ABSTRACT
Note: For the glm() function in R, the available links for each distribution are listed. The following links are available for all distributions in SAS: identity, log, or power(λ) (where λ is specified by the user). For dichotomous outcomes, complementary log-log (link=cloglog), logit (link=logit), or probit (link=probit) are additionally available. For multinomial distributed outcomes, cumulative complementary log-log (link=cumcll), cumulative logit (link=cumlogit), or cumulative probit (link=cumprobit) are available. Once the family and link functions have been specified, the variance function is implied (with the exception of the quasi family). In SAS, overdispersion is implemented using the scale option to the model statement. To allow overdispersion in Poisson, binomial or multinomial models, use the option scale=deviance; the additional aggregate option is required for the binomial and multinomial. Any valid link listed above may be used.
