Course contentsShow
← Data Science
Lesson 672 of 2,14518. Generalized Linear ModelsFree lesson

The Identity Link

Learn the identity link for normal distributions and how it relates to standard linear regression.

The Identity Link

You'll learn: How the identity link function connects linear regression to the broader framework of generalized linear models.

What is the Identity Link?

In the previous lesson, you learned that a link function bridges the linear predictor (the sum of your coefficients times predictors) to the expected value of your response variable. The identity link is the simplest possible link function: it does absolutely nothing!

Formally, the identity link is written as:

g(μ) = μ

This means the link function just returns whatever you put into it, unchanged. No transformation happens at all.

How It Relates to Linear Regression

When you use a normal (Gaussian) distribution for your response variable and pair it with the identity link, you get exactly what you've been doing all along: standard linear regression.

Here's why: in ordinary linear regression, you directly model the mean of Y as:

E(Y) = β₀ + β₁X₁ + β₂X₂ + ...

There's no transformation between the linear combination of predictors and the expected value—they're identical. That's the identity link in action!

An Analogy

Think of link functions as language translators. Most distributions need a translator to convert between the "linear world" (which can produce any real number) and the "distribution world" (which might only allow positive numbers or probabilities). The identity link is like speaking the same language on both sides—no translation needed because the normal distribution already allows any real number.

Key Takeaway: The identity link g(μ) = μ leaves the linear predictor unchanged and is the natural link for normally distributed responses, making standard linear regression a special case of generalized linear models.