Development of the Logistic Regression with Factor Scores (LRFS) Method in Differential Item Functioning (DIF) Detection for Dichotomized Variables

• Activity Number: 71
• Type: Contributed
• Date/Time: Sunday, August 1, 2010 : 4:00 PM to 5:50 PM
• Sponsor: Biometrics Section
• Author(s): Chengwu Yang*+ and Elizabeth Garrett-Mayer and Barbara C. Tilley and Jeanne A. Teresi and Richard N. Jones and Rickey E. Carter and Vanessa K. Hinson
• Companies: Penn State and Medical University of South Carolina and The University of Texas Health Science Center at Houston and Columbia University Stroud Center and Hebrew Rehabilitation Center for Aged and Mayo Clinic and Medical University of South Carolina
• Keywords: differential item functioning (DIF) ; logistic regression ; factor analysis ; measurement bias

Abstract

Differential item functioning (DIF) analysis is a key element in studies of measurement bias, which is common in medical research where the populations are usually heterogeneous. But DIF analysis is not widely used, in part due to its difficulty. Most DIF analysis methods are tedious and formidable. We developed a new method for dichotomized variables using logistic regression with factor scores (LRFS) for estimation of the latent trait in R. We used it to assess DIF in an object naming scale dataset, and compared its procedures and results to five other common methods. The LRFS method, while easier to implement, gives results in agreement with four of the other five. It offers an easy tool to do DIF analysis, has the potential of being adopted by researchers in medicine, and may help improve the quality of medical research that uses measures such as self reports and rating scales.