Date: Thu, 8 Sep 2005 21:42:43 -0700
Reply-To: David L Cassell <davidlcassell@MSN.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: David L Cassell <davidlcassell@MSN.COM>
Subject: Re: Geometric mean rise (GMR) and confidence intervals
Content-Type: text/plain; format=flowed
>To get 95% CI for the ratios you can (should) do bootstrapping . You
>can use proc SURVEYSELECT to generated random sample with replacement.
>Calculate ratio for every random sample.
>Do this many times :)
>And after this just find 2.5 and 97.5 percentile of those generated
Ahh, I see that Andre has fallen prey to my propaganda. Umm, I mean,
he has been reading my notes.
If you use PROC SURVEYSELECT for bootstrapping (which I really do
recommend), be sure to get all the pieces that you need:
proc surveyselect data=YourData out=YourBootData
method=urs /* just as Andre said */
seed=394575664 /* pick your own seed, not 0 */
outhits /* be sure to get all the duplicates of records hit more
than once */
That's it. Your own bootstrap sample, all loaded in 1000 chunks
in the YourBootData file. Then you run your proc again, only this time
with the statement
stuck in it so that each replicate of your bootstrap sample is done
separately. Then you need to output the results, possibly using
ODS. (I can't tell, since you didn't give us any extra info.) After
that, you'll have a data set with one record per replicate, which
you may need to massage slightly before you can feed it into
something like PROC UNIVARIATE to get the percentiles for your CI.
Alternatively, depending on your data and your analysis, it may be
possible to get the desired estimate and confidence interval directly.
But I can't say. PROC NLMIXED might be able to do it for you, but
that's hard to say when my crystal ball is so cloudy.
David L. Cassell
3115 NW Norwood Pl.
Corvallis OR 97330
Donít just search. Find. Check out the new MSN Search!