Actually, when the raw item data follow the scalogram pattern, the true correlations and the corresponding variance components will be recovered when applying a threshold model (Lynch and Walsh 1998). This is also true when the data follow a scalogram pattern but the items are not evenly scattered across the scale and the sum score distribution is skewed: applying a threshold model will recover the true correlations (cf. Derks et al. 2004). However, when applying an ordinary variance component analysis, ignoring its non-normality will yield biased estimates, underestimating the effects of shared environment and overestimating the effects of dominance (cf. Derks et al. 2004). This because when the items are not evenly scattered and the sum score distribution is skewed, the attenuation effect is no longer proportional to the true correlations (Fig. 1D): small correlations are more severely attenuated than large correlations. In case the true DZ correlation equals half the true MZ correlation, DZ:MZ = 1:2, the correlations of the sum scores will show a smaller ratio, DZ:MZ < 1:2, usually an indication of dominance genetic effects or epistasis.
