Robustness of Scoring for a Multistage Test Assessed by Subpopulation Invariance

• In Event: Computerized Testing: New Developments and Comparability Issues

Abstract

This study examined the robustness of scoring for a multistage admission test from the subpopulation invariance perspective. We used the logistic regression function to estimate item parameters of the operational items using the real data from the total group as well as each of the three subgroups. Using those items’ parameters, we produced the score conversions of the more than 4000+ multistage forms for each subgroup separately. The conversions derived from each of the three subgroups were generally comparable to the conversions from the total group. A particular subgroup showed non-negligible statistical bias in scoring caused by item misfit.

Author

• Sooyeon Kim, Educational Testing Service