}} there is no meaningful {{uniform prior}}, and the {{Cauchy distribution}} is often used as a best alternative, but for this you need to decide a midpoint and spread, either of which may introduce {{unintentional bias}}. You also have to be careful about the perspective from which data is measured. For example, with sound do you choose a prior that is uniform over power or decibels? Because the latter is the {{logarithm}} of the former, the uniform prior of one is not uniform for the other.  This is particularly important in {{Bayesian statistics}} as the choice of prior can make a big difference to the results.

See also {{uniform distribution}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/uniformity fallacy">uniformity fallacy</a></span><!--  Array
(
    [slug] =&gt; uniformity fallacy
    [name] =&gt; uniformity fallacy
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The belief or gut feeling that, overall, results will even out, which leads people to reject clumped results as artificial or to suppose that a run of bad luck will be followed by a lucky break.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/unintentional bias">unintentional bias</a></span><!--  Array
(
    [slug] =&gt; unintentional bias
    [name] =&gt; unintentional bias
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Sometimes an advert or a political message may deliberately (or intentionally) put some sort of spin on data and mislead those seeing it. As researchers and professionals, we would all, I hope,  try to avoid this deliberate bias, but we may still have {{unintentional bias}} due to the choice of {{sample}} or other aspects of our study methods. For example, if you interview people in an Apple store and ask them for their favourite phone model, the result may be somewhat skewed! Of course, unintentional bias is not always so obvious; for example, if you interview members of a particular social media network, it may include more younger people, or more well-off people, and this might have an impact on music taste or even favourite food.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/unrepeatable event">unrepeatable event</a></span><!--  Array
(
    [slug] =&gt; unrepeatable event
    [name] =&gt; unrepeatable event
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Sometimes we can repeat an identical event multiple times, for example repeatedly tossing a coin. However, some events are unrepeatable, for example, the weather on 3rd January 2020. Crucially once such an event has occurred it is, by definition, not {{ontologically random}} &ndash; its state is already determined. However, it may still be {{epistemologically random}}, unknown to you, and hence you may sometimes still treat it as having probabilistic or statistical properties.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/upper bound">upper bound</a></span><!--  Array
(
    [slug] =&gt; upper bound
    [name] =&gt; upper bound
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The largest possible value that a variable or dataset can have. For example, when rolling a  die, the upper bound is 6. In contrast, some data is {{unbounded above}}, that is there is no meaningful absolute maximum.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/upper quartile">upper quartile</a></span><!--  Array
(
    [slug] =&gt; upper quartile
    [name] =&gt; upper quartile
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The value such that 75% (three quarters) of the data values are less and 25% more. It is one of three {{quartiles}}, the others being the {{median}} (50% point) and {{lower quartile}} (25% point).
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/usability evaluation">usability evaluation</a></span><!--  Array
(
    [slug] =&gt; usability evaluation
    [name] =&gt; usability evaluation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This is about evaluating how usable a system is; the term can be read quite broadly, or may be distinguished from evaluating {{user experience}}.There are many ways of evaluating usability: {{expert evaluation}} techniques are useful for early design or when users are difficult to contact, but typically the gold standard is expected to use real users. When evaluating with real users there are still many options from {{controlled laboratory experiments}} to {{in the wild}} studies.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/user experience">user experience</a></span><!--  Array
(
    [slug] =&gt; user experience
    [name] =&gt; user experience
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Aspects of a system that affect the user's emotional state; that is creating a (usually) positive experience of using the system.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/user experience design">user experience design</a></span><!--  Array
(
    [slug] =&gt; user experience design
    [name] =&gt; user experience design
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This refers to the parts of the design process and the methods used to ensure a desired {{user experience}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/user experience studies">user experience studies</a></span><!--  Array
(
    [slug] =&gt; user experience studies
    [name] =&gt; user experience studies
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; These are studies that attempt to {{evaluate}} or otherwise understand the {{user experience}} of a study. It is possible to measure {{objective}} aspects such as heart rate, but the emotional and personal nature of {{user experience}} mean that more {{qualitative methods}} are often used, or various forms of subjective questionnaire. 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/user interface">user interface</a></span><!--  Array
(
    [slug] =&gt; user interface
    [name] =&gt; user interface
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The part of a system that the user interacts with. Typically this is some form of screen-based interface with buttons, menus, etc., such as an app or desktop graphical user interface. However, interaction may be through alternatives such as speech, gesture or physical things in the environment that are sensed by the computer.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/user interface development">user interface development</a></span><!--  Array
(
    [slug] =&gt; user interface development
    [name] =&gt; user interface development
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The process of creating the user-facing aspects of a system, from initial scoping and design through to coding and deployment. Typically {{iterative development}} is used where there are cycles where some form of {{prototype}} or {{design sketch}} is evaluated in a {{user study}} or {{expert evaluation}}. The design is then altered in the light of the {{formative evaluation}}, leading into another cycle.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/user interface properties">user interface properties</a></span><!--  Array
(
    [slug] =&gt; user interface properties
    [name] =&gt; user interface properties
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This usually refers to specific generic properties that influence the {{usability}} or {{user experience}} of a system in some way. For example, it is usually the case that interactions are more easily learnt if the interface is {{consistent}}. A {{user interface property}} can be a form of {{generative artefact}} when one deliberately tries to create a system that adheres to the principle, and as such can be hard to {{evaluate}}.
    [has_links] =&gt; 
)
  --></li>
  <li ><span class="name">user interface testing</span> - See <span class="has_content"><a href="https://alandix.com/glossary/hcistats/usability evaluation">usability evaluation</a></span><!--  Array
(
    [slug] =&gt; user interface testing
    [name] =&gt; user interface testing
    [canonical] =&gt; 
    [see] =&gt; usability evaluation
    [description] =&gt; 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/user studies">user studies</a></span><!--  Array
(
    [slug] =&gt; user studies
    [name] =&gt; user studies
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; When a system or context is investigated by using real users. {{User studies}} are often regarded as being the gold standard form of {{evaluation}}, but may not always be possible, for example, if the intended users are difficult to contact or a vulnerable group. 
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/validation">validation</a></span><!--  Array
(
    [slug] =&gt; validation
    [name] =&gt; validation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; The general process of establishing whether something (an interaction technique, system, theory, design principle, toolkit) is in some sense 'valid' &ndash; correct, true, usable. Within HCI there is often a focus on {{empirical evaluation}} of a system, theory or other artefact after it has been created. However, this is effectively impossible for {{generative artefacts}} (things that create other things) such as toolkits, methodologies or design principles, and then other methods, such as {{justification}} based on reasoning and past literature, become more important.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/variability">variability</a></span><!--  Array
(
    [slug] =&gt; variability
    [name] =&gt; variability
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; As a statistical term, this is pretty close to its everyday meaning: how much data items vary from each other &ndash; are they all quite close, or do they vary over a wide range of values? There are various measures of {{variability}}, from the absolute range ({{maximum}}&ndash;{{minimum}}) to the {{interquartile range}}, or the ubiquitous {{standard deviation}}. For a {{theoretical distribution}} you can think of this as how widely spread the graph is, or how variable you would expect data that conforms to the distribution to be.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/variance">variance</a></span><!--  Array
(
    [slug] =&gt; variance
    [name] =&gt; variance
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; A measure of the {{variation}} of {{empirical data}} or of a {{theoretical distribution}}. It is calculated by taking the average of the square of the difference between the data items and their {{mean}}, that is the average square of the {{residual}}. &lt;br&gt;{{variance}}  =   ( &amp;Sigma; (x&lt;sub&gt;i &lt;/sub&gt;&ndash; &amp;mu;)&lt;sup&gt;2&lt;/sup&gt; ) / N  &ndash; where x&lt;sub&gt;i &lt;/sub&gt; is the i-th data item and N is the number of items &lt;br&gt;Often the mean of {{empirical data}} is {{estimated}} based on the data in a {{sample}}; in such cases the simple formula slightly underestimates the real value of the {{variance}} and instead the {{sum of squares}} is divided by N&ndash;1. &lt;br&gt;The {{variance}} is a square measure, which is not easy to interpret (for example the variance of people's height is in square metres). So, more often the {{standard deviation}} is given, which is the square root of the variance. &lt;br&gt; The variance is also typically less {{robust}} than other measures of {{variation}} such as {{interquartile range}}. However, variances have nice mathematical properties &ndash; they 'add up' &ndash; and so are heavily used.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/visualisation">visualisation</a></span><!--  Array
(
    [slug] =&gt; visualisation
    [name] =&gt; visualisation
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Methods and systems to make it easier to see trends or other patterns in data or derived results. This varies from simple {{scatter plots}} to more complex information visualisations such as the {{TreeMap}} or {{parallel coordinates}}. Where {{visualisation}} is combined with machine learning or other advanced data analysis techniques, it is referred to as {{visual analytics}}. Despite the 'visual' word in its name, {{visualisation}} is also sometimes used to refer to ways to communicate information through other senses such as auditory versions of graphs.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/weird people">WEIRD people</a></span><!--  Array
(
    [slug] =&gt; weird people
    [name] =&gt; WEIRD people
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; Western, Educated, Industrialized, Rich, and Democratic &ndash; Henrich et al. pointed out that the vast majority of psychological studies use a very specific type of subject, not at all typical of the global population. IIndeed for many experiments the main subjects are psychology students!  In a {{meta-analysis}} where studies had also been performed in cultural settings which were not WEIRD, they found that cognitive and perceptual results that had been thought to be fundamental to human nature, for example the {{muller-lyer illusion}}, turn out to vary between cultures.
    [has_links] =&gt; 1
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/within-groups studies">within-groups studies</a></span><!--  Array
(
    [slug] =&gt; within-groups studies
    [name] =&gt; within-groups studies
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This is similar to a {{within-subjects design}} except that a group of people need to work together in each condition, for example on some form of shared software or {{CSCW}} system. If anything, group dynamics make the differences between groups larger than {{individual differences}}, so there is even greater advantage in terms of {{statistical power}}. Unfortunately the interference effects between conditions are also greater and it is harder to match groups in the way that you can individually {{match subjects}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/within-subjects experiment">within-subjects experiment</a></span><!--  Array
(
    [slug] =&gt; within-subjects experiment
    [name] =&gt; within-subjects experiment
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; When you ask the same person to take part in two or more conditions of an experiment, in contrast to {{between-subjects design}} where each person takes part in a single condition. For example, if you are comparing two user interfaces, you might ask each subject to perform tasks on both. The advantage of this is that if an individual is particularly skilled or unskilled at the tasks or has simply had a bad day, this cancels out between the two conditions &ndash; that is it cancels out {{individual differences}}. Crucially this increases the {{statistical power}} of your study by {{reducing noise}}. The downside is that there may be {{order effects}} or other forms of {{interference}} between the conditions; for example, a positive {{learning effect}} where the fact that someone has done a task with the first system makes it easier to perform a similar task on the second system. There are ways to ameliorate these effects, for example to {{randomise the order of conditions}}. If a {{within-subjects design}} is not possible and you need to use a {{between-subjects design}}, you might try {{matching subjects}}.
    [has_links] =&gt; 
)
  --></li>
  <li class="has_content"><span class="name"><a href="https://alandix.com/glossary/hcistats/wizard of oz prototyping">Wizard of Oz prototyping</a></span><!--  Array
(
    [slug] =&gt; wizard of oz prototyping
    [name] =&gt; Wizard of Oz prototyping
    [canonical] =&gt; 
    [see] =&gt; 
    [description] =&gt; This is a form of {{prototyping}} where some aspect of the interface that would be expected to be performed by a computer is instead performed by a human researcher, usually secretly. This can be used to try out ideas early before they are fully implemented, or even to explore technology that has not yet been fully developed.
    [has_links] =&gt; 
)
  --></li>
</ul>
</div>
</body>
</html>