02-25-2012 12:25 PM
I'm not so familiar with the SAS program and therefore this question.
I'm trying to fit a model with accident data and I'm using a Negative Binomial model with the crashes as dependent variable.
The results shows a Dispersion factor of "0.3516" . Does this means that there is overdispersion? How the value must be read? With respect to 0 or to 1?
I hope you can help me. Thanks in advance!
02-25-2012 06:36 PM
According to UCLA's website: http://www.ats.ucla.edu/stat/sas/output/sas_negbin_output.htm
Dispersion - This is the estimate of the dispersion parameter. If the dispersion parameter equals zero, the model reduces to the simpler Poisson model; if dispersion is greater than zero the response variable is over-dispersed; and if disperion is less than zero the response variable is under-dispersed.
02-25-2012 10:51 PM
I noticed you stated that you are not familar with the SAS program and you have posted a question in the Enterprise Guide group. I am curious to know whether you have run a SAS program that includes PROC GENMOD or are running an Enterprise Guide task that generates your table of Maximum Likelihood Estimates you have posted. Either way, the following information may be helpful regarding PROC GENMOD.
With regards to your question for your Dispersion parameter... As you are modelling a negative binomial distribution one of the parameters in the model is k (dispersion) which can be estimated (as in your case) or you can fix it using the SCALE option on the MODEL statement - for syntax on how change this check out http://support.sas.com/documentation/cdl/en/statug/63962/HTML/default/viewer.htm#statug_genmod_sect0... . The general definition of dispersion is as Art has found above and for a more detailed definition with regards to a negative binomial model (with the theory) I suggest having a look at http://support.sas.com/documentation/cdl/en/statug/63962/HTML/default/viewer.htm#statug_genmod_sect0...
Hope this helps...
02-26-2012 01:09 PM
While I'm not familiar with the SAS program I'm using the most easy method, running a SAS task which give the table of "Analysis of Maximum Likelihood Parameter Estimates".
Many thanks for your explanation.
02-26-2012 01:46 PM
If you look at the log that task generates you should be able to see what SAS code the Enterprise Guide task generated. That should allow you to find the proper section of the manual to see how the procedure works.