Detection of DIF Based on Non-IRT Generalized Regression Approaches

Abstrakt

Logistic Regression (LR) models have been widely used for detection of Differential Item Functioning (DIF) since work of Swaminathan and Rogers (1980). In this work we consider various extensions of LR models for DIF detection while accounting for other aspects of the model. These include guessing or non-attention parameters in dichotomous and nominal data, which can bring better precision as well deeper understanding of the variables examined. Some of these models have been accounted for within Item Response Theory (IRT) framework. We argue that presented non-IRT models fill a logical gap in methodology and as such are important for educational purposes. Simulation studies suggest good properties as well as advantages in case of smaller samples. Models are implemented within difNLR library in R and offered to wider audience through ShinyItemAnalysis web application.