Logistic Regression Using SAS®: Theory and Application

8.3. GEE Estimation

Beginning with Release 6.12 SAS/STAT, PROC GENMOD makes it easy to get efficient estimates of the coefficients and improved standard error estimates with clustered data by using the method of generalized estimating equations (Diggle et al. 1994). Here’s a brief, nontechnical explanation of this method. In ordinary logit analysis, the maximum likelihood estimates can be obtained by an algorithm known as iteratively reweighted least squares. What that means is that each step in the algorithm is accomplished by weighted least squares, with both the weights and the constructed dependent variable changing at each iteration as functions of the results at the last iteration. In the matrix formulation of weighted least squares, there ...