Statistical analyses were performed with genetic structural equation modeling as implemented in the software package Mx [24]. For continuous variables, a so-called saturated model was fitted to the data in which means (for men and women), variances (for men and women) and five twin correlations were estimated. The regression of age (z-value) or year of birth was (z-value) (separately for man and women) was modeled as a fixed effect, allowing for a linear decrease or increase of the mean with age or cohort. For the dichotomous variables, a threshold model was applied, in which a trait is assumed to have an underlying continuous liability with a standard normal distribution with zero mean and unit variance. Thresholds divide this normal distribution into discrete categories [3]. Different thresholds were estimated for men and women. A regression of the z-score of age or cohort was modeled as a fixed effect on the threshold.
