Modeling sampling duration in decisions from experience

Cognitive models of choice almost universally implicate sequential evidence accumulation as a fundamental element of the mechanism by which preferences are formed. When to stop evidence accumulation is an important question that such models do not currently try to answer. We present the first cognitive model that accurately predicts stopping decisions in individual economic decisions-from-experience trials, using an online learning model. Analysis of stopping decisions across three different datasets reveals three useful predictors of sampling duration - relative evidence strength, how long it takes participants to see all rewards, and a novel indicator of convergence of an underlying learning process, which we call predictive volatility. We quantify the relative strengths of these factors in predicting observers' stopping points, finding that predictive volatility consistently dominates relative evidence strength in stopping decisions.