Subject: | | Re: F-contrasts vs. "planned" t-tests / Vaild approach for ANOVAs |
From: | | Roberto Viviani <[log in to unmask]> |
Reply-To: | | [log in to unmask][log in to unmask]> 7.11.2012 > 17:46>>> > Is that bilateral thalamic and caudate activation? Is that something > that makes sense for your experiment? > I think what you're seeing *could* be due to only having 9 subjects; > however, if you see the standard L frontal activation in most/all of > them, then I don't know what might be causing the discrepancy. Is there > > an especially high amount of motion in these subjects? Does the > normalization look right? > > > On 11/07/2012 11:46 AM, Kirsten Labudda wrote: >> Dear Chris, >> thanks for your quick response. I attached the screenshots of both >> first and second level analyses versions I conducted (version 1: > only >> the activation condition was modelled; version 2: activation and > rest >> was modelled on the first level) in the small group. I used the > movement >> parameters as regressors in the first level analysis. The second > level >> results are thersholded at p<.001 as only very few voxel survived > the >> thershold of p<.05 FWE corrected on the second level. >> Thanks for your help! >> Kirsten >> >> --------------------------------------------------------- >> Dr. Kirsten Labudda, Dipl.-Psych. >> >> Krankenhaus Mara >> >> MRT-Abteilung >> Tel.: 0521-772 777 61 >> & >> Station für Psychosomatische Epileptologie >> Tel.: 0521-772 789 22 >> --------------------------------------------------------- >> >>>>> Chris Watson<[log in to unmask]> 7.11.2012 >> 16:38>>> >> Can you post screenshots of your design matrix, the results of 1st > and >> 2nd-level analyses, etc? >> >> Regarding your design: you only need one regressor, for the task. A >> contrast of "1" will reflect "task> rest". >> You also might want to exclude the movement parameters, if your >> experiment is a block design. Check what the literature says on that >> matter. >> >> >> On 11/07/2012 09:46 AM, Kirsten wrote: >>> Dear fMRI-experts, >>> we wonder about conflicting results of the first and second level >> analysis we conducted with our fMRI data. We used a simple blocked >> verbal fluency task with one activation condition (verbal fluency, > 10 >> blocks) and a rest condition (also 10 blocks). I wonder, whether we > did >> something wrong when using SPM (we used SPM8 and 5 and have the > problem >> with both versions). That’s what we did: After preprocessing >> (realignment, normalization, smoothing), we conducted a first level >> analysis specifying the verbal fluency blocks as activation condition > in >> each subject (by entering the onset scans of each block and its > duration >> in terms of scans) and used the movement parameters as individual >> regressors. We defined two contrasts (verbal fluency: 1 and rest: > -1). >> Is it ok not to model the resting condition separately? We thought so > as >> our design only includes two conditions and with that the vector 1 >> automatically contains the information activation> rest, right? >> Nevertheless, I also conducted the first level analysis with the two >> conditions modeled separately using two T-contrasts then (verbal >> fluency> rest: 1 -1 and rest>verbal fluency -1 1). Both first > level >> procedures lead to very similar results reflecting typical cortical >> language activations. >>> I then used the contrast images (the activation condition>rest, > again >> of both first level procedures described above) in the second level >> analysis to run a one-sample t-test with the contrasts: activation> >> rest: 1 and rest>activation: -1. Surprisingly, the typical cortical >> language activation from the first level analysis completely >> disappeared. Instead, only subcortical activation remained (that was >> present in the first level analysis, too, but it was much weaker > than >> the typical language activation). >>> We have the problem of very incongruent first and second level >> results with SPM5 and 8 and within two different patient groups (one > was >> small having 8 subjects only, but the other group includes 22 >> subjects). >>> Does anybody have an idea why the first and second level results > are >> so divergent? Did we simply do something wrong in SPM? >>> Thanks in advance, >>> Kirsten >>> >> ************************************************************ >> Krankenhaus Mara gGmbH >> Akademisches Lehrkrankenhaus der Universität Münster >> >> Sitz der Gesellschaft >> Kantensiek 11 | 33617 Bielefeld >> >> Amtsgericht Bielefeld HRB 39136 >> >> Geschäftsführer >> Dr. Rainer Norden (Vorsitz) >> Dr. Thomas Krössin >>[log in to unmask] |
Date: | | Wed, 14 Nov 2012 11:34:32 +0100 |
Content-Type: | | text/plain |
Parts/Attachments: |
|
|
|
|
> Okay, the "neuropsychological test battery" was a really bad
> example. If your battery is large enough, and you correct for
> multiple comparisons, it is very unlikely to find any true effects.
> So I would not have corrected on F-test level (would have reported
> both corrected and uncorrected p-values).
>
> Anyway, I had thought that I would have to take into account the
> number of post-hoc tests (as well), but this seems to be wrong then.
It isn't wrong, it is a problem that is not specific to neuroimaging.
If you correct for the volume at each t-test, then you have the same
problem you'd have if it was the neuropsychological test battery.
Here, the F test does not solve the problem because the comparisons
must be planned. You'd rather need a Bonferroni correction.
The opinions on what to do in this situation vary. In genetic
epidemiology, the tendency is to require large samples and
(Bonferroni-corrected) significances much lower than 0.05 for
credibility. In the political sciences, the view has been expressed
that corrections are harmful because of the effect on type II error.
Others like FDR approaches, which become increasingly attractive when
the number of tests becomes high.
>
>
> Back to the fMRI data. Imagine a purely within-subject 3x3-ANOVA,
> which should be reasonable nowadays. E.g. something like "face"
> (happy, sad, fearful), and "sex" (male, female, morph). Maybe I have
> specific hypotheses, but maybe I do not (at least for some levels,
> e.g. concerning "morph"). In the latter case, I would run F-tests
> for "face", "sex" and the interaction. Imagine I get some clusters
> surpassing an otherwise defined voxel-size threshold. What should I
> do then?
>
>
> Or should I run lots of t-tests right from the beginning? I would
> already have to conduct 12 one-sided tests for "face" and "sex". And
> to ensure that the results make sense, I would have to check all
> the interactions as well.
You could declare your t tests as explorative.
You won't escape the problem by adopting one or the other approach to
correction. To have an intuitive understanding of the inevitability of
the problem, see it as a requirement on the resolution of your data.
If you want high resolution (to figure out which of these many
conditions is responsible for variance), you need more data;
otherwise, you'll be looking at noise.
Best wishes,
Roberto Viviani
Dept. of Psychiatry III
University of Ulm, Germany
|
|
|
|