ABSTRACT
This chapter discusses an alternative limited information method, which makes use predicted latent scores to test the causal hypotheses formulated in the structural submodel. In this approach, one starts with the structure estimation of the limits of the measurement submodels and then proceeds to the estimation or prediction of the subjects' scores on the latent variables. Once these predicted latent scores are available, they treated as observed scores, and enter the later analyses based on the structural submodel. This approach is quite general and is, at least in principle, applicable to all types of latent structure models. Furthermore, the structural analyses can perform by standard statistical methods as linear or logistic regression analysis, and do not need specifically developed software. The chapter also discusses the use predicted latent scores from a population point of view and ignores the statistical issues that arise when applying the approach to finite samples.
