A Measure of Random Utility Model Violations and Econometric Test
Updated:
This paper introduces a new measure of stochastic choice consistency for the random utility model (RUM). We extend the notion of RUM rationalizability by allowing “measurement error” on choice probabilities. This relaxation gives a linear programming problem that allows us to compute the minimal measurement error needed to rationalize the dataset. The minimum measurement error is conveniently related to a statistical test of RUMs where the measure can be checked against standard critical values for estimating multinomial distributions. Through simulations, we show that our test correctly rejects RUM for choice probabilities that violate RUM as the sample size increases.