Generalized Linear Mixed-Effects Models in R

The University of Wisconsin and Toyon Research Corporation propose developing a software package in the R statistical environment for estimating generalized linear mixed-effects models (GLMMs) from multilevel, categorical data (e.g., binary, ordinal ormultinomial data). The proposed software package should enable data analysts and researchers to more accurately model and understand this type of survey data. The package is aimed toward a user who may be familiar with mixed-effects models, but is not anexpert in the numerical procedures required to estimate GLMMs.Estimating the parameters for a GLMM can be a computationally intensive task, and remains an active field of academic and industrial research. Furthermore, a brute force approach, which may converge extremely slowly on a desktop computer, is unlikely toestimate an accurate GLMM. In our proposed approach, we avoid a large computational problem from the outset by proceeding in a staged fashion. We begin with an approximate solution, and gradually increase the model accuracy using more numerically intensivemethods. In this staged approach, the most numerically intensive methods operate on a smaller, better defined problem, which results in a fast and accurate solution. The successful completion of this research will result in extensions to the NLME R packagethat will enable statisticians and analysts to build generalized linear mixed-effects models for binary and ordinal data, emphasizing data collected in longitudinal studies.