Predict an ordered outcome — such as low, medium, high or a Likert rating — from one or more predictors.
Ordinal logistic regression models an outcome whose categories have a natural order but unknown spacing: satisfaction rated low, medium, high; a 5-point agreement item; disease stage I to IV. Treating such an outcome as continuous assumes the steps are equal, and treating it as nominal throws the order away. The ordinal model uses the order without assuming equal spacing.
The most common version, which StatRyx fits, is the proportional-odds (cumulative logit) model. It estimates one coefficient per predictor that applies to every cut-point of the outcome: an odds ratio above 1 means higher values of the predictor go with higher outcome categories. The model also estimates thresholds that separate adjacent categories.
StatRyx reports each predictor's coefficient, standard error, z, p, odds ratio and 95% confidence interval, the model log-likelihood and McFadden's pseudo-R². It tests the key assumption — that each predictor's effect is the same at every cut-point — with the Brant test and tells you whether it holds.
If the Brant test shows a violation, the single odds ratio is an average of effects that differ across cut-points. A multinomial model, or separate binary logistic regressions at each cut-point, can then describe the data better.
| Predictor | B | SE | z | p | OR | 95% CI |
|---|---|---|---|---|---|---|
| Waiting time (min) | -0.085 | 0.020 | -4.14 | < .001 | 0.92 | [0.88, 0.96] |
| Staff rating | 1.069 | 0.190 | 5.64 | < .001 | 2.91 | [2.01, 4.23] |
An ordinal logistic regression (proportional odds) showed that longer waiting times were associated with lower satisfaction, OR = 0.92, 95% CI [0.88, 0.96], p < .001, and better staff ratings with higher satisfaction, OR = 2.91, 95% CI [2.01, 4.23], p < .001; McFadden R² = .13, N = 200. The proportional-odds assumption held (Brant test, χ²(2) = 0.18, p = .913).
Yes. You can run it without an account (3 free analyses), and a free account adds 10 more — a one-time allowance, no payment needed. Every result includes the APA 7 write-up. Paid plans (Student from $9/month) add more analyses and Word, PDF and Excel export.
State that a proportional-odds model was used, then give each predictor's odds ratio with its 95% CI and p, a pseudo-R² and N, and the result of the proportional-odds check: "OR = 2.91, 95% CI [2.01, 4.23], p < .001."
That each predictor has the same effect at every cut-point of the outcome — for example, from low to medium and from medium to high. StatRyx tests it with the Brant test.
Yes, it is the standard model for a single Likert item as the outcome. For a scale score averaged over many items, linear regression is usually adequate.
Use ordinal when the categories are ordered and the proportional-odds assumption holds. Use multinomial when the categories have no order, or when the assumption is clearly violated.
Run a Ordinal Logistic Regression on your data →