Detecting DIF With Ideal Point Models A Comparison of Area and Parameter Difference Methods

Ideal point models have become increasingly popular in research and practice, but little is known about how traditional methods for linking and detecting differential item functioning (DIF) perform with data satisfying ideal point assumptions. Very few studies have been conducted on this topic, and the results of some well-known area and parameter difference DIF detection methods have been inconsistent with previous research involving dominance models. Consequently, the authors conducted a Monte Carlo study to help identify sources of these discrepancies. Specifically, they compared the effectiveness of the Lord's chi-square parameter difference method and the differential functioning of items and tests (DFIT) area method for detecting DIF with the Generalized Graded Unfolding Model (GGUM). The results clearly indicated that when DIF was simulated via parameter shifts that produced constant magnitudes of effect for all designated items, Lord's chi-square and DFIT performed similarly well. In addition, the authors found that iterative item characteristic curve (ICC) linking outperformed iterative test characteristic curve (TCC) linking in most experimental conditions. The implications of these findings for DIF detection research and practice with ideal point models are discussed.