Maria Robinson
Dr. James Adelman
Dr. Kenny Yu
A central part of theory testing in mathematical psychology is testing for invariance. One common approach is joint model fitting, in which a single set of parameters is fit to multiple conditions or datasets and assessed by relative fit against a less constrained model. Generalization tests instead treat testing invariance as an a priori prediction problem, where parameters estimated in one context are used to predict behavior in another, and the invariance claim is evaluated by out-of-sample predictive performance. This talk offers a formal and conceptual comparison of these approaches. The core claim is that generalization tests can be stronger tests of models because they are sensitive to model misspecification in addition to failures of invariance. This talk considers what constitutes model misspecification, how parameter estimates can be optimal for a given fitting criterion even under misspecification, and when violations due to misspecification are consequential for theory assessment. The talk builds directly on work by Yu and Robinson (in preparation).
This is an in-person presentation on July 18, 2026
(17:20 ~
18:20 EDT).
