The unconditional model is a useful starting point, providing preliminary information on the quality of the indicators via inspection of factor loadings, tracelines, and information functions (see P.J. Curran, J. McGinley, D.J. Bauer, A.M. Hussong, A. Burns, L. Chassin, K. Sher, R.A. Zucker, manuscript in preparation). Yet the unconditional model also makes a number of unrealistic assumptions, including that the distribution of the latent variable is homogeneous across subpopulations (i.e., studies). It also assumes that the relationships between the latent variable and the indicators are equal across subpopulations. We will test each of these assumptions in turn.
