statistical bias

Terms from Artificial Intelligence: humans at the heart of algorithms

The glossary is being gradually proof checked, but currently has many typos and misspellings.

In statistics the term 'bias' has a precise mathematical definition. Given an underlying quantity X we want to know (say the average height of adults in a country), we will use some form of estimator E (say the average height of a random sample of a hundred adults). We say that the estimator E is unbiased if the 'expected' value of E is the correct value of X. Here 'expected' is in the statistical sense of the mean of E averaged over all possible samples. Any particular sample of adults might be unusually tall or short, but the average of the average hight of all such samples is the true value. In contrast if expected value of E is not correct, th estimator is biased.
Note that this is related to, but not the same as the day-to-day meaning of bias, which may imply conscious or unconscious prejudice, or judging individuals based on the characteristics of a group of which they are part such as gender, ethinicity or social class.

Used in Chap. 20: page 317