Handling missing data on DIF detection under the MIMIC model


This study employs a two-step procedure which selects an anchor set and assesses the items outside the anchor for DIF (M-IT/M-PA; Wang & Shih, 2010). Results show that accuracy rates in selecting an anchor with multiple imputation were generally higher than with full-information maximum likelihood.

When data is MCAR, DIF detection yields high power even with 30% of missing, large sample size and moderate DIF effect. Under MAR and MNAR, however, DIF detection was affected when sample size was small and DIF effect was moderate.

Work presented at the 2015 National Council of Measurement in Education (NCME) as a conference talk. Please contact the author for further details.