This presentation is part of: Poster Session - Utility Theory; Health Economics; Patient & Physician Preferences; Simulation; Technology Assessment
DEMOGRAPHIC DIFFERENCES IN ITEM RESPONSES TO PHYSICAL FUNCTIONING AND MENTAL HEALTH DOMAINS OF THE SF-36
Yanni F. Yu, MA, Andrew P. Yu, MA, and Jeonghoon Ahn, Ph.D. University of Southern California, Pharmaceutical Economics and Policy, Los Angeles, CA
Purpose: Differential item functioning (DIF) is the observation that individuals with equal ability from different groups present unequal probabilities of endorsing a certain item. The purpose of this study is to investigate whether demographic characteristics have uniform DIF (effect extent is same for all ability levels) or non-uniform DIF (effect extent varies with ability level) effect on items in physical functioning (PF) and mental health (MH) domains of the SF-36, and to examine overall validity of PF and MH items from the DIF perspective. Methods: Those who completed SF-36 between 04/1994 and 02/1995 (n=7538) were extracted from a cohort of Southern California Kaiser Permanente members. Ordinal logistic regressions (OLR) were applied separately to investigate uniform and non-uniform DIF effect of seven demographic variables including age, gender, ethnicity, education, income, marital and employment status through three steps: entering 1) matching variable, 2) group variable, and 3) interaction term of matching and group variables. The first and last two models were compared to determine the presence of uniform and non-uniform DIF using log-likelihood test and the effect size of DIF based on difference of R-square (<0.035 negligible, 0.035~0.070 moderate, or >0.070 large). Results: A total of 210 scenarios (combinations of 15 items and 7 group variables) were examined. After matching on the latent ability level represented by total score of PF or MH domain, all PF and MH items displayed uniform (U) or non-uniform (NU) DIF across at least one demographic group. The items were more sensitive to difference of age (12 items-U, 8-NU), ethnicity (11-U, 11-NU), and income (10-U, 10-NU), but less to marital status (7-U, 4-NU). The extent of variation in item responses due to different education (10-NU) and employment status (11-NU) were more likely to change with the level of underlying ability than the extent of variation due to difference of other demographic attributes. As an overall measure, the effect sizes of present DIF across all demographic groups were small (¥ÄR2 <0.035), evidencing satisfactory DIF validity. Conclusions: In this study, PF and MH items showed no strong DIF associated with different demographic characteristics. From the DIF perspective, the study revealed the satisfactory validity of PF and MH domains on the item level.
