I have different results in PROC GLIMMIX with the same dataset and code

[Jastrid]

Hi everyone,

I’m using PROC GLIMMIX in SAS to analyze my dataset. Two months ago, I ran my code and obtained certain results. However, the results have changed when I run the same code today with the same dataset.

I already checked the distribution of variables and quartiles (proc rank and PROC FREQ), and they appear identical.
There are no random functions (RAND()) in my code, and I used ddfm=bw in PROC GLIMMIX.
The random effect variable (random intercept / sub=area;) has not changed.
Despite all this, my coefficients and confidence intervals have changed.

Has anyone experienced something similar? What other factors could affect the results inPROC GLIMMIX, even when using the same data and code?

Thanks in advance for your help!

[mkeintz]

Whenever someone raises the problem of "different results ... from the same code and data", I would want to see the log's of both runs, to confirm that SAS also thinks it is the same code.

And, for those in this community who have familiarity with PROC GLIMMIX (I am not), you might benefit from showing the logs from both runs, as well as the results from each.  Help the community help you.

[Sa_S]

Hi @Jastrid 

For this issue, you might be best served by contacting Technical Support and requesting SAS Content Assessment Tool support.  Here is a link for your convenience:

https://support.sas.com/en/technical-support.html#contact 

Thank you for using SAS!

[PaigeMiller]

Another possibility is that the data changed in between your two runs of GLIMMIX. I know you said it didn't change, but it is indeed something that needs to verified. You can't just say it, you have to check and convince yourself the data didn't change.

But, as stated above, you need to provide us with the log of the two different runs of GLIMMIX and the output from both runs.

[SteveDenham]

I have seen some behavior such as this. Recall that the optimization method is non-linear and that the log likelihood values are defined as hexadecimal during calculations. That leads to truncation and semi-chaotic behavior if the program is run on a different machine or even on the same machine with a different operating system. Also, check to see if your version of SAS has changed since the original run. There were some substantive changes made to GLIMMIX going from previous versions to SAS v9.4TS1M8 that may be contributing to this. See this problem Note: https://support.sas.com/kb/70/094.html 

SteveDenham