Sampling Properties of Gower's General Coefficient of Similarity

Abstract

The sampling properties of Gower's general coefficient of similarity were investigated. Artificial communities were created and the effects of sample size and species number on the sampling distribution of the coefficient were studied using a computer simulation technique. It was found that the index tends to overestimate low similarity and to underestimate high similarity. The bias of the index increased with species number, while the standard deviation was independent of species number. Increasing sample size increased the accuracy and the precision of the index in estimating community similarity.