topic Re: poisson predicted probability for counts in Statistical Procedures

poisson predicted probability for counts

Salih — Mon, 09 Jul 2018 19:01:30 GMT

I have been trying to get predicted probability for each count. I am putting code "predicted=" after model statement but that is giving me the mean of poisson. How can I get predicted probability based on poisson for each count and graph it?

data world;
input scores @@;
datalines;
4 3 3 3 2 4 2 0 3 3 3 2 4 2
1 1 1 3 7 3 2 2 2 5

..........
;

proc genmod data=world;
model scores= /dist=poisson link=log;
output out=new predicted=predictedscores;
run;
quit;

Re: poisson predicted probability for counts

PGStats — Mon, 09 Jul 2018 19:38:02 GMT

If you have a SAS/ETS license, proc countreg provides option prob= in the output statement to compute those probabilities.

Re: poisson predicted probability for counts

Salih — Mon, 09 Jul 2018 20:30:37 GMT

I don't have SAS/ETS. I am using SAS studio university edition. I tried countreg, didn't work.

Re: poisson predicted probability for counts

sld — Tue, 10 Jul 2018 02:57:21 GMT

I'd say that your code is doing exactly what you've specified: you've specified an intercept-only model, so predictions for all observations will be equal to the intercept.

I don't have an idea about what you would rather be doing, but I'm guessing it might be more along the lines of your other question

https://communities.sas.com/t5/SAS-Statistical-Procedures/proc-genmod-graphics-for-count-data-model-fit-assessment/m-p/476567#M24805, where you've gotten two good solutions.

Re: poisson predicted probability for counts

PGStats — Tue, 10 Jul 2018 03:27:20 GMT

Diy then,

data probs;
set new;
prob = pdf("Poisson", scores, predictedScores);
run;