Which model makes assumptions about ordering of categories and is used with ordinal data?

• A. Adjacent category model
• B. Proportional odds model
• C. Multinomial logistic regression
• D. Continuation-ratio model

Answer: (B)

When you have an ordinal outcome, the model that explicitly uses the order of categories and assumes the same predictor effect across all category thresholds is the proportional odds model. It builds on the cumulative odds for being at or below each category: logit[P(Y ≤ k)] = α_k + βᵀX for each threshold k, with the same β across all thresholds. This means the effect of a predictor is constant no matter where you are in the ordered categories, which directly relies on the ordering of the outcome. Because probabilities are modeled cumulatively, the natural ranking of the categories is integral to the approach, making it the standard choice for ordinal data when you want a single set of covariate effects that apply across all levels.

Other models approach ordinal data differently and don't impose this same proportional effect across all thresholds. An adjacent-category model compares neighboring categories rather than using cumulative thresholds; a continuation-ratio model follows a sequential conditioning process, which reflects a different way of exploiting order. Multinomial logistic regression treats the outcome as unordered, ignoring the natural ranking entirely.