Take the 2-minute tour ×
Cross Validated is a question and answer site for people interested in statistics, machine learning, data analysis, data mining, and data visualization. It's 100% free, no registration required.

I want to fit a negative binomial model for my count data. I am wondering what the differences for the computation of these three PROCs are in SAS.

share|improve this question
2  
On the face of it, this question ought to have a routine answer in the SAS manuals and help files. Is there something in those references that you need help with? –  whuber Aug 10 '12 at 15:26
    
@whuber, I just a little bit confused on how the manual describes this. They seems to perform the same task and I am not sure if the calculation would be different. –  Seen Aug 10 '12 at 18:49
    
@Seen I think you would get a more helpful answer if you could give more details in your question. There are many possible differences between these procedures, if you explain what is confusing or what specifically you are interested in, we can give a more focused answer. –  Joshua Aug 11 '12 at 2:40

1 Answer 1

There are many overlapping procedures in SAS. If all you want to do is fit a negative binomial model, any of those should be fine. Differences arise in that each of those procedures has additional abilities. For example, proc genmod has flexible residual correlation structures, proc countreg offers bounds and constraint options, proc fmm fits finite mixture models, which are a very flexible class of models but it has less post estimation capacities built in.

If a procedure does everything you want, there is probably little reason to use a different one. However if it does not or it is difficult, there may be. For example proc genmod has an estimate statement whereas proc countreg does not, although one could still get estimates by output the parameter estimates and computing them manually.

share|improve this answer
    
Well I just want to find out a starting point to choose procs in SAS...(However, proc genmod can do zero inflated negative binomial models as well) –  Seen Aug 11 '12 at 2:40
    
@Seen good point about proc genmod (will edit my answer). I think in part for choosing procs in SAS, to the extent that it does what you want, it probably does not matter much. I wonder if it would be reasonable to create some sort of comparison table. Every SAS proc does so much that a good, up-to-date comparison of all or most features would be far from trivial, but that does not mean it could not be undertaken. –  Joshua Aug 11 '12 at 2:49
    
that was what I was looking for in the SAS manual. I like the UCLA stat computing website as they have tutorials using the procs I mentioned above as different methods for count data modelling, while I don't see the comparisons among them. –  Seen Aug 11 '12 at 2:55
1  
For what it is worth, in the UCLA stat consulting group, we tend to use genmod. We talked about demonstrating the same analysis in various procs (or other functions for other packages) but the thing is there is often many means to an end. Other examples in SAS are proc reg and proc glm which overlap substantially, proc glimmix, proc mixed, and proc nlmixed also come to mind. They all started differently so they each have their own quirks but they have all been extended so they overlap. –  Joshua Aug 11 '12 at 3:14
    
+1, it's good to have you on-board here, @Joshua, you're contributing a lot of value. It's a shame you seem to have joined during the slowest point of the year, I suppose. Don't let that discourage you. –  gung Aug 17 '12 at 2:39

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.