Hi Alberto,
On 2 Mar 2011, at 19:22, Alberto Inuggi wrote:
> Hi all
>
> I'm analyzing resting state data in a group of children and I would like to correlate RS networks strenght with several attentional/inhibitory related scores.
> Assuming just one covariate. I have two EVs. the group mean and the covariate for each subject.
> the first contrast [1 0] give me the group mean
>
I assume you're doing this using dual regression. Just like in the case of higher-level GLMs you should demean the scores.
> 1)
> should I explicitly create two contrasts for investigating the positive [0 1] and negative [0 -1] correlation between my components and my covariate, or I just create the [0 1] one and melodic output the correlation sign as either a blue or red map for deactivation (neg correlation) and activation (pos correlation) respectively??
>
The between-subject analysis in dual regression is based on performing t-test within the randomisation framework, i.e. you only get one-tailed tests. As such, if you are interested in the inverse contrasts then you could simply add the reverse contrast as suggested. There is a quicker way of doing this, though, which just takes the positive contrast p-values and turns them into 1-p values. That way you avoid performing the randomisation again.
> 2)assuming one covariate and only one [0 1] contrast
>
> I cannot properly understand the meaning of the different combinations among
> a) Full model F-test
> b) group mean contrast
> c) the covariate contrast.
>
I suggest you go through the relevant section in the FEAT documentation (higher-level analysis). Briefly, every test corresponds (and should be conceptualised as such) to a question you have about the signal. in b) you are asking where in the brain the mean strenght/intensity is significantly >0. In c) you are asking where in the brain the covariate explains strength/intensity over and above the mean and in a) you are asking for voxels where the two regressors linearly combined produces any regression fit significantly larger or smaller than 0, i.e. where there is any effect at all.
> I can have different significance combinations.
>
> full model F test, group mean, covariate correlation.
> full model F test, group mean
> full model F test, , covariate correlation.
>
Yes, you are asking different questions
> and so on....
>
> In order to may affirm that a network is present in all the subjects do i need the significance of both full model F-test and group mean contrast or only the latter one?.
Most commonly a t-contrast on the group mean is used.
>
> In order to may affirm that a component strength correlate with a covariate do I need all the three significance (full model F-test, group mean contrast and covariate contrast) or only the latter two.
Only c)
hth
Christian
>
> for example which is the difference between having the three significance or just mean and covariate, or just full model and covariate, or simply the covariate.
>
> sorry for all these question,
> thanks in advance
>
> Alberto
|