Constant sex difference across populations in liability of nonmetric traits

Abstract

Studies have revealed the existence of statistically significant sex differences in the frequency of nonmetric traits, but no agreement seems to exist about their variability among populations. This problem was examined using the multifactorial threshold model. Considering the assumption of additive effects of factors on the liability and the nature of effect of sex difference on the development of nonmetric traits, it would be reasonable to assume that the sex difference in the mean of liability is constant across populations. This hypothesis was tested and the magnitude of sex difference was examined using the world-wide dataset collected by Ossenberg and the dual-liability threshold model formulated by the author with a modification to accommodate side difference in the probability of trait occurrence. The data were divided into 16 samples regarded as randomly sampled from regional populations. The data of 31 bilateral traits were analyzed using maximum likelihood estimation procedures. After confirming the homogeneity of the variance of liability between sexes and across populations, the homogeneity and significance of sex difference in the mean of liability were tested. The results indicate the homogeneity of sex difference across populations. The assumed constant sex difference was statistically significant in 17 traits at the 1% level, and its magnitude exceeded half the averaged distance between eight groups of populations in 12 traits. Population comparisons without distinguishing sex are justifiable if they use the traits with enough weak sex difference in comparison with population differences. Since the sex difference has proved to be basically constant across populations, the estimates of the assumed constant sex difference reported in this study would provide references for selecting traits appropriate for each comparison. The Breslow–Day test of homogeneity of sex difference indicated the inapplicability of the genotype model to the data, supporting Ossenberg’s proposal for the use of side counts.