[BioC] multtest Paired T-test error

[Katherine Pollard]

Ken,

For a "pairt" test, you need the labels to be 0's and 1's *and* to be
ordered with pairs next to each other. So, this should work:

labels<-c(0,1,0,1,0,1)
stats<-mt.teststat(m,labels,test="pairt",nonpara="n")

Also, be sure to reorder the columns of m accordingly.

Best,
Katie

> ------------------------------
>
> Message: 3
> Date: Thu, 27 Jan 2005 18:51:56 +0000
> From: "Ken Termiso" <jerk_alert at hotmail.com>
> Subject: [BioC] multtest Paired T-test error
> To: bioconductor at stat.math.ethz.ch
> Message-ID: <BAY101-F399A175AAEC5F20717A3AEE8780 at phx.gbl>
> Content-Type: text/plain; format=flowed
>
> On a Mac G5 with OSX 10.3 (updated) and R 2.0.1 I'm using multtest with good
> success until I tried to run a paired t-test...the other tests executed
> fine, but mt.teststat with test="pairt" didn't seem to like my class
> labels...
>
> m <- exprs(m579rma)
> m.cl <- c(0,0,0,1,1,1) # first 3 cols are control chips, last 3 cols are exp
> chips.
> m <- as.numeric(m)
> dim(m) <- c(22277,6)
> teststat_pairt_para_rma <- mt.teststat(m,m.cl,test="pairt",nonpara="n")
>
> Error in mt.checkclasslabel(classlabel, test) :
> 	Some errors in specifying classlabel for the paired t test for the block 1
> located at ( 1 2 )
> your classlabel= 0
>
>
>
> I then tried to see if it didn't like the zeroes in my classlabel...
>
>
>
> mpairedt.cl = c(1,1,1,2,2,2)
> teststat_pairt_para_rma <-
> mt.teststat(m,mpairedt.cl,test="pairt",nonpara="n")
>
> Error in mt.checkclasslabel(classlabel, test) :
> 	in paired t test, we only handle two groups
> your classlabel= 1
>
>
>
> As per another message in the BioC archives entitled 'Bug Report:
> mt.teststat with test="pairt"', I rearranged my chips so that the order was
> con,exp,con,exp,con,exp instead of con,con,con,exp,exp,exp..."m_pairt" is
> the name of this rearranged matrix of my chips...
>
>
>
> m_pairt.cl <- c(1,2,1,2,1,2)
> teststat_pairt_para_rma <-
> mt.teststat(m_pairt,m_pairt.cl,test="pairt",nonpara="n")
> Error in mt.checkclasslabel(classlabel, test) :
> 	in paired t test, we only handle two groups
> your classlabel= 1
>
>
>
> Should I just take the square root of the F-stat P-values to get the
> P-values for the paired t-test, or is there a workaround?
>
> Thanks in advance,
> Ken