Model Fitting: Generalized Linear Models |
The Method Tab
You can use the Method tab (Figure 24.8) to specify aspects of the generalized linear model such as the response distribution and the link function.
You can specify the following aspects of the model:
- Response distribution
- specifies the distribution of the response variable. This corresponds to the DIST= option in the MODEL statement.
- Predict probability of
- specifies whether to model the probability of the first or last level of the response variable. This item is available only when the response distribution is binomial or multinomial. This corresponds to the DESCENDING option in the PROC GENMOD statement.
- Link function
- specifies the link function. This corresponds to the LINK= option in the MODEL statement. The following table specifies the default link function for each response distribution. Table 24.1: Default Link Functions
Distribution |
Default Link Function |
---|
binomial |
logit |
gamma |
inverse (power( - 1)) |
inverse gaussian |
inverse squared (power( - 2)) |
multinomial |
cumulative logit |
negative binomial |
log |
normal |
identity |
Poisson |
log |
When the choice of response distribution is multinomial, the choice of link functions is limited to the cumulative logit, the cumulative probit, and the cumulative complementary log-log.
- Power
- specifies the number to use for a power link function. This item is available only when the link function is the power function.
- Order response by
- specifies how to order the response variable. This corresponds to the RORDER= option in the PROC GENMOD statement.
- Classification variables parameterization
- specifies the parameterization method for the classification variables. This corresponds to the PARAM= option in the CLASS statement. The dialog box supports the GLM, effect, and reference coding schemes.
- Estimate scale parameter as
- specifies the method for estimating the dispersion parameter. This corresponds to the SCALE= option in the MODEL statement.
Copyright © 2009 by SAS Institute Inc., Cary, NC, USA. All rights reserved.