How Many Variables are Too Few? Effect of Sample Size in STET, a Method to Test Conspecificity for Pairs of Unknown Species.

Abstract

One of the questions most often asked in paleoanthropology is whether the amount of variation in a fossil sample is too much to be from a single species. STET (STandard Error Test) uses standard error of the coefficient from a linear regression model relating a pair of specimens to ask if variation in a fossil sample needs to be explained as the presence of multiple species. Previous studies have used this method to test the null hypothesis of no differ-ence in various hominid samples and showed that the variation is not too great to reject the hypothesis of single species. In this paper, properties and limits of STET are explored using skeletal data of known sex and species.Data include 53 cranial and postcranial measurements of modern humans (n=87) and chimpanzees (n=43). STET values are calculated for all possible interspecific (n=3,741) and intraspecific pairs (n=4,644) to generate distribu-tions of STET. Results show that interspecific STET values are always greater than intraspecific STET values when all 53 variables are used. When variable numbers are arbitrarily decreased by random sampling, STET becomes less effective. Random sampling of variables is repeated 1,000 times to assess the effectiveness of STET. The amount of over-lap between interspecific and intraspecific STET values is used as error rates from 0.01 to 0.10. Minimum number of variables necessary for STET to be effective ranges from 30 (0.10 error rate) to 48 (0.01 error rate).