Meta-Analysis of the Mantel-Haenszel Index for the Detection of Differential Item Functioning

This dissertation proposes an approach for synthesizing the Mantel-Haenszel (MH) Differential Item Functioning (DIF) index. It is shown that meta-analytic summaries of MH DIF estimates have potential implications for test development through observations of DIF changes. DIF changes could occur if the DIF magnitudes of items vary over time or across different groups. This approach is an extension of meta-analysis techniques for odds ratios. Unlike the typical odds ratio, the MH DIF is a common odds ratio, indicating the average of odds ratios across all matched ability levels. Also, the dissertation examines how the Q test of homogeneity for MH DIF indices behaves as a function of the magnitude of variations in DIF using a simulation study. This proposed approach is used with real data. My example uses data on the reading achievement of standard curriculum students (SCS) and students with disabilities (SWD) on a state-wide assessment. Thus, this dissertation makes a contribution to the understanding of SWD DIF trends in reading achievement by investigating meta-analytic summaries of the MH DIF index. In addition, this dissertation investigates whether the values of DIF for SCS vs. SWD are dependent on the type of reading item tested. Educational implications are discussed.