Description
Statistical models of choice among alternatives, based on random utility theory, that estimate how attributes drive decisions in stated or revealed choice data.
Discrete choice modelling explains and predicts choices among a finite set of alternatives using random utility theory: each alternative has a utility made of observed attributes and an unobserved random component, and people choose the alternative with the highest utility. Models range from multinomial logit to nested logit, probit and mixed logit, which allows preferences to vary across people. McFadden's work on transport choices made it central to econometrics. Unlike conjoint, it applies equally to revealed preference data, such as observed choices in logs, so it can model how users choose among navigation paths, plans or options. Rigour comes from correct specification of the choice set, attention to violations such as independence of irrelevant alternatives, handling preference heterogeneity, and validating out of sample.
Attributes
- Discipline
- Behavioral Science & Economics
- Loss type
- Latent
- Stage
- Strategy, Definition
- Data
- Quantitative
- Effort
- Weeks
Steps
- Define the choice situation, the set of alternatives and the attributes that vary.
- Collect stated choice data from designed experiments or revealed choices from logs, recording the full choice set.
- Specify utility functions for each alternative with attributes and user characteristics.
- Estimate a baseline multinomial logit, then test nested or mixed logit to address IIA violations and heterogeneity.
- Assess fit and validate predictions on held-out choices.
- Compute elasticities, willingness to pay or predicted shares under design scenarios.
Sources
- Discrete Choice Methods with Simulation
- Stated Choice Methods (Louviere, Hensher & Swait)
- Feature Prioritization Methods (slide deck)
Also known as
- Discrete choice analysis
- Random utility models
- Multinomial logit
- Mixed logit
Categories
- Analytics & Behaviour
- Behavioral Science & Economics