I am running a generalized linear model with the GENLIN procedure (Analyze->Generalized Linear Model) where the response variable has a negative binomial distribution with a log link. I have clicked the radio button for "Negative binomial with log link" in the Counts section of the "Type of Model" dialog. However, I wish to have the value of the ancillary (dispersion) parameter estimated, rather than fixed at 1. I do not see an option to request that the parameter be estimated. If I paste the command to a syntax window by clicking the Paste button, I see that the parameter is fixed at 1 in the /MODEL subcommand of GENLIN, as in:
I was able to edit the command, replacing 1 with MLE, as in:
Is there a way within the Generalized Linear Model dialogs to request MLE estimation of the ancillary parameter for the negative binomial distribution?
You can request maximum likelihood estimation (MLE) of the ancillary parameter for the negative binomial distribution within the GENLIN dialogs. In the "Type of Model" dialog, click the radio button for "Custom" in the lower part of the dialog, rather than "Negative binomial with log link" in the "Counts" section. Once you've clicked the Custom button, you can choose "Negative binomial" from the "Distribution" scroll bar in the Custom area and choose "Log" from the "Link function" scroll bar in that area. Having chosen Negative Binomial, the "Parameter" box below the Distribution scroll bar will be activated. The "Specify value" radio button will be selected by default, with the default value of 1 in the Value box. Click the ."Estimate value" radio button below that value box. These steps will request MLE estimation of the ancillary parameter or the negative binomial distribution.
The GENLIN /MODEL syntax that is generated by these steps is
/MODEL ....DISTRIBUTION=NEGBIN(MLE) LINK=LOG