Discrete Choice Simulation — Stated Preference Policy Modelling
Discrete Choice Simulation (Stated Preference / SP Simulation) · Also known as: stated preference simulation, SP simulation, revealed preference modelling, Ayrık Seçim Simülasyonu (Stated Preference / SP Simulation)
Discrete choice simulation is a behavioural modelling method — grounded in random utility theory formalised by Daniel McFadden in the 1970s and extended to simulation-based estimation by Kenneth Train — that estimates how individuals choose among mutually exclusive alternatives and then uses those estimated preference parameters to forecast how choice shares would shift under hypothetical policy or market scenarios. It is the dominant quantitative tool in transport demand analysis, health economics, environmental valuation, and marketing research.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Discrete choice simulation is appropriate when individuals face mutually exclusive alternatives and the goal is to understand attribute-level preferences and forecast how choice shares respond to policy or product changes. The method handles categorical, ordinal, and continuous attribute variables. It suits cross-sectional and longitudinal (panel) data structures. Stated-preference designs are used when real-world variation is insufficient or when future alternatives must be evaluated; revealed-preference data are used when actual choice records exist. The IIA assumption of the basic multinomial logit must be plausible; if alternatives share unobserved similarities, a nested or mixed logit is required. At least 100 respondents are needed; mixed logit demands more.
Strengths & limitations
- Provides behavioural micro-foundation: utility parameters are directly interpretable as trade-off rates and willingness-to-pay values.
- Enables policy counterfactuals without real-world experiments — any combination of attribute levels can be evaluated once the model is estimated.
- Mixed logit captures preference heterogeneity and correlation among alternatives, eliminating the restrictive IIA assumption of basic logit.
- Applicable across transport, health, environment, and marketing domains with the same modelling framework.
- Hypothetical bias: stated-preference responses may overstate willingness-to-pay relative to real choices.
- Mixed logit estimation requires Monte Carlo integration, which can be computationally intensive and sensitive to the number of simulation draws.
- The IIA property of basic MNL is often violated in practice; diagnosing and addressing it requires model selection skill.
- Willingness-to-pay estimates can be sensitive to the utility specification and to the assumed distribution of random parameters in mixed logit.
Frequently asked
What is the difference between stated-preference and revealed-preference data?
Revealed-preference (RP) data record actual past choices observed in real markets — for example, which transport mode a commuter actually used. Stated-preference (SP) data collect hypothetical choices from survey respondents presented with carefully designed choice scenarios. RP data reflect real decisions but are limited to attribute variation that exists in the real world; SP data allow the analyst to vary attributes beyond observed ranges and to test alternatives that do not yet exist, at the cost of potential hypothetical bias.
When should I use mixed logit instead of multinomial logit?
Use multinomial logit (MNL) as a starting point when the independence-of-irrelevant-alternatives (IIA) assumption is defensible — that is, when alternatives do not share unobserved similarities. If a Hausman-McFadden test or contextual reasoning suggests IIA is violated, or if you expect meaningful preference heterogeneity across individuals, switch to mixed logit. Mixed logit can accommodate both issues but requires specifying the distribution of random parameters and using Monte Carlo simulation during estimation.
How many simulation draws are needed for mixed logit?
Train (2009) recommends starting with 100–500 Halton or Sobol quasi-random draws for estimation. Stability should be verified by replicating estimation with a larger draw count and confirming that parameter estimates and standard errors do not change materially. Simulation noise inflates standard errors and can bias estimates if draw counts are too low; for final reported models, 1,000 or more draws are common practice.
How is willingness-to-pay calculated from a discrete choice model?
Willingness-to-pay (WTP) for an attribute is the negative ratio of that attribute's utility coefficient to the cost (price) coefficient: WTP = −(β_attribute / β_cost). In mixed logit with a random cost coefficient, this ratio is a distribution rather than a single number; the WTP-space parameterisation avoids numerical instability by directly modelling the WTP as the random parameter rather than deriving it from a ratio.
Sources
- Train, K.E. (2009). Discrete Choice Methods with Simulation (2nd ed.). Cambridge University Press. DOI: 10.1017/CBO9780511753930 ↗
- Ben-Akiva, M. & Lerman, S.R. (1985). Discrete Choice Analysis: Theory and Application to Travel Demand. MIT Press. ISBN: 978-0262022170
How to cite this page
ScholarGate. (2026, June 1). Discrete Choice Simulation (Stated Preference / SP Simulation). ScholarGate. https://scholargate.app/en/simulation/discrete-choice-simulation
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Conjoint AnalysisExperimental design↔ compare
- MicrosimulationSimulation↔ compare
- Mixed LogitEconometrics↔ compare
- MONTE-CARLO-SIMULATIONDecision-making↔ compare
- Multinomial LogitEconometrics↔ compare