Proceedings of the Annual Meeting of the Cognitive Science Society

Latent decision strategies are central to human behavior and cognition. They may differ across persons and fluctuate over time within a single person. This paper introduces a framework that extracts latent, trial-level decision strategies from think-aloud verbalizations, with the resulting strategy estimates mapped onto formal strategy identification via quantitative model selection. We validated this approach with two intertemporal choice experiments, in which participants verbalized their thoughts while making binary choices between smaller-sooner and larger-later options in free-choice (Experiment 1) or instructed (Experiment 2) conditions. We used three popular LLMs to rate participants' alternative-based versus attribute-based evaluations based on their think-aloud protocols at the trial level. Results suggest that think-aloud protocols effectively capture individual variations in decision strategies, as reflected in comparisons between computational models, and can detect strategy shifts due to instructions. Our framework offers a scalable, powerful tool for understanding latent cognitive processes underlying human choice behavior.