Longitudinal Differential Item Functioning Detection Using Bifactor Models and the Wald Test

Abstract

The use of longitudinal data for studying cross-time changes is built on the key assumption that properties (e.g., slopes and intercepts) of the repeatedly-used items remain unchanged over time. True changes in the latent variables are indistinguishable from item-level changes when items exhibit differential item functioning (DIF) across time points. To date, no research has extended the modified Wald test for longitudinal DIF detection. The current Monte Carlo simulation study proposes and evaluates a new approach, which pairs the versatile bifactor model with the modified Wald test, for detecting longitudinal DIF. Power and Type I error associated with DIF tests under the new approach are reported for conditions with varied proportions of known anchors and different types of standard error estimation procedure. The new approach is also compared to DIF methods based on the misspecified unidimensional model which assumes independence in the factors and items. An applied example is provided, along with the flexMIRT script and the R code used respectively for model calibration and DIF analysis. Limitations of the current study and future research directions are discussed.