ATLASResearch
methods
Quantitative/ Analysis

Probit regression

Use a normal-link model for discrete outcomes

Probit regression maps a linear predictor to a probability through the normal cumulative distribution. Binary probit concerns two outcomes; ordered probit uses thresholds for ordered categories. Multivariate probit permits dependence between discrete responses. Coefficients are on a latent scale and are usually interpreted through predicted probabilities or marginal effects.

WHEN IT FITS

Use when a discrete outcome is naturally modelled by a latent threshold process or a probit formulation fits the substantive framework. Specify ordering, dependence and the meaning of marginal effects.

Strengths

  • Natural latent-threshold interpretation
  • Extensions can model correlated discrete outcomes

Limitations

  • Coefficients do not directly express probability changes
  • Identification and distribution assumptions matter in multivariate extensions

Know the boundary

Logit and probit coefficients are not directly comparable; neither link alone supplies causal identification.

USED ACROSS
Banking & financeBusiness & MBAPsychologyComputer scienceEducation