Statistics and Its Interface

Volume 12 (2019)

Number 2

Variable selection for correlated bivariate mixed outcomes using penalized generalized estimating equations

Pages: 265 – 274

DOI: https://dx.doi.org/10.4310/SII.2019.v12.n2.a7

Authors

Ved Deshpande (Department of Statistics, University of Connecticut, Storrs, Ct., U.S.A.)

Dipak K. Dey (Department of Statistics, University of Connecticut, Storrs, Ct., U.S.A.)

Elizabeth D. Schifano (Department of Statistics, University of Connecticut, Storrs, Ct., U.S.A.)

Abstract

We propose a penalized generalized estimating equations framework to jointly model correlated bivariate binary and continuous outcomes involving multiple predictor variables. We use sparsity-inducing penalty functions to simultaneously estimate the regression coefficients and perform variable selection on the predictors, and use cross-validation to select the tuning parameters. We further propose a method for tuning parameter selection that can control a desired false discovery rate. Using simulation studies, we demonstrate that the proposed joint modeling approach performs better in terms of accuracy and variable selection than separate penalized regressions on the binary and the continuous outcomes. We demonstrate the application of the method on a medical expenditure data set.

Keywords

false discovery rate, generalized estimating equations, medical expenditure, penalized estimation

2010 Mathematics Subject Classification

Primary 62J07. Secondary 62J12.

Full Text (PDF format)

Received 14 September 2017

Published 11 March 2019