Stochastic Choice#
Analyze probabilistic choice data with random utility models. Based on Chambers & Echenique (2016) Chapter 13.
Random Utility Model#
Function: fit_random_utility_model(log)
In a Random Utility Model (RUM), utility has a deterministic and stochastic component:
where \(V_i\) is the systematic utility and \(\varepsilon_i\) is a random shock.
Choice Probability:
Logit Model#
Function: fit_luce_model(log)
With Type I Extreme Value errors, choice probabilities follow the logit form:
This is equivalent to Luce’s choice axiom.
McFadden’s Axioms#
Function: test_mcfadden_axioms(log)
RUM-consistent choice must satisfy:
Regularity: \(P(i | A) \geq P(i | B)\) if \(A \subseteq B\) and \(i \in A\)
IIA (for logit): \(\frac{P(i | A)}{P(j | A)} = \frac{P(i | B)}{P(j | B)}\) if \(i, j \in A \cap B\)
McFadden’s Theorem
Choice probabilities are consistent with RUM if and only if they satisfy a set of linear inequalities (Block-Marschak conditions).
Reference: McFadden (1974), Block & Marschak (1960), Chambers & Echenique (2016) Ch. 13