SEARCH

SEARCH BY CITATION

Abstract

During much of his academic career, Raymond Pearl advocated the employment of quantitative and statistical methods in biology and applied them in his own voluminous research output, much of it in various aspects of human biology. If he were to return today, he would be pleased to note the fruits of his advocacy—the widespread quantification of biological and anthropological research. He would undoubtedly be impressed with the tremendous progress in statistical analysis in the 63 years since he ceased his work and with that amazing tool, the personal computer. But he would also note the dangers inherent in applying these powerful methods without sufficient regard for the experimental or observational designs they require and for their basic assumptions. Among the errors Pearl would note are: 1) the use of canned computer programs in ignorance of or inattention to their implied assumptions; 2) testing many contrasts in one dataset without making allowances for the lack of independence of these tests; 3) ignoring the effects of lack of independence (autocorrelation) when working with spatially, temporally, or phylogenetically related data; 4) presenting distorted ordination plots; and 5) using an unrealistic null distribution in computation or simulation, as when employing sample sizes too small to test the hypotheses the investigator is interested in. All of these errors are common not only in publications on physical anthropology and human biology, but also in the wider biological literature. Am. J. Hum. Biol. 16:113–124, 2004. © 2004 Wiley-Liss, Inc.