NMST432 Advanced Regression Models

Link functions for binary data


Back to course page


The logistic link (logit)

Based on the distribution function of the standard logistic distribution.

\[g(\mu)=\frac{\mu}{1-\mu}\]

\[g^{-1}(\eta)=\frac{\exp(\eta)}{1+\exp(\eta)}\]

Plot of the inverse logistic link function:

plot of chunk unnamed-chunk-1

The probit link

Based on the distribution function of the standard normal distribution.

\[g(\mu)=\Phi^{-1}(\mu)\]

\[g^{-1}(\eta)=\Phi(\eta)\]

Plot of the inverse probit link function:

plot of chunk unnamed-chunk-2

The cauchit link

Based on the distribution function of the standard Cauchy distribution.

\[g(\mu)=\frac{1}{\pi}\arctan(\mu)+\frac{1}{2}\]

\[g^{-1}(\eta)=\tan[\pi(\eta-\frac{1}{2})]\]

Plot of the inverse cauchit link function:

plot of chunk unnamed-chunk-3

The complementary log-log link

Related to the distribution function of the extreme value (Gumbel) distribution. The link is not symmetric.

\[\quad g(\mu)=\log(-\log(1-\mu))\]

\[g^{-1}(\eta)=1-\exp(-\exp(\eta))\]

Plot of the inverse complementary log-log link function:

plot of chunk unnamed-chunk-4

Comparison of inverse link functions

plot of chunk unnamed-chunk-5


Back to course page