An evaluation of different approaches for behavior genetic analyses with psychiatric symptom scores

We used a simulation study to evaluate six approaches for behavior genetic analyses of psychiatric symptom scores. For the selection of the correct model, the best results were obtained with approaches using transformed scores in combination with a procedure involving p-values. With normalizing transformations, the chi 2 test statistic gave a reasonable impression of the overall fit of the model but was less accurate when used as a difference test. The asymptotic distribution free estimation methods yielded chi 2s that were much too large. All data analysis techniques yielded substantially biased parameter estimates. The most biased results were obtained with normalizing transformations. The least biased results were obtained with tobit correlations, but because of its large standard errors the most precise estimates were obtained with polychoric correlations and optimal scale scores. An empirical study showed that a recognition of the role of methodological factors was helpful to understand part of the differences between assessment instruments, raters, and data analysis techniques that were found in the real data.