Analyses

Ordinal regression

Predict an ordered category, such as a 1-to-5 rating.

Edit on GitHub

When to use it

Use ordinal regression when the outcome has ordered categories and the gaps between them are not equal scores. A 1-to-4 rating is the usual case. The model has one slope per predictor and a threshold between each pair of adjacent categories.

Assumptions

The outcome is ordered. The proportional-odds idea is that one slope works at every threshold. Tensr does not test that assumption.

Running it in Tensr

Analyze → Regression → Ordinal. In chat: “Ordinal regression of rating on practice.”
Dependent is the ordered outcome. Independents need at least one predictor.
Confidence level starts at 0.95. Include constant is on the request. The dispatcher does not pass it through.

Options

Prop

Type

Reading the output

A 1–4 satisfaction rating predicted from study hours. Higher hours were built to push people up the scale, with a lot of overlap between adjacent ratings.

TermEstimatep-value
hours0.551< .001
0.0/1.00.744.351
1.0/2.00.599< .001
2.0/3.00.498.002

Ordinal regression for satisfaction, n = 96. Ordinal regression for satisfaction, n = 96. Metrics: AIC = 250.7; Log-likelihood = -121.35.

Reporting (APA 7)

Ordinal regression for satisfaction, n = 96. Report the estimate in the table. This procedure is not summarised by one p-value.

Coming from SPSS

Analyze → Regression → Ordinal.

SPSS PLUM prints thresholds and a slope, with standard errors and Wald tests. Tensr’s table uses Term, Estimate, and p-value.

A numeric score uses linear regression. Two categories use logistic regression.