Why Pearson's r is Not a Good Similarity Coefficient for Comparing Collections
George L. Cowgill · American Antiquity · 1990
Pearson's coefficient of linear correlation, r, sometimes is used to express the similarity between two archaeological collections when each collection is characterized by the percents of the artifact types it includes. Examples show that r can be extremely misleading when used in this way, and patterns of similarity among a group of collections can be wildly misrepresented. The Brainerd–Robinson statistic is one example of a similarity coefficient that is much better than r for comparing collections.