The Effects of Missing Time-Varying Covariates in Multilevel Models

Last Modified

Creator

Bainter, Sierra A.
- Affiliation: College of Arts and Sciences, Department of Psychology and Neuroscience

Abstract

Multilevel models are commonly used in psychological research to examine developmentally-motivated hypotheses concerning the within- and between- person effects of a time-varying covariate on some outcome. Whereas multilevel models are flexible to accommodate incomplete data on the outcome, missing time-varying covariates present significant challenges to researchers. Unless multiple imputation is used, missing time-varying covariates will lead to a loss of data. This project evaluated the effects of missing time-varying covariates and imputation of missing time-varying covariates in multilevel models using a multifaceted simulation study. My results showed that missing time-varying covariates can lead to biased parameter estimates. However this bias is likely minor compared to bias already present in complete data due to unreliable estimates of the person-mean of the time-varying covariate. The results presented here are clear motivation for researchers to choose alternative estimation strategies that can account for measurement error in the person mean, whether or not time-varying covariates are missing.

Date of publication