Fast Pure R Implementation of GEE: Application of the Matrix Package

Authors

Lee S. McDaniel, University of Wisconsin-Madison
Nicholas C. Henderson, University of Wisconsin-Madison
Paul J. Rathouz, University of Wisconsin-Madison

Date of this Version

6-2013

Document Type

Article

Citation

The R Journal (June 2013) 5(1); Editor: Hadley Wickham

Comments

Copyright 2013, The R Foundation. Open access material. License: CC BY 3.0 Unported

Abstract

Generalized estimating equation solvers in R only allow for a few pre-determined options for the link and variance functions. We provide a package, geeM, which is implemented entirely in R and allows for user specified link and variance functions. The sparse matrix representations provided in the Matrix package enable a fast implementation. To gain speed, we make use of analytic inverses of the working correlation when possible and a trick to find quick numeric inverses when an analytic inverse is not available. Through three examples, we demonstrate the speed of geeM, which is not much worse than C implementations like geepack and gee on small data sets and faster on large data sets.