Contingency Table Models for Assessing Item Bias

Abstract

Two strategies for assessing item bias are discussed: Methods based on Itemx Group interaction and methods that compare the probabilities of a correct response for different groups conditional on ability level. In latent trait models, correct response probabilties are compared conditional on latent ability; inScheuneman’s (1979) method these probabilities are compared conditional on the observed test score. Scheuneman’s method is modified to fit the general theory of loglinear and logit models for contingency tables. A distinction is made between uniform and nonuniform item bias, and a method to assess item bias and distinguish between uniform and nonuniform bias is described. Data reported by Scheuneman are used to demonstrate the method in detail; the differences obtained with Scheuneman’s method are discussed. In addition, the method is applied to two tests administered byvan der Flier (1980) in Kenya and Tanzania, and the results are compared to those obtained with Scheuneman’s approach.