Investigating the Impact of Imputation Methods on the Non-Uniform DIF Detection in the MIMIC Model

• In Event: AERA Roundtable Session 10
  In Roundtable Session: Missing Data in Measurement

Abstract

This simulation study investigated how three methods for dealing with missing data (listwise deletion [LW], two-way imputation [TW], multiple imputations [MI]) impacted the performance of non-uniform DIF detection in the MIMIC model under three mechanisms of missingness. The preliminary results indicate that the TW and MI imputation methods produced inflated type-I error rates than the complete data in the case of large missingness whereas type-I error rates of LW are like the complete data across the simulation conditions. Given the type-I error rates, we may expect that MI may be computationally expensive to be an alternative to the LW deletion method to address the problem of nonuniform DIF detection by the MIMIC model in the existence of missing data.

Authors

• Onur Demirkaya, University of Illinois at Urbana-Champaign

• Ge Jiang, University of Illinois at Urbana-Champaign