6.1. Introduction

In chapter 2, we saw how to estimate fixed effects linear regression models using several different methods and several different SAS procedures, including REG, GLM, TSCSREG, GENMOD, and MIXED. Now we are going to estimate the same model with PROC CALIS, which is designed to estimate linear structural equation models with latent variables, sometimes known as LISREL models. Why do we need another SAS procedure when we already have five that will do the job? The reason is that by estimating the linear fixed effects model in CALIS, we can do several important things that are not possible with the other procedures:

• estimate models that are a compromise between fixed and random effects models

• construct a likelihood ratio test for fixed versus random effects

• estimate fixed effects models that include reciprocal effects of two response variables

• estimate fixed effects models with lagged values of the response variable

A separate chapter is devoted to this method because the data structure and the conceptual framework is quite different from that used for most of the methods described in chapter 2. I will first explain how to use PROC CALIS to estimate the random effects model described in chapter 2. Then we will see how that model can be modified to produce the fixed effects model.