Dear Guillaume:
Thank you so much for your prompt reply! I just wonder, if I would like to test the main effects as well, then I can’t model it like this right? Would it be ok to do it as I proposed then?
Thanks a lot!
All the best,
Katarina
Sent from my iPhone
> On 18 Oct 2019, at 12:28, Flandin, Guillaume <[log in to unmask]> wrote:
>
> Dear Katarina,
>
> The simplest here is to compute the "ind vs dir" contrast for each
> subject from the two groups at the first level, and enter these images
> in a two-sample t-test at the second level. An F-contrast [1 -1] will
> then test for the group by condition interaction.
>
> Best regards,
> Guillaume.
>
>
>> On 18/10/2019 11:19, Katarina Bendtz wrote:
>> Dear experts:
>>
>> I would like to ask for your opinion on my design, if anyone has the time. I have an fMRI experiment with 2 conditions: ”ind” and ”dir” and two groups: group 1 and group 2. All participants was presented with both conditions (repeated measures), but each participant belongs to only one group. I am interested in the group difference for the difference between ind and dir, so basically the interaction between condition and group.
>>
>> My idea is to construct contrasts at first level for ”ind vs baseline” and ”dir vs baseline” and then take these to the second level with a design matrix that looks like this:
>>
>> First ”block” (as in the first consecutive rows) of rows: ”ind vs baseline” contrast values for all subjects in group 1
>> Second block of rows: ”dir vs baseline”, group 2
>> Third block of rows: ”ind vs baseline”, group 1
>> Fourth block of rows: ”dir vs baseline”, group 2
>>
>> The first column represents the main effect of group and is = [-1 -1 1 1]’
>> The second column represents the main effect of condition and is = [1 -1 1 -1]’
>> The second column represents the interaction effect (the one I am interested in) an is = [1 -1 -1 1]’
>> The fourth row is just the grand mean = [1 1 1 1]
>>
>> To compute the ANOVA, I would then use the SPM F-contrasts [1 0 0 0], [0 1 0 0] and [0 0 1 0] for the main effects and the interaction, respectively.
>>
>> As I understand it, SPM will this way calculate the F-value based on the extra squared residuals for each effect by comparing to the reduced model(s), so that the F-value I get from the F-contrasts are the hypothesois test of these effects just like if I would have tested different GLMs, with and without these factors modelled, just like in any ANOVA.
>>
>> Do you think I have understood it correctly or have I missed anything?
>>
>> Thank you so much,
>> All the best,
>> Katarina
>>
>> Sent from my iPad
>>
>
> --
> Guillaume Flandin, PhD
> Wellcome Centre for Human Neuroimaging
> UCL Queen Square Institute of Neurology
> London WC1N 3BG
|