Sample interview questions: How do you handle issues of missingness in longitudinal mixed-effects models with time-varying covariates?
Sample answer:
-
Multiple Imputation:
Generate multiple (e.g., 5-10) complete datasets by imputing missing values using appropriate methods (e.g., predictive mean matching, multiple regression, or stochastic regression imputation). Analyze each imputed dataset separately using mixed-effects models and combine the results using Rubin’s rules to obtain overall estimates and standard errors.
-
Full Information Maximum Likelihood (FIML):
Use FIML, which assumes missing data occurs at random (MAR) and utilizes all available information to estimate model parameters. FIML is implemented in some statistical software (e.g., SAS, Stata, or R packages like ‘nlme’) and provides unbiased estimates under MAR.
-
Mixed-Effects Model with Missing Covariates (MMCM):
MMCM explicitly models the missingness process by including auxiliary variables that indicate … Read full answer