Empirical comparisons of analytic strategies for MIMIC DIF analysis: A potential solution for biased anchor set

The purpose of this Monte Carlo study was to evaluate the performance of the multiple indicators and multiple causes (MIMIC) confirmatory factor analysis (CFA) for detecting differential item functioning (DIF). Specifically, this study compared different application strategies including two conventional testing approaches (forward-inclusion, backward-elimination) and five test statistic values (uncorrected or Bonferroni-corrected LR, △CFI of 0.01 or 0.002, △SRMR of 0.005) across conditions of different item type, test length, sample size, impact, and DIF type and DIF size in a target item and an anchor set. In addition, the author proposed an alternative testing approach (effects-coded backward-elimination) as a potential solution for arbitrary choice of a DIF-free anchor set. Simulation results indicated that when an anchor set was truly biased, only the proposed approach performed adequately under several conditions. False positive rates were controlled at the nominal alpha level (w