Online Program

Saturday, February 21
CS24 Special Uses of R Sat, Feb 21, 11:00 AM - 12:30 PM
Maurepas

Sparse Matrix Computation in R with an Application to GEEs (302940)

View Presentation View Presentation

Nicholas C. Henderson, University of Wisconsin, Madison 
*Lee S. McDaniel, LSUHSC, School of Public Health 
Paul J. Rathouz, University of Wisconsin, Madison 

Keywords: Generalized estimating equations, R, sparse matrices

Most generalized estimating equation solvers in R only allow for a few predetermined options for the link and variance functions due to their C implementations. We introduce a package, geeM, which is implemented entirely in R and allows for user-specified link and variance functions.This allows the analyst to easily investigate the sensitivity of the analysis to the choice of link and variance functions.The geeM package also serves as an example of the benefits of sparse matrix representations, which are contained in the Matrix package. We discuss the speed improvements possible through the use of sparse matrices, using the inversion and multiplication of matrices involved in GEE solving for illustration. Through three data examples, we demonstrate that a pure R implementation can be faster than C implementations like geepack and gee on large data sets.