Probit Regression Model
Also known as: probit regression, normit model, Probit Modeli
The probit model is a regression method for a binary (0/1) outcome that maps a linear index of the predictors through the standard normal cumulative distribution function to produce a probability. It is a classical discrete-choice alternative to logistic regression, developed in standard econometrics treatments such as Greene's Econometric Analysis (2018).
Key highlights
- Handles a binary outcome cleanly, keeping fitted probabilities inside the valid 0-1 range.
- Rooted in a latent-variable model with a normally distributed error, giving it a clear theoretical interpretation in econometrics.
- Well supported by classification diagnostics such as the ROC curve and AUC, and by marginal-effects reporting for interpretation.
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
Use the probit model when the dependent variable is binary and the goal is prediction or classification, with a reasonable sample size (at least about 50 observations). It assumes a genuinely binary outcome and predictors that are not strongly collinear. It is a natural choice in econometric settings where a normally distributed latent error is a sensible story, and it works on cross-sectional or longitudinal data. It is less suitable for very small samples or when predictors are highly collinear.
Strengths & limitations
- Handles a binary outcome cleanly, keeping fitted probabilities inside the valid 0-1 range.
- Rooted in a latent-variable model with a normally distributed error, giving it a clear theoretical interpretation in econometrics.
- Well supported by classification diagnostics such as the ROC curve and AUC, and by marginal-effects reporting for interpretation.
- Coefficients are not directly interpretable as effects on the probability; marginal effects (dy/dx) must be computed separately.
- Requires a moderate sample (about 50 or more observations) and breaks down when predictors are strongly collinear.
- Estimation relies on numerical maximum likelihood, which can fail to converge when the outcome is near-perfectly separated by the predictors.
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
How does probit differ from logistic regression?
Both model a binary outcome through a linear index, but probit passes the index through the standard normal CDF (Φ) while logistic regression uses the logistic CDF. The two usually give very similar fitted probabilities; probit is often preferred when a normally distributed latent error is the natural assumption, logit when interpretable odds ratios are wanted.
Why can't I read probit coefficients directly?
Because the coefficients act through the nonlinear normal curve, a one-unit change in a predictor does not move the probability by a fixed amount. To interpret the impact, report marginal effects (dy/dx), which translate each coefficient into its effect on the predicted probability.
How large a sample do I need?
Around 50 observations is a sensible minimum. Maximum-likelihood estimates are unstable in small samples, and convergence problems become more likely, especially when the outcome is nearly separated by the predictors.
How do I assess the fit?
Check goodness of fit with a Hosmer-Lemeshow test, and evaluate classification performance with the ROC curve and AUC. These complement the coefficient and marginal-effect estimates.
Sources
- 1.Greene, W. H. (2018). Econometric Analysis (8th ed.). Pearson.ISBN 978-0134461366
You have read it. What now?
Cite this page
ScholarGate. (2026, June 1). Probit Model. ScholarGate. https://scholargate.app/econometrics/probit-model