A study on the robustness of linear mixed effects models to violation of error assumptions

The linear mixed effects model is widely used for analyzing continuous longitudinal data, which involves repeated measurements taken over time.Such data may exhibit heteroscedasticity or correlation among error terms.If these issues are not addressed, the model may be misspecified by assuming homoscedasticity and independence.Additionally, in longitudinal studies, some subjects may not be observed until the planned final time point, potentially leading to censoring.This study presents simulation results to examine how the linear mixed model-assuming homoscedasticity and independence-estimates fixed effects in both complete and censored data under various conditions of heteroscedasticity or correlated error terms.It also offers guidelines for addressing these situations.