Estimating interaction and nonlinear effects with SAS

This chapter presents a small SAS program that takes as input the PROC CALIS program setup for a structural model that estimates only linear effects. It analyses based on the original raw data, data that were rank transformed, and subsets of the data excluding abstaining participants or those who met liberal criteria for being multivariate outliers. Polynomial regression techniques have been used to describe the functional relationship of quadratic or interaction effects in addition to main effects of variables. Although there is a clear conceptual utility in estimating such quadratic and interaction effects, such regression models have been shown to produce biased and/or inconsistent estimates if the predictor variables contain significant amounts of measurement error. Like J. Jaccard and C. K. Wan, K. G. Joreskog and F. Yang have also noted that the advent of the ability to program nonlinear constraints in structural equation modelling programs eliminates the need to specify the cross-product parameters.