Course contentsShow
← Data Science
Lesson 674 of 2,14518. Generalized Linear ModelsFree lesson

The Probit Link

Understand the probit link as an alternative to logit for binary data using the normal CDF.

The Probit Link

What you'll learn: You'll understand the probit link function as an alternative to logit for modeling binary outcomes using the normal distribution's cumulative curve.


Core Idea

When modeling binary data (yes/no, success/failure), we need a link function to connect our linear predictor to probabilities between 0 and 1. You've seen the logit link, which uses the logistic curve. The probit link does the same job but uses the cumulative distribution function (CDF) of the standard normal distribution instead.

The probit link is written as:

probit(p) = Φ⁻¹(p)

Where Φ⁻¹ is the inverse of the standard normal CDF (also called the quantile function). It transforms probabilities into z-scores from the normal distribution.


Why Use Probit Instead of Logit?

Both logit and probit produce S-shaped curves that map linear predictors to probabilities. They're very similar in the middle but differ slightly in the tails:

  • Probit has thinner tails (based on the normal distribution)
  • Logit has slightly heavier tails (based on the logistic distribution)

In practice, they often give nearly identical results. The choice depends on context:

  • Use probit when there's a theoretical reason to assume normally distributed latent thresholds
  • Use logit for easier coefficient interpretation (odds ratios)

Interpretation Note

Unlike logit coefficients (which relate to odds), probit coefficients are harder to interpret directly—they represent changes in z-scores rather than odds ratios.


Key Takeaway: The probit link uses the normal CDF to map linear predictors to probabilities for binary outcomes—it's functionally similar to logit but assumes a normal distribution underlying the binary response.