Estimating and Testing Mediation Effects with Censored Data

This study investigated influences of censored data on mediation analysis. Mediation effect estimates can be biased and inefficient with censoring on any one of the input, mediation, and output variables. A Bayesian Tobit approach was introduced to estimate and test mediation effects with censored data. Simulation results showed that the Bayesian Tobit approach can be used to deal with censoring in estimating and testing mediation effects under certain conditions. The proposed method was illustrated by analyzing an empirical data set with more than 50% of censored data on the output variable. Some guidelines for applying the proposed method were discussed.