Investigating the Population Sensitivity Assumption of Item Response Theory True-Score Equating Across Two Subgroups of Examinees and Two Test Formats

Dorans and Holland (2000) and von Davier, Holland, and Thayer (2003) introduced measures of the degree to which an observed-score equating function is sensitive to the population on which it is computed. This article extends the findings of Dorans and Holland and of von Davier et al. to item response theory (IRT) true-score equating methods that are commonly used in the nonequivalent-groups anchor test (NEAT) design. Using data from the Advanced Placement Program Calculus AB exam, which contain multiple-choice (MC) and free-response (FR) sections, the authors investigate the population sensitivity of the IRT equating functions computed for the MC section only and for the MC and FR sections together. The degree of population sensitivity is also compared across three equating methods: the IRT true-score equating method and two observed-score equating methods, chained equipercentile and Tucker linear equating.