Ordered Choice Models: Ordered Logit and Probit

Graduate Depth 77 in the knowledge graph I know this Set as goal
ordered-choice ordinal logit probit

Core Idea

Ordered logit and probit apply when the dependent variable has more than two ordered categories (e.g., satisfaction from 1-5). These models assume a latent continuous variable with threshold values determining the observed ordinal outcome.

Explainer

From your study of logit and probit, you know how to model binary outcomes: a latent index y* = Xβ + ε crosses a single threshold to produce a 0 or 1. Ordered choice models extend this logic to outcomes with more than two ordered categories. The key word is *ordered*: the categories have a natural ranking (strongly disagree < disagree < neutral < agree < strongly agree; or bond ratings AAA > AA > A > ...) but the distances between categories are not assumed to be equal. You can't treat these as OLS outcomes because forcing equal spacing on ordinal categories imposes false precision — the gap between "disagree" and "neutral" need not equal the gap between "neutral" and "agree."

The architecture is a latent variable model with multiple thresholds. There is still an unobserved continuous variable y* = Xβ + ε representing the underlying propensity (satisfaction, creditworthiness, pain severity). But now there are J−1 thresholds μ₁ < μ₂ < ... < μ_{J-1} that partition the real line into J intervals. The observed outcome is category j whenever μ_{j-1} < y* ≤ μ_j. The thresholds are estimated alongside the β coefficients by maximum likelihood. Because the thresholds are free parameters, the model lets the data determine how wide each "band" is, rather than imposing equal spacing as OLS would implicitly require.

Interpretation requires care. As in binary logit/probit, the coefficients β tell you the direction of effect: a positive β_k means that increasing x_k shifts y* upward, making higher categories more probable. But the marginal effect on any specific category is non-monotone — a positive shift can increase the probability of the highest category, decrease the probability of middle categories, and increase the probability of the lowest category simultaneously, depending on where the probability mass is concentrated. This is why you should compute marginal effects for each category rather than simply citing the coefficient. For ordered logit, the latent error follows a logistic distribution (so the cumulative probabilities use the logistic CDF); for ordered probit, it follows a standard normal. The choice rarely matters much in practice, but both are estimated by maximum likelihood, which you already know how to work with.

The proportional odds assumption (sometimes called the parallel regression assumption) is a key identifying restriction in ordered logit: the β coefficients are the same across all thresholds — only the intercepts differ. This means a one-unit increase in x shifts the latent index by the same amount regardless of which threshold you're comparing. This is a testable restriction, and when it fails, you may need a generalized ordered logit that allows the slopes to vary across thresholds. Think of ordered choice models as a principled bridge between binary discrete models (too few categories) and OLS (assumes cardinal, continuously distributed outcomes) — they occupy the middle ground that much real survey and administrative data actually inhabits.

Practice Questions 5 questions

Prerequisite Chain

Counting to 10Counting to 20Understanding ZeroThe Number ZeroCounting to FiveOne-to-One CorrespondenceCombining Small Groups Within 5Addition Within 10Addition Within 20Two-Digit Addition Without RegroupingTwo-Digit Addition with RegroupingAddition Within 100Repeated Addition as MultiplicationMultiplication Facts Within 100Division as Equal SharingDivision as Grouping (Measurement Division)Division: Grouping (Repeated Subtraction) ModelDivision: Fair Sharing ModelDivision as Equal SharingDivision as GroupingBasic Division FactsDivision Facts Within 100Two-Digit by One-Digit DivisionDivision with RemaindersRemainders and Quotients in DivisionDivision Word ProblemsIntroduction to Long DivisionFactors and MultiplesPrime and Composite NumbersEquivalent FractionsRelating Fractions and DecimalsDecimal Place ValueReading and Writing DecimalsComparing and Ordering DecimalsAdding and Subtracting DecimalsMultiplying DecimalsDividing DecimalsDividing FractionsMixed Number ArithmeticOrder of OperationsInteger Order of OperationsVariable ExpressionsCombining Like TermsOne-Step EquationsTwo-Step EquationsSolving Multi-Step EquationsEquations with Variables on Both SidesAngle Pairs: Complementary, Supplementary, and VerticalParallel Lines and TransversalsCorresponding AnglesAlternate Interior AnglesTriangle Angle Sum TheoremExterior Angle TheoremTriangle Inequality TheoremSimilar Triangles: AA SimilaritySimilar Triangles: SSS and SAS SimilarityProportions in Similar TrianglesRight Triangle Trigonometry IntroductionTrigonometric Ratios ReviewRadian MeasureConverting Between Degrees and RadiansThe Unit CircleGraphing Sine and CosineGraphing Tangent and Reciprocal Trigonometric FunctionsDerivatives of Trigonometric FunctionsAntiderivativesIndefinite IntegralsBasic Integration RulesRiemann SumsDefinite Integral DefinitionProbability Density Functions and Continuous DistributionsCumulative Distribution FunctionsContinuous Random VariablesNormal DistributionClassical OLS Assumptions (Gauss-Markov)Multiple RegressionLogit and Probit Models for Binary OutcomesOrdered Choice Models: Ordered Logit and Probit

Longest path: 78 steps · 462 total prerequisite topics

Prerequisites (2)

Leads To (0)

No topics depend on this one yet.