potency@violet.berkeley.edu (Tom Slone) (03/15/89)
I am planning to work with a large body of raw data in the behavioral sciences
(~500K data points). I would appreciate advice on sophisticated statistical
programs for the Macintosh and/or IBM PC for use in analyzing this data. The
data is temporal, i.e., each data point represents information for 1 minute.
If you have experience with using sophisticated statistical programs on the Mac
or PC, I would appreciate your advice.
potency@violet.berkeley.edu
{decvax|hplabs|ihnp4|decwrl|nbires| \
sdcsvax|tektronix|ulysses}!ucbvax!violet!potencyphil@wubios.wustl.edu (J. Philip Miller) (03/17/89)
In article <21641@agate.BERKELEY.EDU> potency@violet.berkeley.edu (Tom Slone) writes: >I am planning to work with a large body of raw data in the behavioral sciences >(~500K data points). I would appreciate advice on sophisticated statistical >programs for the Macintosh and/or IBM PC for use in analyzing this data. The >data is temporal, i.e., each data point represents information for 1 minute. >If you have experience with using sophisticated statistical programs on the Mac >or PC, I would appreciate your advice. > I have had reasonable experience with using such large datasets on SAS on a PC. You definitely do not want an original XT with its 10Meg disk. You want a 286 or 386 processor, you want 2-4Meg of LIM Spec 4.0 memory, and obviously for this type of data you want a large, fast disk. Historically SAS has proven to be very useful when you have datasets like this that you are likely to want to process in a variety of different ways, computing summary indices, and then do analyses of the resulting indices. SAS has all the tools to do this type of processing. The ETS modules also have tools for doing time series models, if this is part of your analytic strategy. -- -*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-* J. Philip Miller - Div of Biostat - Washington Univ Medical School phil@wubios.WUstl.edu - Internet phil@wubios.wustl - bitnet (314) 362-3617 c90562jm@wuvmd - alternate bitnet
rns@se-sd.sandiego.ncr.com (Rick Schubert) (03/21/89)
In article <433@wubios.wustl.edu> phil@wubios.UUCP (J. Philip Miller) writes: >I have had reasonable experience with using such large datasets on SAS on a >PC. You definitely do not want an original XT with its 10Meg disk. You want >a 286 or 386 processor, you want 2-4Meg of LIM Spec 4.0 memory, and obviously >for this type of data you want a large, fast disk. Would an 80*87 be useful and/or advisable for SAS (or other statistical packages, for that matter)? -- Rick Schubert (rns@se-sd.sandiego.NCR.COM)
hes@ecsvax.UUCP (Henry Schaffer) (03/22/89)
In article <1858@se-sd.sandiego.ncr.com>, rns@se-sd.sandiego.ncr.com (Rick Schubert) writes: > ... > Would an 80*87 be useful and/or advisable for SAS (or other statistical > packages, for that matter)? > > -- Rick Schubert (rns@se-sd.sandiego.NCR.COM) Yes - highly recommended for heavy duty statistical analysis. ("heavy duty" means that you spend more time doing the arithmetic than reading the data in from the disk.) --henry schaffer n c state univ
phil@wubios.wustl.edu (J. Philip Miller) (03/22/89)
In article <1858@se-sd.sandiego.ncr.com> rns@se-sd.sandiego.NCR.COM (Rick Schubert(AEP)) writes: >In article <433@wubios.wustl.edu> phil@wubios.UUCP (J. Philip Miller) writes: >>I have had reasonable experience with using such large datasets on SAS on a >>PC. You definitely do not want an original XT with its 10Meg disk. You want > >Would an 80*87 be useful and/or advisable for SAS (or other statistical >packages, for that matter)? > yes, in general it is useful, though perhaps not as useful as many folk would suspect. Stat packages typically spend a lot of their time executing control logic, formatting, and other non-Floating point operations. When you are talking about AT class machines where an appropriately 80287 is only a couple of hundred dollars, then almost always it is worth it since it will give you a 20-50% reduction in execution time for many computationally intensive operations. For 80386 processors, particularly for those running at 25+ MHz, then you are talking over $1k for the chip and it may not be such a great investment. Since it is dependent on the particular instruction mix, it is always wise to do some benchmarks on your type of problem. For many analyses, I find the computer is just waiting for my slow brain & fingers so the fastest machine is really not making my job go any faster :-( -phil -- -*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-* J. Philip Miller - Div of Biostat - Washington Univ Medical School phil@wubios.WUstl.edu - Internet phil@wubios.wustl - bitnet (314) 362-3617 c90562jm@wuvmd - alternate bitnet