Investigating differential item functioning by chronic diseases in the SF-36 health survey: a latent trait analysis using MIMIC models

OBJECTIVES: Differential item functioning (DIF) is present when respondents of unique subgroups endorse certain items differently given the respondents have the same underlying ability. This study investigates the presence of DIF regarding chronic illnesses among items of the physical functioning (PF) and mental health (MH) domains of the SF-36 health survey. METHODS: Multiple indicators multiple causes (MIMIC) model was applied to data extracted from the Kaiser Permanente database for members who completed the SF-36 during 1994-1995 (N = 7538). DIF effects were evaluated for sociodemographic variables and for indicators of 5 chronic conditions: hypertension, rheumatic conditions, diabetes, respiratory diseases, and depression. An iterative strategy with backward selection was applied to build DIF models, which were estimated by weighted least squares. The Hochberg procedure was applied to P values for multiple tests. RESULTS: After controlling for the selected covariates and the latent ability, DIF was present in 3 items for hypertension, one for respiratory diseases, and one for diabetes. Adjusting for DIF did not modify the overall pattern of exogenous variables' effects on PF or MH, except Hispanic and other ethnicity on PF, education on MH became insignificant; and black ethnicity on PF, old ages and other ethnicity on MH became significant. CONCLUSIONS: Considering the number of items and disease subgroups compared, the presence of DIF was minimal among items of the PF and MH domains of the SF-36. DIF had little effect on comparisons of sociodemographic or disease groups.