Hi Shuran
> Dear Cyril,
>> Thank you for your reply! It took me a while to try to understand your comments. I have a few more questions if you don't mind to provide further guidance.
>> Is my understanding wrong like this - in ANOVA analysis, there is no direction for the activated clusters, the F results only indicate regions in which differences exit among groups?
yes F are bidirectional A different from B
> Then how to interpret prompt "only positive intensity displayed" in xjview for F contrasts? I'm also confused about prompt "only negative intensity displayed" for the T contrasts (positive effect of ***).
I don't use xjview but for a F 'positve' or 'negative' don't exist, that is which ever prompt you use you should get the same or something for + and nothing - (depending on how it's coded)
for a T because it is directional say A>B you have positive and negative values - if you have a voxel A=10 and B=-5 then the contrast +1 of your t-test (which effectively compute A-B) is +5 ; now imagine A=5 and B=10
then the contrast = -5 and you can see this result in SPM by using a contrast -1 -- now if you think about it all you have done in SPM is reversing the contrast, ie the two images are simply mirror around 0 ;
if you look in xjview at your contrast +1 but set the promt at negative intensity, then you are effectively looking at the contrast -1
> What's more, I'm not very clear about your advice that
> (2) you look at the effects separately in each clusters (spm - plot); as for any ANOVA report the effect sizes and direction of the effect per cluster.
> effect size=size per cluster (e.g. 100 voxels)? direction- should I perform post-hoc T tests for F contrasts to clarify the direction of activations?
> Could you detail a little bit or give an example?
the stats are computed for each voxel so with a F test you could have A>B in one region and A<B in another region - you have to look at the effects separately in each clusters - this means you have to plot the
different conditions
Concretely in SPM,
(1) set a F contrast [1 0 0 0 ; 0 1 0 0 ; 0 0 1 0; 0 0 0 1] which simply looks at where there is stuff going on
(2) look at your ANOVA results, select a cluster (via result table) click on plot and select this F contrast, this will show the effect size for each condition (A1 A2 B1 B2) in this cluster ; that you can
understand what is going one
the effect size in SPM are the raw beta parameters and this is not the best because of various reasons, I would suggest reporting % signal change - see my recent tutorial here
http://www.frontiersin.org/journal/10.3389/fnins.2014.00001/abstract
you can use MarsBar to compute these percentages - to my knowledge this is the most accurate tool to get this, although is doesn't implement what I describe in my paper (and my toolbox isn't ready yet)..
Cyril
--
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.
|