Quasi-least squares (QLS) is a computational approach for estimation of the correlation parameters that is in the framework of generalized estimating equations (GEE). This talk will give an overview of QLS that includes: why QLS is useful; how we specify a QLS model; a listing of software (and related papers) that we have developed for implementation of QLS in Stata, R, Matlab, and SAS; details regarding the QLS estimation procedure; and finally, some brief discussion of our current research in this area that concerns methods for selection of working correlation structures for GEE/QLS analysis of correlated binary data. A particular focus of this presentation is implementation of QLS for members of a class of correlation structures that have been thoroughly studied for the analysis of longitudinal data, but whose application for binary outcomes has been limited. For example, we have proposed a simple model for simulation of correlated binary data for members of this class of structures, some of which to our knowledge, have not been featured in any prior manuscript on the simulation of correlated binary data. The unifying theme of this presentation is that QLS, like GEE, is a relatively simple and straightforward approach that as such, can be very useful with respect to both methods research and statistical consultations.
Return to Biostatistics Working Group