Endogenizing Choice Set Formation Using the Perturbed Utility Model

Abstract

Defining the choice set is a critical challenge in discrete choice models with large sets of alternatives, such as destination choice. This study proposes a novel approach based on perturbed utility maximization, which explicitly represents zero probabilities for certain alternatives, enabling the simultaneous estimation of deterministic consideration sets and model parameters. Tests with synthetic data reveal that ignoring choice set restrictions in multinomial logit models leads to overestimation of choice probabilities and marginal effects, especially for higher values, resulting in biased policy evaluations. In contrast, the proposed method effectively mitigates these biases and enhances estimation accuracy. Application to a real-world destination choice model demonstrates the method’s scalability and robustness, confirming its ability to handle large choice sets and adapt consideration set sizes based on contextual factors.