Enhancing Precision in PISA's 2022 Creative Thinking Assessment: Toward Aim‐Directed Item‐ and Factor‐Level Analyses

Le résumé fourni par la source

ABSTRACT Large‐scale assessments often involve measurements with missing data where for a given student all items of a construct can be missing by design. In such circumstances, researchers are required to make methodological decisions in secondary analysis. A common approach is the use of plausible value estimation, which is designed for population‐level inference and thus works well for comparing groups. However, it is less suited to item‐ or factor‐level inquiries, where alternative methods may be preferable. Multiple imputation and full information maximum likelihood are two such alternatives. These methods handle missing data without relying on the same external conditioning variables used in plausible value generation, and instead operate directly on the observed item‐level data structure. To date, they remain underused, possibly because their simplicity contrasts with prevailing preferences for complex modeling. This paper compares two analytical paradigms—imputation‐based methods (plausible values, multiple imputation) and likelihood‐based methods (full information maximum likelihood)—in the context of the 2022 PISA Creative Thinking assessment. We highlight the usefulness of plausible values for population‐level comparison, multiple imputation when auxiliary information can be justifiably varied, and full information maximum likelihood for confirmatory modeling. We encourage practitioners to favor simpler approaches, as these often yield comparable results while improving transparency.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.