Observing Unobserved Heterogeneity: Using Process Data to Enhance Choice Models

People use different strategies when making choices. Modeling this choice process heterogeneity, however, is difficult using just the data provided by most standard choice experiments. We try to capture process heterogeneity by augmenting choice models with variables derived from information-acquisition data gathered unobtrusively during choice tasks. These variables supplement standard logit specifications which identify how an individual used the attributes and attribute values to screen and rank alternatives in making a choice. The approach improves in-sample fit, prediction in a holdout sample, and residuals indicate that the models are providing better specified estimates of choice probabilities.