Binomial Logistic Regression — odds ratios + ROC AUC + classification
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 the log-odds of a binary outcome as a linear combination of one or more predictors.
Reports coefficients (β with SE, z, p, 95% CI), per-predictor odds ratios with 95% CI, model fit (McFadden's pseudo-R², AIC, BIC, log-likelihood), Hosmer-Lemeshow goodness-of-fit OR modern calibration intercept-slope-ICI (TRIPOD-AI 2024), ROC AUC for discrimination, and a classification table at the default p ≥ 0.5 cut-point. The engine auto-checks separation, multicollinearity (VIF), linearity-of-logit for continuous predictors, and influential observations.
Worked example
Which factors predict whether a student passes?
Pass/fail (binary) was regressed on study hours and prior GPA for 200 students; logistic regression reports odds ratios.
Each extra study hour raised the odds of passing by 38% (OR = 1.38, 95% CI [1.23, 1.56], p < .001); prior GPA also predicted passing (OR = 2.12). The model discriminated well, AUC = .78.
A logistic regression showed study hours predicted passing (OR = 1.38, 95% CI [1.23, 1.56], p < .001), as did prior GPA (OR = 2.12); model AUC = .78.
Try it yourself: Load this ready-made sample and follow the run above.
When to use it
- Case-control study (retrospective)Heart-disease cases (n = 200) vs matched controls (n = 200); predictors: age, cholesterol, smoking, BMI.
- Prospective cohort with binary outcome1,000 employees followed for 2 years; outcome = burnout (yes/no); predictors = baseline stress, workload, support.
- Cross-sectional classification / prediction modelPredict pass/fail on a certification exam from study hours, practice scores, and years of experience in 800 historical examinees.
When NOT to — use instead
- Continuous outcomeLogistic is for binary outcomes. → Linear Regression \u2014 OLS (continuous y + predictors)
- 3+ unordered outcome categoriesLogistic is binary. → Multinomial Logistic \u2014 nominal 3+ categories (per-class OR vs reference)
- Ordered (3+) outcome categoriesFor ordinal outcomes use ordinal logistic regression with proportional-odds assumption. → Ordinal Logistic \u2014 ordered categories + Brant proportional-odds test
- Paired / matched binary outcomeUse conditional logistic regression or McNemar's test. → McNemar's Test (paired 2\u00d72)
Hypotheses
Parameter tested: vector of population log-odds coefficients β
Assumptions (and what to do if they fail)
Check: min(n_y0, n_y1) / k.
If violated: Insufficient events per predictor inflate OR uncertainty and can flip signs.
Check: AUC value.
If violated: AUC ≈ 0.5 means the model has no discriminative power beyond chance.
Check: Hosmer-Lemeshow p-value.
If violated: Poor calibration (p < .05) means predicted probabilities differ from observed event rates.
Check: Max Cook's D > 1 = severe; > 4/n = noteworthy.
If violated: Influential cases can flip OR signs in small samples.
Check: Max VIF.
If violated: High VIF inflates OR standard errors and CIs.
Ready to run a Binomial Logistic Regression — odds ratios + ROC AUC + classification on your own data?
Guided setup, automatic assumption checks, effect sizes, figures and an APA write-up.
Run this test →