EQS and LISREL Examples Using Survey Data

Abstract

LISREL 8 provides a nonlinear constraint capability that is different from that available in COSAN, and the constraint coding effort is reduced. Researchers in the social sciences have called for the estimation of continuous interaction and quadratic effects to improve the interpretation of significant first-order effects. Estimation techniques for interaction and quadratic latent variables in structural equation models can be classified into direct and indirect approaches. Kenny and Judd proposed that latent variable interactions and quadratics could be specified adequately using all unique products of the indicators of their constituent linear latent variable. Subgroup analysis generally involves splitting the sample and assessing differences in model fit when the model is restricted to the resulting groups of cases. The number of product indicators in a model with interactions and quadratics involving overidentified constituent latent variables can become large.