When you do the p->t conversion, are you assuming a two-sided t? Do you want to try the --synth option (it's a lot easier when you *know* they should be t).
d.g.lawyer@medisin.uio.no wrote:
Actually, it is an excellent question, and gets at the heart of my research-- rather than saying all bets are off, using the skew of the t-values to gauge the effect (or number of hypothesis/vertices where we have an effect).
My big problem is that I can't investigate the distribution of the t-values if I don't know which set of t-values is valid. Converting p to t should give me the same t-values as the output t-values, but it doesn't, not even close.
+glenn
This might seem like an odd question, but why do you expect the t values to be t-distributed? Remember, they will only be t-distributed under the null. If you have an effect, and I assume you do, then all bets are off. Try doing the same thing with synthetic guassian noise (mri_glmfit will do this for your if you just add --synth to the cmd line).
doug
d.g.lawyer@medisin.uio.no wrote:
Hi,
I'm seeing some odd behavior in t-values and p-values exported from FreeSurfer. In geeky detail:
fit a linear model using FreeSurfer, saving t and p-values convert output files to ascii load ascii files into R convert the FreeSurfer "p-values" into real p-values via lh.pval <- 10^(-1*abs(freesurfer.lh.pvals)) # this is necessary as FreeSurfer writes the -log10 of the pvalue, with the sign # demonstrating direction of effect. convert these pvalues to tvalues lh.convert.t <- qt(lh.pval,XX) # where XX is the degrees freedom # I have several studies, 60 < XX < 200 A histogram of lh.convert.t is roughly OK, could be zero-mean.
HOWEVER, the t-values exported by FreeSurfer are not. In one study, the range of the converted t-values was [-3.8, 3], but the range of the raw t-values was [-0.6,3.8]. In a second study, the range of the raw t-values was [0,17]. As you could guess from these ranges, histograms of the two versions of t-values also differ radically.
Regarding version numbers and such, I have three experiments I have checked for this phenomenon. Two are old, and were run using version 1.2 (output files in .w format, etc), and the third study is new (version 3.X, outputs are F.mgz and sig.mgz). We run the linux/RHEL versions.
My questions: why are the t-values output by FreeSurfer so oddly distributed? and why don't they agree with what i get when I convert the p's to t's? Silly user error on my part?
Thanks much for any insights
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
-- Douglas N. Greve, Ph.D. MGH-NMR Center greve@nmr.mgh.harvard.edu Phone Number: 617-724-2358 Fax: 617-726-7422
In order to help us help you, please follow the steps in: surfer.nmr.mgh.harvard.edu/fswiki/BugReporting