In regression analysis, one of a wide class of model in which the fitted value is a transformation of a linear predictor and the frequency distribution is not necessarily the normal distribution. Apart from the standard linear regression model (see regression analysis), the most important cases are (a) for integer counts, the logarithmic transformation and the Poisson distribution, and (b) for proportions, the logistic transformation and the binomial distribution.
GLMs may be used in regression analysis where inspection of residuals indicates that the distribution is other than the normal distribution, and may also be used to analyze contingency tables.
The analysis uses the method of maximum likelihood, solved by iterative use of weighted least squares estimation.
Subjects: Probability and Statistics — Computing.