Errors of inference in the detection of major gene effects on psychological test scores.

Computer simulation methods were employed to generate abilities of 10 sets of 250 nuclear families, each comprising a pair of randomly mated parents and two children. It was assumed that the distribution of abilities in the population was normal and caused entirely by additive polygenic effects. A simulated psychological test was administered to each sample to generate test scores for each subject. A different test, consisting of 40 items of varying difficulty and discriminating power, was used in each sample. The "mixed model," specifying a single major gene with polygenic and environmental background variation, was tested for each data set. Likelihood ratios were computed to test for the contribution of a major locus and its conformity to Mendelian segregation. Only one out of 10 samples was consistent with pure multifactorial inheritance. Of the remaining nine samples, four showed non-Mendelian segregation and five were consistent with current statistical criteria for establishing the contribution of a major gene to variation in psychological test scores. This high frequency of false conclusions suggests that the naïve application of such methods to behavioral data is often likely to be misleading. Raw test scores alone are not sufficient to test the mixed model. The development of tractable models for behavioral traits requires the responses of subjects to individual items.