Comparison of Irt Likelihood Ratio Test and Logistic Regression Dif Detection Procedures

The Type I error rates and the power of IRT likelihood ratio test and cumulative logit ordinal logistic regression procedures in detecting differential item functioning (DIF) for polytomously scored items were investigated in this Monte Carlo simulation study. For this purpose, 54 simulation conditions (combinations of 3 sample sizes, 2 sample size ratios, 3 DIF magnitudes, and 3 DIF conditions) were generated and each simulation condition was replicated 200 times. In general, the Type I error rates of IRT likelihood ratio test and ordinal logistic regression procedures were in good control across all simulation conditions. The power of likelihood-ratio test was high for medium or large sample sizes and moderate or large DIF magnitude conditions. The power of this procedure increased as the sample size or DIF magnitude increased. On the other hand, the power of ordinal logistic regression procedure was unacceptably low for all DIF conditions except for the large sample size and large DIF magnitude condition.