Beyond a Single Coefficient: A Model‐Based Framework for Local Predictive Validity along the Score Scale

Abstract Most studies on predictive validity summarize the relationship between admission test scores and academic performance using a single coefficient, such as a correlation or a linear regression slope. While informative, such summaries do not reveal how predictive strength varies across the score scale. In this article, we extend the concept of the marginal effect to the context of predictive validity, formalizing it as the local rate of change in expected academic performance given test scores. The outcome considered is the first‐year progress rate, a bounded variable between 0 and 1 that often takes values at the extremes. To respect this distributional structure, we implement a zero/one inflated Beta regression model, which simultaneously accounts for point masses at 0 and 1 and the continuous component in between. This framework enables us to examine how predictive capacity evolves across different score ranges, uncovering regions where test scores are more or less informative. By highlighting such heterogeneity, the approach provides richer insights into how admission tests function and offers valuable tools for analyzing predictive capacity and informing admission policies.