Exact Poisson, proc genmod

Cruise — Sat, 19 Jan 2019 21:43:21 GMT

Hi All,

I'm running Poisson regression model. My outcome is the number of events per census block group and the predictor is the number of toxic industrial facilities per census block group. My events are very rare where maximum count is 3 and min is 0.

I wanted to run exact poisson regression model shown in the code below. But I get the error:

WARNING: There is not enough memory available to build the network for exact computations.
ERROR: The SAS System stopped processing this step because of insufficient memory.

But I'm skeptical if it's a memory issue. Because model doesn't run even at the first obs=100 observations.

Any suggestions and ideas? Thanks in advance!

PROC GENMOD DATA=MYDATA;
CLASS AGECAT(REF='10')/PARAM=REF ref=first;
MODEL N_BLOCK = X_COUNT SES AGECAT/ DIST=POISSON LINK=LOG OFFSET=LN;
exact X_COUNT;
RUN;

SAS Output

Analysis Variable : N_BLOCK
Minimum	Mean	Median	Maximum
0	0.0090629	0	3.0000000

Re: Exact Poisson, proc genmod

Cruise — Sat, 19 Jan 2019 21:44:02 GMT

Hi All,

I did my due diligence and found out that zero inflated negative binomial was recommended for the rare events with lots of zero in the data. This improves the model fit from AIC 16080 to AIC 16010. i hope this might help if someone had similar doubt i had. Any objections if it was the solution to my problem? I'll greatly appreciate if any feedbacks.

ODS OUTPUT ParameterEstimates=ZINB_PAR;
proc genmod data=MYDATA;
   CLASS AGECAT(REF='10')/PARAM=REF;
   model N_TRACT = PREDICTOR POVERTY AGECAT/ dist=ZINB offset=LN;
   zeromodel AGECAT;
   ODS TRACE ON; 
run;

@PaigeMiller @PGStats

topic Re: Exact Poisson, proc genmod in Statistical Procedures

Exact Poisson, proc genmod

Re: Exact Poisson, proc genmod