Evaluating Type 1 Error in Differential Item Functioning Methods Based on Item Parameter Distributional Shape

• In Event: Roundtable Session 28
  In Roundtable Session: Topics in Differential Item Functioning

Abstract

Factors involved in generating dichotomous item response data based on a distributional shape are not widely addressed in the Monte Carlo differential item functioning literature. This simulation study assessed how the use of a normal and uniform distribution to generate 2PL data impacted Type 1 error of Lord’s Chi-Square (Lord, 1980) and the Mantel-Haenszel (Holland & Thayer, 1988) procedure. Results indicate there is little to no difference in Type 1 error based on item parameter distributional shape suggesting that at least these two statistical tests are not overly sensitive to these item assumptions. This finding is applicable to psychometricians, test constructors, policy makers, and educational researchers.

Authors

• Emily A. Price, Ohio University

• Gordon P. Brooks, Ohio University - Athens

• George A. Johanson, Ohio University