Choice Set Imputation

Abstract

The problem of choice set formation for decision makers is an important subject in discrete choice modeling, especially when the choice set contains a large number of elemental alternatives. In general, the choice set of an individual could be a randomly sampled choice set; however, this is claimed to be a behaviorally unacceptable practice because of the fallacious assumption of individuals’ full knowledge of potential random choices. This issue brings up the need to devise methods to logically allocate credible choice alternatives for individuals. Although the use of these methods could be dependent on specific applications, this study attempted to identify the distinction between model estimation and prediction steps in the context of residential location choice modeling. From a theoretical point of view, the paper proposes a modified weighted stratified sampling approach that is an improved version of random sampling for model estimation. The approach is believed to be a better replicate of the universal choice set than other sampling methods, and it is capable of resulting in consistent estimates even with small sample sizes. The estimated model was applied in a simulation framework with a hazard-based imputed choice set approach for prediction.