Generalizing heuristic switching models and a (boundedly) rational route away from randomness

The behavioral economics literature on evolutionary discrete choice models typically relies on the standard logit framework. However, this approach imposes significant limitations on the types of economic environments it can represent as it, e.g., does not allow for heterogeneity in preferences regarding observables (random taste variation) and assumes independence of irrelevant alternatives (IIA). We relax the assumptions underlying standard logit and address two key questions: (i) to what extent do the theoretical insights of Brock and Hommes (1997) (BH) hold in more general economic settings? (ii) can the standard logit's shortcomings in capturing relevant experimental findings be resolved by using more flexible forms of discrete choice models? We find that a probit-based model that meaningfully relaxes the IIA assumption fits experimental data with four choice alternatives considerably better than standard logit, especially if the model additionally allows for random taste variation. Further, we demonstrate that while the key insights of BH remain valid in broader environments, allowing for taste variation can provide a route away from the chaotic dynamics emerging in BH.