Using Differential Functioning of Items and Tests (DFIT) to Examine Targeted Differential Item Functioning

Current studies of differential item functioning (DIF) look at look at how groups differ in responding to items across an entire trait continuum. This is important for detecting the presence of consistent patterns of responses across items between groups of people. Current tests of DIF are limited in that they only detect differences between groups across all levels of the trait. However, selection decisions are usually made within specific ranges of trait levels. The purpose of this research was to determine if restricting theta values in an existing framework would be better at detecting DIF as current methods for restricted ranges of the trait continuum. This Monte Carlo study used a 3 (difficulty DIF) by 4 (discrimination DIF) by 2 (canceling versus noncanceling DIF) design. Traditional differential functioning of items and tests (DFIT) framework analyses were used and then rerun using the targeted ranges of theta. The targeted ranges were defined as the 100 lowest and 100 highest theta values. Type I error rates and power analyses were examined. Results indicate that it is possible to detect DIF accurately at specific trait levels when DIF was not detected across the entire range of theta values. This research has implications for using cut scores at particular levels of a trait for items that have not been assessed using the new, targeted ranges. Limitations and future research are discussed