statsoft53@aol.com (StatSoft53) wrote:

>You may consider a software package that takes better advantage of the >hardware on which it is running. I ve just timed STATISTICA for Windows: >the problem you have described (a multiple regression with 40 variables >and 35,000 observations) which, as you reported, takes SAS 3 hours, took >STATISTICA 25 seconds (the same hardware/system was used).

<sales pitch deleted>

The problem is not SAS, the problem is your Windows setup, I'm fairly certain. SAS for OS/2 is several times faster than the windows versions, but the difference is (in my experience so far) "only" about 4 times faster for the OS/2 version than for SAS for Windows. Based on that and my experience with a similar data set, I'd expect your job to take 30 seconds if you were to use SAS for OS/2, and about two minutes for this problem with SAS running under WinNT. You certainly do need to see what is wrong with your NT setup -- three hours is really excessive. I can't beleive that the SAS for OS/2 is more that 4 or 5 times faster than the Windows version.

I've run lots of SAS multivariate analysis with a data set of about 42,000 observations and I've come to expect few runs to take over 5 minutes even for fairly nasty SAS programs with a lot of procs (100mhz Pentium, 32meg memory -- the latter helps both OS/2 and WinNT a lot).

Cheers, John P. Ball john.ball@szooek.slu.se