Differential Item Functioning: When Does It Matter?

• Activity Number: 48
• Type: Invited
• Date/Time: Sunday, July 29, 2012 : 4:00 PM to 5:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Karen Bandeen-Roche*+ and Janne Petersen
• Companies: Johns Hopkins Bloomberg School of Public Health and Hvidovre University Hospital
• Keywords: latent ; measurement ; regression ; latent variable ; surrogate

Abstract

Latent variable models provide an appealing framework for studying and addressing measurement difficulties. Differential item functioning (DIF) is a difficulty that occurs when the relationship between a latent construct and the observable surrogates through which it is inferred varies by reporter characteristics: that is, differs by context. In our experience, however, analytic findings and inferences regarding relationships between latent constructs and their potential determinants or outcomes frequently differ little when addressing, and not addressing, DIF. Our paper presents analytical and numerical studies to delimit distributional features determining differences between estimator estimands and standard errors when employing latent variable models that do, and do not, address DIF, in cases where DIF exists. Illustration is made using epidemiological and clinical datasets on aging and HIV. We aim to deepen insight into the models and analyses at issue and improve practice when using these.