Ordinal Regression

Description

Models Likert and other ordered-category responses as ordinal, not interval, estimating how predictors shift the odds of higher response categories.

Likert items are ordered categories: the distance between 'agree' and 'strongly agree' is not known to equal that between 'neutral' and 'agree'. Treating them as interval numbers and comparing means can produce misleading effects, including spurious differences or even sign reversals, as shown by Liddell and Kruschke (2018). Ordinal regression, typically the proportional odds or cumulative link model, models the probability of responding at or below each category threshold as a function of predictors. Coefficients express how predictors shift the latent response, interpretable as odds ratios. Assumptions such as proportional odds are tested, and mixed-effects or Bayesian versions handle repeated measures. Its rigour lies in matching the model to the measurement level of the data.

Attributes

Loss type
Latent
Stage
Evaluation
Effort
Days

Steps

  1. Confirm the outcome is an ordered categorical variable, such as a single Likert item or ordered rating, and inspect the response distribution.
  2. Specify a cumulative link model (logit or probit) with relevant predictors such as design variant, segment or task.
  3. Fit the model in a suitable tool (for example R ordinal, statsmodels OrderedModel, SPSS PLUM or brms) and add random effects for participants if responses are repeated.
  4. Test the proportional odds assumption, for example with a Brant test or by comparing to a partial proportional odds model.
  5. Report odds ratios with confidence or credible intervals and predicted probabilities for each category, rather than mean scores.
  6. Visualise predicted category distributions by condition to communicate results to stakeholders.

Sources

  • Statistics for Behavioral Sciences
  • Discovering Statistics via SPSS (Field)
  • Survey Methodology (Groves)
  • scipy.stats
  • pingouin
  • mord
  • PyMC
  • bambi
  • statsmodels
  • matplotlib
  • seaborn
  • SPSS REGRESSION
  • PLUM
  • LOGISTIC REGRESSION procedures; Hayes PROCESS macro v4
  • Liddell & Kruschke (2018) Analyzing ordinal data with metric models: What could possibly go wrong? Journal of Experimental Social Psychology
  • The Inquiry Series, Vol III, Ch 42
  • The Inquiry Series, Ch 42 — Ordinal Logic — Treating Likert & Survey Data Correctly
  • The Inquiry Series, C2 — Quantitative UX Analysis in Python — Descriptive to Inferential
  • The Inquiry Series, C8 — SPSS for Regression, Moderation & Mediation
  • Quantitative User Experience Research — Chris Chapman

Also known as

  • Ordinal logistic regression
  • Proportional odds model
  • Cumulative link model
  • Ordinal regression
  • Proportional odds models
  • Non-parametric Likert analysis
  • Ordinal Data Analysis

Categories

  • Analytics & Behaviour
  • Statistics & Modelling