A Simulation Study of the Effects of Ability Range Restriction on IRT Item Bias Detection Procedures.

The effects of variations in degree of range restriction and different subgroup sample sizes on the validity of several item bias detection procedures based on Item Response Theory (IRT) were investigated in a simulation study. The degree of range restriction for each of two subpopulations was varied by cutting the specified subpopulation ability distribution at different locations and retaining the upper portion of the distribution. It was found that range restriction did have an effect on the accuracy of the bias detection procedures. The signed area index was least influenced by variations in range restriction, whereas the base low area index was found to be invalid regardless of the degree of range restriction. The findings for variations in sample size were mixed. (Author)