How Discrimination and Preference Parameters Affect Block Option Selection in Multidimensional Forced-Choice Tests

The effectiveness of multidimensional forced-choice (MFC) assessments depends not only on the measurement model but also on how options are paired into blocks. This simulation study evaluated four block option selection (BOS) strategies for fixed-form MFC assessments under the multi-unidimensional pairwise preference model with two parameters: an option discrimination parameter and an option preference parameter. The study explicitly compared the separate and joint contributions of option discrimination and option preference parameters to the accuracy of latent trait estimates. The four algorithms were based on discrimination only (∏α), preference only (Δβ), and two joint algorithms that used both discrimination and preference information (∏α + Δβ and ∏α/Δβ). A fully crossed 2 × 3 × 4 design manipulated trait mean distinctness, inter-trait correlation, and BOS algorithm. Results showed that the Δβ and ∏α/Δβ algorithms consistently produced the two lowest mean absolute error values across all five traits and six trait-profile conditions. Inter-trait correlation was a stronger contextual factor than trait mean distinctness: as correlations increased, the advantage of algorithms using preference information over ∏α became larger. These findings suggest that option discrimination and preference parameters make distinct contributions to BOS and that incorporating option preference information, either alone or jointly with discrimination information, can improve the accuracy of latent trait estimates in fixed-form MFC assessments.