We initially focused on how the distribution of the latent variable may differ across subpopulations. For our example, we expect that depression levels may vary across participants as a function of study, age, history of parental alcoholism, and gender. Such differences are sometimes referred to as impact (Holland & Wainer, 1993). To help us better understand potential impact effects, we generated factor scores from the unconditional model and plotted them as a function of the covariates. For instance, to get an initial idea of potential age-related changes in depression, we plotted factor scores as a function of age and study, determining that the age trends in the scores differed across studies but could be well-approximated in each study by a cubic function.
