Objective: Measures of shoulder function may differ by dominance of affected shoulder, surgical history, gender, or race. We present a technique for determining whether observed differences in function between groups are due to biased test items or real differences in function.
Study design and setting: Four hundred patients who were receiving rehabilitation for a variety of shoulder impairments completed a survey of shoulder function. Thirty-seven items measuring shoulder function were analyzed for differential item functioning (DIF) related to demographic characteristics using an ordinal logistic regression (OLR) and item response theory (IRT) approach. When DIF was identified in an item, we modified the IRT analysis to calibrate item parameters separately in appropriate demographic groups. We compared adjusted and unadjusted patient ability measures in each demographic group.
Results: Several items were found to have a modest amount of DIF related to the different demographic characteristics, especially gender; however, adjusting measures for DIF had little impact on overall measures of shoulder function and made almost no difference in average shoulder function across demographic groups.
Conclusion: In this pool of shoulder function items, adjustment for DIF made almost no difference in measures of function across demographic groups.