Stability of differential item functioning over a single population in survey data

This study investigates the stability of differential item functioning (DIF) in survey data. Surveys are conducted periodically, and their results are often reported by aggregating responses. Estimating the stability of DIF across subsets of a survey population can be an important indicator in determining the likelihood of DIF stability over different populations. Data from four previously administered surveys were used in the analysis of the stability of gender-related DIF. The surveys were the Family Survey, Economic Expectations and Attitudes, Ministry With Young Adults Initiative, and Attitudes Toward the Environment Survey. Two samples of 500 participants each were randomly selected and used from each survey. The Mantel-Haenszel and the logistic regression procedures were used separately to detect DIF. Results showed that DIF in survey data is highly stable over subsamples from a single population.