Proc GENMOD doesn't print criteria for assessing goodness of fit

Reply
Frequent Contributor
Posts: 75

Proc GENMOD doesn't print criteria for assessing goodness of fit

Here's the log of the program

1120  proc genmod data=trial8;

1121  class aid;

1122  model w2pa = f_w2pa w1pa f_w1pa/dist=poisson link=log type3;

1123  repeated subject=aid/type=ind;

1124  run;

NOTE: Class levels for some variables were not printed due to excessive size.

NOTE: Algorithm converged.

NOTE: Algorithm converged.

NOTE: PROCEDURE GENMOD used (Total process time):

      real time           0.06 seconds

      cpu time            0.06 seconds

I think the program looks and runs fine, yet it doesn't print criteria for assessing goodness of fit (Deviance, Pearson Chi-squared, etc.) It does print QIC, QICU.

Is there some thing wrong with this? Can I force it to print G-o-F?

In here http://support.sas.com/kb/22/630.html one sentence in the last paragraph says: "Note that the Pearson and deviance statistics, if displayed, apply only to the initial model that begins the GEE estimation, not to the final GEE model". Does it also mean that those statistics may not be displayed? If so, in which cases are they printed and which are they not?

Respected Advisor
Posts: 2,655

Re: Proc GENMOD doesn't print criteria for assessing goodness of fit

Posted in reply to NonSleeper

I guess the simple answer is "No you can't force it to print G-o-F," at least for the criteria you mention.  There are methods for assessing goodness of fit, using the ASSESS statement, and the link you provided gives some info on the QIC.  I assume that all of this has to do with non-independence when a repeated model is fit, and the use of quasi-likelihoods.

You could shift to a conditional model (rather than the marginal GEE) by moving to GLIMMIX, and fitting the clustering by subject as a G-side effect using METHOD=QUAD or =LAPLACE, which will give several ICs, if I recall correctly.

Steve Denham

Ask a Question
Discussion stats
  • 1 reply
  • 204 views
  • 4 likes
  • 2 in conversation