| Fit Analyses |
Method
Observations with missing values for Y, X, Weight, or Freq variables are not used. Observations with nonpositive Weight or Freq values are not used. Only the integer part of Freq values is used.
To view or change the response distribution and link function, click the Method button in the variables dialog. This displays the dialog shown in Figure 39.3.
Figure 39.3: Fit Method Options Dialog
You can choose the response distribution and link function of the Y variables. If you choose a binomial distribution, specify either
- a Y variable with values 1 or 0 indicating success or failure
- a Y variable giving the number of successes in a certain number of trials, and a Binomial variable to give the corresponding number of trials
If you choose a power link function, specify the power value in the Power entry field.
If you select an Offset variable, it is treated as an X variable with coefficient fixed at 1.0.
You can choose the scale parameter for the response distribution. If you choose a Constant scale, specify the constant value in the Constant entry field.
With overdispersion in the model, you can specify the Quasi-Likelihood option to fit the generalized linear model using the quasi-likelihood functions.
If you choose a normal response distribution with a canonical link (identity for normal distributions), you can specify the Exact Distribution option to fit the linear model using the usual exact distributions for the test statistics.
You can specify the Fisher's Scoring option to use the Fisher's scoring method in the maximum-likelihood estimation for the regression parameters.
By default, SAS/INSIGHT software uses the Normal response distribution and Canonical link with the Exact Distribution option to perform a fit analysis for the linear model.
Copyright © 2007 by SAS Institute Inc., Cary, NC, USA. All rights reserved.