Hi,
I have to recreate analyses done in 2008 with SAS 9.1.3.
I have the datasets and the code - but it turns out that the results are slightly different.
Does anybody know whether there was a change in the default settings of proc mixed between 9.1.3 and 9.4?
Thanks for any help,
Michael
And it may be something as subtle as differences in operating system registry entries. We have had examples with identical code and data give "different" results on two power stations. The only difference between the two machines, so far as we can tell, are in the registry.
This only happens with iterative procedures (MIXED, GLIMMIX, MCMC) that we have tested. And the only solution that anyone in our IT group has come up with would be to scrub one machine down and reinstall all software.
From a purely theoretical point of view, it isn't too surprising, given representation errors accumulating over a non-linear iterative process.
But our QA unit doesn't like it.
Steve Denham
What version of 9.4 are you running? Please provide your source code so we can see what statements and options you are using.
In general, for changes to SAS./STAT in SAS 9.4, see the "What's New in SAS 9.4" page:
Each release has a link that starts "For more information, see..."
For older releases, go to your favorite search engine and do the following searches, up through the version of SAS that you are running:
(See "How old is your version of SAS?" for the relationship between Analytical product numbers and the underlying SAS version.)
"What's new" "SAS/STAT" "proc mixed" 9.2
"What's new" "SAS/STAT" "proc mixed" 9.22
"What's new" "SAS/STAT" "proc mixed" 12.1
I looked at all these, but I didn't see anything obvious.
Are you running the program on the same machine? Same Operating System? Some of the differences in calculations can be from different hardware and operating environment.
Also there are likely to be minor tweeks to algorithms in the 10 years or so since 9.1.
How different is different? Consider that values at the 6 decimal changing may not have any practical difference. If the mean price of a house in some market is reported at $185,245.123488245 for many purposes any of the decimals are not very meaningful in the real world.
And it may be something as subtle as differences in operating system registry entries. We have had examples with identical code and data give "different" results on two power stations. The only difference between the two machines, so far as we can tell, are in the registry.
This only happens with iterative procedures (MIXED, GLIMMIX, MCMC) that we have tested. And the only solution that anyone in our IT group has come up with would be to scrub one machine down and reinstall all software.
From a purely theoretical point of view, it isn't too surprising, given representation errors accumulating over a non-linear iterative process.
But our QA unit doesn't like it.
Steve Denham
Hi Steve,
even if that does not solve my problem I know now that it is - given the environment where I am working - not solvable ...
What we do is rerunning the analyses and document the differences (as they are minor this will be acceptable).
Thanks,
Michael
Build your skills. Make connections. Enjoy creative freedom. Maybe change the world. Registration is now open through August 30th. Visit the SAS Hackathon homepage.
Register today!Learn the difference between classical and Bayesian statistical approaches and see a few PROC examples to perform Bayesian analysis in this video.
Find more tutorials on the SAS Users YouTube channel.
Select SAS Training centers are offering in-person courses. View upcoming courses for: