Definition:Generalized Linear Model

From ProofWiki
Jump to navigation Jump to search

Definition

A generalized linear model is a mathematical model in which a function $\map g \mu$ of the expectation $\mu$ of a random variable $Y$ is a linear function of one or more cause variables.




Link Function

Consider a generalized linear model which is a function $\map g \mu$ of the expectation $\mu$ of a random variable $Y$.

The function $\map g \mu$ is known as the link function.


Examples

One Variable

The generalized linear model for one cause variable $x$ is of the form:

$\map g \mu = \beta_0 + \beta_1 x$

where $\map g \mu$ is the link function.


Bernoulli Variable

Let $Y$ be a Bernoulli variable with expectation $\mu = p$.

Then the link function $\map g \mu$ of the generalized linear model for $Y$ is given by:

$\map g \mu = \map \ln {\dfrac p {1 - p} }$


Poission Distribution

Let $Y$ be a random variable with a Poisson distribution.

Let $Y$ have an expectation $\lambda$ which varies with $x$.

Then the link function $\map g \lambda$ of the generalized linear model for $Y$ is given by:

$\map g \lambda = \ln \lambda$


Probit Analysis

Probit analysis is a special case of a generalized linear model.

Note that, like least squares regression, it has been around since before the formulation of the generalized linear model.


Also see

  • Results about generalized linear models can be found here.


Historical Note

The generalized linear model was conceived and developed by Robert William Maclagan Wedderburn and John Ashworth Nelder from $1972$.


Sources