Bias Reduction in Robust Mean–Geometric Mean Linking via SIMEX

Robust mean–geometric mean (MGM) linking is a method for comparing the performance of two groups on a test involving dichotomous items and is particularly suited to settings with fixed and sparse differential item functioning (DIF). However, robust MGM linking has been shown to yield biased estimates in finite samples because the estimated item parameters are affected by sampling error, which in turn induces bias in the estimated linking parameters. To address this issue, the simulation extrapolation (SIMEX) method is applied to robust MGM linking to reduce bias in the linking parameter estimates. Results from a simulation study demonstrate that SIMEX reduces bias in robust MGM linking. Moreover, SIMEX with a linear extrapolation function also reduces the variance of the parameter estimates in the absence of DIF effects. These findings indicate that the application of SIMEX in robust MGM linking methods can be generally recommended for empirical research aimed at removing DIF items from group comparisons.