A generalized linear model relates the expectation value 
of a response variable to a linear predictor 
through a specified link function 
,
 |
The conditional distribution of the response is chosen from an exponential family. The link function and response distribution are parts of the model and determine how changes in the predictors affect the mean response.
Ordinary linear regression with normal distribution errors and the identity link is one example. Binomial responses with the logit transformation as link and Poisson distribution responses with the logarithmic link are two other common examples. A generalized linear model is distinct from a general linear model, which is a linear model for a continuous response vector.
