Conditional Logit Model (McFadden)
Also known as: McFadden's Choice Model, Discrete Choice Logit, Alternative-Specific Logit, Koşullu Logit Modeli
The Conditional Logit Model, introduced by Daniel McFadden in 1974, is a discrete-choice econometric model designed to explain an individual's selection among a finite set of mutually exclusive alternatives. Unlike multinomial logit, it uses covariates that vary across alternatives — such as price, travel time, or product attributes — making it ideally suited for revealed-preference studies in transportation, marketing, and labor economics.
Key highlights
- Closed-form choice probabilities that are computationally tractable for large choice sets
- Grounded in random utility theory, providing a structural economic interpretation of coefficients
- Handles alternative-specific regressors naturally, unlike standard multinomial logit
- Log-likelihood is globally concave, ensuring reliable maximum-likelihood estimation
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
Use the conditional logit model when your outcome is a choice among three or more mutually exclusive alternatives and the key predictors are characteristics of the alternatives themselves (not of the decision-maker). The model requires that the Independence of Irrelevant Alternatives (IIA) assumption holds — that is, the relative odds of choosing between any two alternatives are unaffected by the presence or attributes of other alternatives. IIA can be tested with the Hausman–McFadden test. If IIA is violated, consider nested logit or mixed logit. When individual-specific covariates matter, combine both types of regressors or switch to the mixed logit framework.
Strengths & limitations
- Closed-form choice probabilities that are computationally tractable for large choice sets
- Grounded in random utility theory, providing a structural economic interpretation of coefficients
- Handles alternative-specific regressors naturally, unlike standard multinomial logit
- Log-likelihood is globally concave, ensuring reliable maximum-likelihood estimation
- The IIA assumption is often implausible when alternatives are close substitutes (e.g., two bus lines versus driving)
- Coefficients on alternative-specific variables are constrained to be identical across alternatives, which may be restrictive
- Cannot separately identify the effect of individual-specific variables without normalization or a reference alternative
- Does not accommodate random taste variation across individuals; mixed logit is needed for that
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
What is the difference between conditional logit and multinomial logit?
Conditional logit uses regressors that vary across alternatives (e.g., price of each option) with a single coefficient vector shared across alternatives. Multinomial logit uses regressors that vary across individuals (e.g., income) with alternative-specific coefficients. In practice, many applied models include both types; software packages such as Stata's clogit and R's mlogit handle both specifications.
How do I test the IIA assumption in conditional logit?
The standard test is the Hausman–McFadden (1984) specification test: estimate the full model and a restricted model that drops one alternative from the choice set, then compare the resulting coefficient vectors. A significant chi-squared statistic suggests IIA failure. Small-sample versions and alternative tests (e.g., Small–Hsiao) are also available in R and Stata.
Can I include both individual-specific and alternative-specific variables?
Yes. Individual-specific variables (e.g., age, income) must be interacted with alternative dummies to create alternative-varying regressors, with one alternative serving as the reference category (its interaction coefficient constrained to zero for identification). This extended specification is sometimes called the mixed conditional logit in older literature, distinct from the random-parameters mixed logit.
Sources
- 1.McFadden, D. (1974). Conditional logit analysis of qualitative choice behavior. In P. Zarembka (Ed.), Frontiers in Econometrics (pp. 105–142). Academic Press.ISBN 978-0-12-776150-3
You have read it. What now?
Cite this page
ScholarGate. (2026, June 2). Conditional Logit. ScholarGate. https://scholargate.app/econometrics/conditional-logit