A Regularization Method for Detecting Differential Item Functioning Under the Framework of Generalized Linear Models

• Activity Number: 120 - SPEED: Variable Selection and Networks
• Type: Contributed
• Date/Time: Monday, July 31, 2017 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Learning and Data Science
• Author(s): Jing Jiang* and Zhushan Li
• Companies: Boston College and Boston College
• Keywords: differential item functioning ; regularized logistic regression ; item response theory model

Abstract

The purpose of the study is to present a regularization method for estimating differential item functioning (DIF) parameters under generalized linear models. DIF occurs when the probabilities of correctly responding to an item are unexpectedly different for individuals from different groups with a same latent ability level. Traditional DIF detection approaches usually require all items except the one under detection to be DIF-free, which is possibly wrong. Otherwise, failing to identify invariant anchors will lead to inflated type I errors. This problem can be solved by simultaneous estimation of DIF parameters in one model by using regularized logistic regression. Simulation studies were conducted to compare this proposed method with other DIF detection techniques such as Mantel-Haenszel method and logistic regression method, and the results indicated the feasibility and applicability of the proposed method.