Subject: | | Re: Group conjuction analysis across three experiments |
From: | | Atesh Koul <[log in to unmask]> |
Reply-To: | | [log in to unmask][log in to unmask]> wrote:
> > Dear list, > > (I'm re-posting this as the first time I posted it without a subject title, which isn't very helpful for other users) > > I would be grateful for some advice on the following: I am conducting an evoked response DCM analysis on grand-average ERPs from a visual object processing study on young and old adults. > > I have some idea from previous studies of the network involved in this task from previous studies, and this is largely confirmed in an MSP source analysis on this data. > > My hypotheses relate specifically to the values of delay and connection strength parameters across groups. However I would like to use model selection to find the best model for the data before doing statistics on parameter values. This would be both to find the best connection topology and the optimal subset of nodes (via their connections). > > So Question 1: is this an advisable / recommended analysis strategy, given the above? > > If roughly 'yes': > > My experiment has a 2x2x2 factorial design (living vs. nonliving objects; basic vs. domain level naming; young vs old adults). The first two factors are within-subject, the third one is between subjects. These result in 8 ERPs in the grand average. > > So Question 2: When using BMS to compare alternative models and/or families, should I also specify the full set of contrasts in the between trial effects? Or does my specification to use all 8 ERPs in the .mat with the (1,2,3,4,5,6,7,8) vector already set up the cells of the factorial design? Or is this factor information not necessary for the model evidence? > > Thanks, > > John > > > -- > Mr. John Griffiths, MSc > > PhD Candidate > > Centre for Speech, Language, and the Brain > > Department of Experimental Psychology > > University of Cambridge, UK > >[log in to unmask] |
Date: | | Tue, 10 Jul 2012 20:22:09 +0530 |
Content-Type: | | text/plain |
Parts/Attachments: |
|
|
|
|
Dear Krishna,
Thanks for your reply. My experiment has 2 contrasts A and B in one
spm.mat, two other contrasts C and D in second spm.mat and similarly E and
F in third spm.mat. All I need to do is a conjunction of A to F. Given
this scenario, can you please elaborate a little more on the first
approach of Random effects analysis.
Atesh
> Dear Atesh,
>
> The conjunction principle as in Nichols et al. paper is that the minimum t
> value of all the contrasts will become the t value of the conjunction.
> Hence the correct method of doing conjunction is to do a random effects
> analysis and use the contrasts [1 0 0], [0 1 0] and [0 0 1] and do a
> conjunction between them. The alternative option is to do three one-sample
> t-tests, threshold them and do an imcalc to get a map where all the three
> overlap. if you want to know how much is the overlap, example in contrast
> A, the t-value may be 5 where as in contrast B, the t value may be 3. This
> you can do by using MRIcron or slover by projecting blobs of different
> colors e.g. red, blue, green and selecting the transparency of the colors,
> you can get the overlap areas using the usual RGB combinations. In general
> the Random effects analysis will be useful if you want to further probe
> into the results of your experiment such as finding the difference between
> contrasts A and B etc.
>
> The other methods like implicit masking should not be used for
> conjunctions, because the t-value you get are from the main contrast
> selected. Masking only filters out voxels not active in the second
> contrast.
>
> HTH
>
>
|
|
|
|