topic Proc glm (changes or updates) in Statistical Procedures

Proc glm (changes or updates)

stanmarciano — Sun, 09 Aug 2020 13:18:48 GMT

This is not a technical/code related question (more of a general question related to proc GLM). I had some codes written for a project a couple of years back (2 factor anova model). I used Proc GLM. A simple version of the code is below:

proc glm data = mydata;
class Independent_Var1 Independent_Var2;
model Dependent_Var=Independent_Var1|Independent_Var2/ss3 solution;
lsmeans Independent_Var1|Independent_Var2;
run;

I recently dug out this long forgotten project and started working on it again. I tried to replicate my previous findings (dataset is exactly the same). However, I am not getting my previous results back. There are slight changes in the F statistic and p-values. For example for Independent_Var1, the original p-value was 0.08 and I now get 0.12. Again, with the same dataset, I would expect to get my previous results.

Can anyone help me understand why I am getting somewhat different results. Have there been any changes or updates to Proc GLM over the last couple of years or so? Is there a resource to track changes to the proc's? I do keep my data and codes organized so I am confident that I am using my original data as well as the code.

Thanks in advance.

EDIT: During this time, I have gone from using SAS 9.2 to SAS 9.4.

Re: Proc glm (changes or updates)

PaigeMiller — Sun, 09 Aug 2020 13:45:36 GMT

There are three possible reasons why the results might have changed

The data is different
The code is different
PROC GLM has changed

You can do the research to confirm that #1 and #2 have not happened. Simply saying they haven't changed (and using bold letters) doesn't convince me. If you conclude that #3 is the only possibility, we would need to see the results before and the results after.

Re: Proc glm (changes or updates)

Rick_SAS — Mon, 10 Aug 2020 12:49:48 GMT

If you post your actual code and data, we can probably help more. Since the F statistic for the ANOVA model was developed 100 years ago, I don't think it is likely that the computation changed between SAS 9.2 and 9.4.

Re: Proc glm (changes or updates)

SteveDenham — Mon, 10 Aug 2020 13:12:04 GMT

If the method was iterative (say PROC MIXED), I would also look at hardware and OS differences. But to find this for PROC GLM seems odd. Are you sure there is no data preprocessing in the original project that has been dropped out in the current project?

SteveDenham