|
Daniel,
I could not resist. Here is some code that you might want to adapt.
/****************************************************************************************************
* Program name: *
* Purpose: Demo of selecting sample size using bootstrap
*
* Description: Let us asume that we need to select a sample size that will
detect an error in a *
* population unsing bootstrap method. In this example the rate
of errors is about 20%. *
* In this case we use a sample of a 15 and want to estimate the
probability of getting *
* at least one error. We draw 500 samples of 15 each and look
at the distribution. *
* In this exmaple I used proc gchart to show the distribution
from where I can read or *
* see the probability to get ate least one error if I should
decide to have a sample of*
* 100. The sample size (100 in this case) can be increased or
decreased utill the *
* desired probability are reached.
*
* (In my experiments a sample size of ten will give one error 87% of the
time *
* A sample of 15 will give a 96% probability of detecting
the error *
* Written by: Francois van der Walt francoisw@gji.com.au Date:
June 2009 *
****************************************************************************************************/
data pop;
do i=1 to 1000; /* 1000 is the population size and can be changed */
if ranuni(0) <.2 then ind="Error"; /* .2 is the percentage errors in the
population - change as needed*/
else ind="Correct";
output;
end;
data bootstrp;
do k=1 to 500; /* 500 is the number of bootstraps */
errcount=0;
corrcount=0;
do j=1 to 15; /* 15 is the size of each bootstrap sample */
selobs=ranuni(0)*1000; /* 1000 is the population size and should match
data pop do statement */
set pop point=selobs;
if ind="Error" then ErrCount+1;
else CorrCount+1;
end;
output;
end;
stop;
run;
proc gchart data=bootstrp;
hbar ErrCount/discrete;
run;
If you have questions send me an e-mail.
Kind regards
Francois
On Mon, 22 Jun 2009 15:02:39 -0700, Daniel Nordlund <djnordlund@VERIZON.NET>
wrote:
>> -----Original Message-----
>> From: SAS(r) Discussion [mailto:SAS-L@LISTSERV.UGA.EDU] On
>> Behalf Of Shiling Zhang
>> Sent: Monday, June 22, 2009 2:54 PM
>> To: SAS-L@LISTSERV.UGA.EDU
>> Subject: Re: bootstrap confidence intervals
>>
>> Here is a basic example on how to use bootstrap resampling to do a
>> regression.
>>
>> SAS surveyselect procedure provides a easy way to obtain a bootstrap
>> resampling data set. The left over things are standard in regression.
>>
>> %let n=20;
>>
>> data t1;
>> do i =1 to &n;
>> x=rannor(567);
>> y=1 + 1 *x +rannor(567);
>> output;
>> end;
>> run;
>>
>> proc surveyselect seed=789 data=t1
>> METHOD=urs rate=1
>> rep=1000 outhits out=bootstrapped;
>> run;
>>
>> proc print data=bootstrapped(obs=100);
>> run;
>>
>> proc reg data=bootstrapped outest=est noprint;
>> by replicate;
>> model y = x ;
>> run;
>> quit;
>>
>> proc means data=est;
>> var intercept x;
>> run;
>>
>> proc reg data=t1 ;
>> model y = x ;
>> run;
>> quit;
>>
>>
>>
>> On Jun 22, 9:11 am, adel_ta...@YAHOO.FR ("adel F.") wrote:
>> > Hi,=0A=0AI=A0would like to =A0run a logistic regression
>> models using bootst=
>> > rap to get bootstrap confidence intervals for my odds
>> ratios and bootstrap =
>> > confidence intervals for AUC.=0A=0AIs there a SAS
>> procedure, which can help=
>> > me do that.=0A=0AThanks a lot,=0A=0AAdel=0A=0A=0A
>
>In addition to other advice/suggestions you have received, I highly
>recommend the paper by David Cassell at the link below.
>
>http://www2.sas.com/proceedings/forum2007/183-2007.pdf
>
>Hope this is helpful,
>
>Dan
>
>Daniel Nordlund
>Bothell, WA USA
|
