Print

Print


Hi everyone - 

I wanted to ask a follow-up question on this discussion. It is about a single-subject (first) level, however.
In a single subject analysis, I've set up one condition (with 10 onsets for 10 trials) and behavioural performance as a parametric modulator, and set up a t-contrast like: 0 1 0.. (so that the contrast highlights the behavioural performance regressor). 
My question is, what are the dots on the glass brain output actually telling me? That in these regions, increases in behaviour are related to increases in activation perhaps?

Thanks

On Mon, Jun 15, 2015 at 7:45 PM, MCLAREN, Donald <[log in to unmask]> wrote:
Almost correct, at the group level the test statistics don't test for "commonality", but rather where is the group "average" is different from 0 (F-test) or greater than 0 (T-test [1]) or less than 0 (T-test [-1]).

If all subject show activation/deactivation at the same location, then it will likely show up in the T-/F-tests. You may also find additional areas in the group maps that never appeared in the single subject maps. For example, if all subjects had an activation contrast of 0.001, then its very unlikely that this would show up in the individual maps, but because all subjects had the same contrast value, then it would be very significant at the group level.

-Donald

On Mon, Jun 15, 2015 at 6:16 AM, Joelle Zimmermann <[log in to unmask]> wrote:
Hi Helmut and Martin,

Thank you both for your advice! :-) That makes perfect sense.

In that case, doesn't an F-contrast [1] tell me where the subjects commonly activate and and commonly deactivate? 
Whereas a T-contrast [1] tells me where subjects commonly activate, and a T-contrast [-1] tells me where subjects commonly deactivate? 

Thanks,
Joelle

On Mon, Jun 15, 2015 at 5:19 PM, H. Nebl <[log in to unmask]> wrote:
Dear Joelle,

The F-contrast [1] is undirected, the t-contrast in SPM are always directed = one-sided. In your case with a single group the two tests produce equivalent results, with F = T^2.

If you just set up one one-sided t-contrast (e.g. [1]) because you're only interested in this direction, thresholded at a certain alpha, you will indeed get more significant findings than with one F-contrast. This is nothing surprising though. The F-contrast [1] thresholded at alpha is equivalent to two one-sided t-contrasts [1] and [-1] thresholded at alpha/2 each. Or put it differently, if you conduct two one-sided t-contrasts [1] and [-1] thresholded at alpha each, this would be equivalent to an F-contrast thresholded at alpha*2.

In neuroscience it is very common to look at both directions but to threshold the t-contrasts at .001 each, thus not accounting for no. of tests. This is sometimes interpreted as the t-contrast to be more sensitive, but actually it's due to not accounting for no. of performed tests.

Concerning your results, given that you only have five subjects there might just be too much variance to detect a consistent response / significant clusters/voxels, even if all subjects show a response going into the same direction. There might of course be other issues (outliers, noise, ...).

Best

Helmut