Evaluating Bayesian Variable Selection Approaches for Nonlinear Random Effects Models

• In Event: AERA Roundtable Session Friday 7:45 am - Gold 4
  In Roundtable Session: RT4: Modeling Complex Data and Patterns: Bayesian, Longitudinal, and Machine Learning Approaches

Abstract

Longitudinal data modeling is particularly helpful for understanding individual and mean trends of development over time. One feature of longitudinal design is the collection of extensive information on participants to explain the underlying growth process. However, selecting which covariates (i.e., predictors, or independent variables) to include in a statistical model is challenging with the goals of avoiding both overfitting and underfitting. The present study demonstrates and compares multiple Bayesian variable selection methods, including two broad categories: (1) shrinkage methods and (2) spike-and-slab methods, via an extensive Monte Carlo simulation study. Our goal is to provide recommendations for researchers and practitioners about Bayesian variable selection methods regarding their effectiveness in variable selection, model convergence, and parameter estimation precision and accuracy.

Authors

• Yue Zhao, University of Minnesota

• Nidhi Kohli, University of Minnesota

• Eric F. Lock, University of Minnesota