The fourth advantage of IRT has to do with floor and ceiling effects. A problem of analysing sum scores that represent indices of psychopathology is that these scores show a skewed distribution in the general population (Van den Oord et al. 2003; Derks et al. 2004). These skewed distributions result from the fact that many behavioural phenotypes are assessed using questions that relate to symptoms that are relatively rare in the population. These distributional violations may have important implications for the inference regarding relative variance components when analysing sum scores (Derks et al. 2004). In an IRT framework one is essentially free to specify the distribution of the latent trait (in some cases, it can even be estimated). In most cases, with polygenic traits, a normal distribution seems the most reasonable alternative (a mixture approach may be more suitable for traits with only a few large QTL effects). When in turn the variance of the normally distributed latent trait is decomposed into genetic and non-genetic variance, the inference is unbiased if the assumptions of the model are correct.
