Hello, I am new to genmod and I am trying to calculate RR with an modified Poisson approach using robust error variance (by using the repeated statement and the subject identifier).
I have two questions:
1.) my output does not show me the output of the exp option on the estimate statement completely and not in the way as it was in other examples I am following. The xp estimate is missing, however L'Beta estimate and alpha etc are given ind the specific row. And I have no idea why? Can anyone help?
2.) Secondly, in other examples I found to ways to state the estimate statement "estimate 'beta' smoking 1/exp" and "estimate 'beta' smoking 1-1/exp" where exactly is the difference?
I will add my model below unfortunately the output cannot be displayed in a readable manner.
Thank you in advance!!
(1) since the actual statistical results (e.g. p-value) are the same for beta and L'beta, sas only shows them on one line.
(2) in most sas procs that allow
estimate 'smoking' smoking 1 / exp;
syntax, you only need the single 1. the use of 1 -1 comes from situations where the whole matrix was required (like what you see when you use a class variable and sas prints out the contrast matrix). i personally find the matrix version confusing (too hard to count the zeros when you have a lot of categories) and like the syntax i used above, which is clear. also, i think sas will even react to the 1 -1 syntax as you show it with a warning that there is extra stuff in the estimate statement.
Thank you very much Ellen for your explanation! That helped a lot!!
Can I ask you another question. I learned that a modified modified Poisson regression analysis gives better result than a normal Poisson regression analysis for longitudinal data, which tends to give conservative confidence intervals. SO, this is what I am trying to do with SAS. I tried this on my data, however, I always get the same warning:
WARNING: The generalized Hessian matrix is not positive definite. Iteration will be terminated.
ERROR: Error in parameter estimate covariance computation.
ERROR: Error in estimation routin
I f I run the analysis without the modification of:
repeated subject=id/type=ind;
or
repeated subject=id/type=unstr;
it works fine and I get results.
So my question is - how big could the bias on CIs actually be and secondly how can I overcome this warning?
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software