Differential item functioning in a large scale standardized mathematics assessment: The interaction of gender and ethnicity

DIF studies in the arena of mathematics achievement have investigated gender separately from ethnicity. The author of this dissertation argues that analyzing DIF at the global level does not serve the purpose of illuminating actual gender and ethnic performance differences. The purpose of this observational study is twofold: (1) to investigate how DIF was associated with gender and ethnicity for the total groups, (2) to conduct DIF analyses separately within gender and ethnicity. This research is unique in that it undertakes a 2-way DIF approach in taking account of the interaction of gender and ethnicity.;Data from the 1999 Delaware Mathematics Achievement Test were used for the analysis. The present study was set within the context of gender (male and female), four ethnic groups (African American, Asian American, Hispanics, and White), four grades, 3, 5, 8, and 10, for multiple-choice item formats. Two statistical procedures were utilized: Mantel-Haenszel and logistic regression DIF detection procedures.;Results obtain from this research show that in most comparisons, a roughly equal number of DIF items favoring the reference and the focal groups at either direction are found. However, a number of gender and ethnic DIF items that are previously undetected in a total analysis yet were flagged when 2-way procedures are applied (subsequent analyses of subgroups). Two specific instances are observed in Grade 3 and in Grade 5. This study has important theoretical and practical implications. It is clear that two-way analyses of DIF, beyond the traditional one-way DIF analyses, could benefit large-scale standardized testing programs. This more complete approach to DIF analysis enhances our understanding of the nature of DIF and may offer clues as to the causes of DIF, which may not be evident through one-way analyses.