← Statistics for Mathematical Biology

Generalised linear models

Generalised linear models extend linear regression to responses whose distributions are not well represented by constant-variance normal errors.

They relate the mean response \(\mu=E[Y]\) to predictors through a link function:

\[g(\mu)=\beta_0+\beta_1x_1+\cdots+\beta_px_p.\]

Examples

Logistic regression uses a Bernoulli or binomial response and the logit link:

\[\log\frac{p}{1-p}=\beta_0+\beta_1x_1+\cdots.\]

Poisson regression is commonly used for count data and typically uses a logarithmic link.

Key idea. GLMs preserve a regression structure while adapting the probability distribution and link function to the type of biological response.