power-law distribution

Terms from Statistics for HCI: Making Sense of Quantitative Data

A form of long-tail distribution where the frequency is inversely proportional to some power: that is freq(n) = 1/nλ. 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 λ 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 must not use tests designed for the Normal distribution, even when the sample size is very large.

Defined on page 51

Used on pages 50, 51, 53, 93, 108

Also known as power law, power-law data