PROC GENMOD

[AudreyM]

When i run my Proc Genmod, i have the message "ERROR: Floating Point Zero Divide". 

PROC GENMOD DATA = _1;
  CLASS TRTP1-TRTP3 country1-country21 NEXTGR11-NEXTGR12 FEVPGR11-FEVPGR13;
  MODEL AVAL = TRTP1-TRTP3 country1-country21 NEXTGR11-NEXTGR12 FEVPGR11-FEVPGR13 /
  OFFSET=LNFUDUR2 DIST=negbin LINK=log;
  BAYES SEED=22022021 COEFFPRIOR=uniform THIN=100 OUTPOST= dataset2 NMC=100000;
RUN;

i am trying to find solutions and i have seen that if i modify AVAL values by AVAL+1 then the model runs (it means that instead of having a range of values between 0 and 5, it is now between 1 and 6).

Is that an acceptable solution (i have tried and a small number of data and i have a slight variation of results)?

Thanks

[Ksharp]

I saw you used "OFFSET=LNFUDUR2" , if AVAL is a count variable(or Poisson distribution),
you can't have 0 be the value of AVAL ,it must be a positive integer number.

[StatDave]

You might try adding the NOLOGNB option in the MODEL statement. If that, or your solution, results in an estimated negative binomial dispersion parameter near zero, then you probably should just use the Poisson distribution. Or, if you don't need a Bayesian analysis, you could try PROC GLIMMIX instead. By the way, if your groups of same-named predictors are dummy variables that you created to represent multi-level categorical variables, then you can instead simply put the individual multi-level variable names in both the CLASS statement and the MODEL statement. For a variable in both statements, the CLASS statement creates dummy variables for you and puts them in the model.