Multinomial Logistic — nominal 3+ categories (per-class OR vs reference)
VerifiedAdvanced & specialized
Run this test straight away on a free built-in teaching dataset — no data of your own needed — or bring your own. Either opens the guided workspace: variable setup, assumption diagnostics, results with effect sizes and confidence intervals, figures, and APA-ready reporting.
Loading teaching datasets…
Or use your own dataset
Loading your datasets…
Models a NOMINAL (unordered) categorical outcome with 3+ classes via the generalised logit model: log(P(Y=c)/P(Y=ref)) = β₀_c + Σ β_ic x_i for each non-reference class c.
For each non-reference class reports per-predictor coefficients, odds ratios with 95% CI (interpreted relative to the reference class), model fit (likelihood-ratio χ², McFadden R², AIC, BIC), per-class predicted probabilities, and classification accuracy at the argmax cut-point. Pre-checks IIA (independence of irrelevant alternatives) and multicollinearity.
Worked example
What predicts which of three transport modes a commuter chooses?
Commute mode (car / bus / bike — unordered categories) was regressed on distance and income with multinomial logistic regression (car as the reference).
Greater distance lowered the odds of choosing bike over car (relative-risk ratio = 0.46 per SD) — longer trips favoured driving; income also shaped mode choice.
A multinomial logistic regression showed distance lowered the odds of cycling versus driving (RRR = 0.46 per SD), with income also shaping mode choice.
Try it yourself: Load this ready-made sample and follow the run above.
When to use it
- Nominal categorical outcome (3+ unordered classes) + predictorsBrand choice (A/B/C/D, A as reference) ~ price + advertising + age (n = 800 consumers).
- Diagnostic / triage classification with multiple categoriesTriage diagnosis (cardiac / pulmonary / metabolic / other) at ED admission ~ symptoms + biomarkers + vitals (n = 1200).
When NOT to — use instead
- Binary outcomeUse logistic regression — multinomial is overkill for k = 2. → Binomial Logistic Regression \u2014 odds ratios + ROC AUC + classification
- Ordered (3+) outcome categoriesUse ordinal logistic — preserves the ordering and is more parsimonious. → Ordinal Logistic \u2014 ordered categories + Brant proportional-odds test
- Continuous outcomeUse linear regression. → Linear Regression \u2014 OLS (continuous y + predictors)
- Count outcomeUse Poisson / negative-binomial regression. → Poisson Regression \u2014 IRR for count outcomes (auto-flags overdispersion)
Hypotheses
Parameter tested: matrix of class-specific coefficients β (k_predictors + 1 rows × k_categories − 1 columns)
Assumptions (and what to do if they fail)
Check: min(per_class_n) / k.
If violated: Sparse classes inflate class-specific OR uncertainty; can produce extreme ORs.
Check: Accuracy − baseline > 0.
If violated: If model accuracy ≤ baseline, predictors do not improve classification.
Check: Count of |std residual| > 2.
If violated: Outliers can flip per-class OR signs in small samples.
Check: Max VIF.
If violated: High VIF inflates per-class OR standard errors.
Ready to run a Multinomial Logistic — nominal 3+ categories (per-class OR vs reference) on your own data?
Guided setup, automatic assumption checks, effect sizes, figures and an APA write-up.
Run this test →