Keynotes | K&Ts | GACs | Talks | Posters | Search

Poster B in Poster Session B: Tuesday, August 4, 2:00 – 3:45 pm, Kimmel Center, Shorin & Rosenthal Rooms

Characterizing Decision Strategies from Think-Aloud in Risky Choice

Cierra Ferguson¹, Hua-Dong Xiong², Hanbo Xie¹, Robert Wilson¹; ¹Georgia Institute of Technology, ²Honda Research Institution US

Presenter: Cierra Ferguson

Models of risky choice typically infer decision processes from gamble attributes and observed choices rather than directly observing the reasoning that unfolds during decision-making. Although such models can predict behavior well, their latent variables are researcher-specified, motivating interest in think-aloud protocols as a window into the human decision process. However, prior think-aloud work has often relied on either manual analysis that is difficult to scale or predictive use of language models that offers limited insight into reasoning structure. Here, we test whether think-aloud verbal reports can be distilled into interpretable representations that both reveal structure in reasoning and capture decision-relevant information that improves prediction of choice behavior. We compared baselines with think-aloud-based models ranging from interpretable abstractions of participants’ verbal reports to sanitized and full transcript representations. Think-aloud-based representations improved prediction over an LLM baseline that received only option values and probabilities, with progressively richer representations yielding higher accuracy. Some models that used abstract representations of think-aloud also outperformed the Expected Value and Prospect Theory baselines. These results suggest that think-aloud can be transformed into interpretable abstractions that reveal sequential reasoning structure and additional decision-relevant signal not captured by gamble attributes and observed choices alone. More broadly, this framework uses think-aloud to bridge formal models of choice with the reasoning processes that drive behavior.

Topic Area: Decision-Making, Cognitive Control & Event Cognition