[MINC-users] mni.compute.FDR question

[Jason Lerch]

Yup, that's exactly what it means: no results at that FDR - see min(q 
$q) for the best possible FDR threshold in that particular dataset.

Jason

On 9-Nov-08, at 5:27 PM, Mishkin Derakhshan wrote:

> Hi,
> I'm receiving what I think is an error (but might actually be a
> result?) when I use the mni.compute.FDR function in R from the
> mni.cortical.statistics package, but I'm not sure.
>
> Here is my R code, and the warning messages:
>
> library(mni.cortical.statistics)
> gf <- mni.read.glim.file("input.glim",TRUE,FALSE,"csv")
> dt <- mni.build.data.table(gf)
> ms <- mni.mean.statistics(gf, 'y ~ group', vertex.table=dt)
> summary(ms)
> vs <- mni.vertex.statistics(gf, 'y ~ group', vertex.table=dt)
> mni.write.vertex.stats(vs, "output.vertstats", mean.stats=ms,  
> glim.matrix=gf)
> q <- mni.compute.FDR(vs$tstatistic[,2], df=23, fdr=0.05)
>
> Warning messages:
> 1: In max(r$ix[r$x == TRUE]) :
>  no non-missing arguments to max; returning -Inf
> 2: In qt((sorted.p.values$x[max(r$ix[r$x == TRUE])])/2, df) :
>  NAs introduced by coercion
> 3: In return(fdr.threshold, q) : multi-argument returns are deprecated
>
>
> When I check the fdr threshold value it is NA. Does this mean that
> there are no significant differences between my two groups, or is
> there an error with my data? The reason I say this is because I ran
> this exact code three time but with different groups:
> 1. CONTROLS VS SPMS - i get an fdr threshold
> 2. CONTROLS VS RRMS - i get NA
> 3. CONTROLS VS RENAMED-CONTROLS - I also get NA
>
> which is how I reverse engineered my theory that there was no
> significant difference. Of course I want to know if this is the case
> for sure before I draw any conclusions about my data.
>
> thanks,
> mishkin
> _______________________________________________
> MINC-users at bic.mni.mcgill.ca
> http://www2.bic.mni.mcgill.ca/mailman/listinfo/minc-users