Our simulations showed the dramatic extent of the attenuation effect and the bias in estimating variance components due to imperfect measurement. Particularly when sum score distributions are skewed, underestimation of shared environmental effects and overestimation of dominance genetic effects may occur. The bias in variance components was also illustrated with an empirical data set: instead of finding a heritability estimate of 40% for attention problems with a sum score (Van den Berg et al. 2006c), a heritability estimate of 73% was obtained when including a measurement model and estimating it simultaneously with the variance decomposition model. This example provides an additional illustration of the bias in variance components due to the analysis of sum scores. However, it should be noted that model fit was not assessed, nor was the assumption of measurement invariance tested. This requires further study.
