The standard univariate twin model can be extended to include several phenotypes and several genetic and environmental factors. In the second analysis, we used a bivariate model (Cholesky) to examine the assumption that both assessments of the LTH-AD reflect the same underlying liability, that is, that the second assessment was not influenced by occasion-specific familial factors. This was performed to justify the use of the third type of models, the so-called measurement models (see Fig. 1). Here, the underlying assumption is that each assessment of LTH-AD serves as a separate assessment of the same phenotype. Thus, the covariance among individuals between the 2 assessments can be modeled as a latent common factor of reliable LTH-AD liability. Assessments at both time 1 and time 2 are assumed to be imperfect indicators of a true latent liability of AD (Fig. 1A). The λ1 and λ2 denote the linear regressions between the latent liability to AD and the assessments (i.e., the square root of the test–retest reliability). The other 2 paths influencing the LTH-AD, κ1 and κ2, denote the effects of the residual
