Multinomial Logistic Regression
Also known as: multinomial logistic regression, polytomous logistic regression, softmax regression, Çok Kategorili Lojistik Regresyon
Multinomial logistic regression is a maximum-likelihood method for a nominal (unordered) dependent variable with more than two categories. Building on McFadden's 1974 treatment of qualitative choice, it gives each category its own set of coefficients relative to a reference category.
Key highlights
- Handles a nominal outcome with more than two categories in a single, coherent maximum-likelihood model.
- Produces interpretable per-category coefficients and relative-risk (odds) ratios against a reference category.
- Predicts a full probability distribution over all categories for each observation.
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 multinomial logit when the dependent variable is nominal with three or more unordered categories and you want to model or predict category membership from one or more predictors. It assumes the categories are genuinely unordered, that the independence of irrelevant alternatives (IIA) holds, and that each category has enough observations (a rule of thumb is at least ten events per predictor). A reasonable sample is needed, with roughly 100 observations or more as a starting point. If the categories have a natural order, an ordered logit is more appropriate.
Strengths & limitations
- Handles a nominal outcome with more than two categories in a single, coherent maximum-likelihood model.
- Produces interpretable per-category coefficients and relative-risk (odds) ratios against a reference category.
- Predicts a full probability distribution over all categories for each observation.
- Relies on the independence of irrelevant alternatives (IIA) assumption, which can fail when some categories are close substitutes.
- Needs a sizeable sample with enough observations in every category — sparse categories make estimates unstable.
- Treats the categories as unordered, so it ignores any natural ordering and is not suited to ordinal outcomes.
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
How is multinomial logit different from ordinary logistic regression?
Ordinary logistic regression handles a binary outcome with two categories, while multinomial logistic regression extends the same idea to a nominal outcome with three or more unordered categories by estimating a separate coefficient set for each non-reference category.
What is the independence of irrelevant alternatives (IIA) assumption?
IIA states that the relative odds between any two categories do not depend on the other available categories. It can fail when some categories are close substitutes, in which case the comparisons can be biased and an alternative model may be needed.
What is the reference category and how do I read the coefficients?
One category is fixed as the reference and its coefficients are set to zero. Each remaining coefficient describes the log-odds of its category against that reference; exponentiating it gives the relative-risk (odds) ratio.
When should I use ordered logit instead?
If the outcome categories have a natural order — such as a Likert scale or a low-medium-high rating — an ordered logit uses that ordering and is usually more efficient than treating the categories as unordered.
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 978-0127761503
You have read it. What now?
Cite this page
ScholarGate. (2026, June 1). Multinomial Logit. ScholarGate. https://scholargate.app/econometrics/multinomial-logit