A follow-up on PermutationTest2 for the S-News archive...
Thank you to Tim Hesterberg who identified a problem in my original
post. In re-typing my commands for the posting, I inadvertently made
my array lengths 3 instead of length 2. Tim's description below
explains in a straight-forward way how the PermutationTest2 works.
---------------------------------------------------
I can't reproduce that; when I do
a<-c(1,1,1)
b<-c(10,10,10)
c<-permutationTest2(a,data2=b,statistic=mean,paired=T)
I get:
unique(c$replicates)
[1] -3 9 3 -9
This makes sense; the values correspond to switching 0, 1, 2, or 3 of
the
pairs.
I can reproduce your results if I make a and b length 2.
Tim Hesterberg
On Apr 7, 2006, at 8:52 PM, Chris Fiebrich wrote:
I'm using PermutationTest2 to compare two samples. My data is
paired, so I am using the "paired=T" option. I wasn't convinced
this algorithm was working the way I thought, so I did a simple test:
a<-c(1,1,1)
b<-c(10,10,10)
c<-permutationTest2(a,data2=b,statistic=mean,paired=T)
When I view c$replicates, I see lots of 0s, 9s and -9s. How can it
calculate 0 for one of the replicates if it is keeping the pairs
together? I would think the only permutations would be:
1-10 and 10-1 (or -9 and 9). Obviously, it's also using 10-10 and
1-1 to get the "0" replicates.
When I use paired=F, I also get replicates that include 0, 9 and
-9. I understand that.
Does anyone understand how the paired=T option works?
--------------------------------------------------------------------
This message was distributed by s-news@lists.biostat.wustl.edu. To
unsubscribe send e-mail to s-news-request@lists.biostat.wustl.edu with
the BODY of the message: unsubscribe s-news
|