Dear Katerina,

On Tue, 10 Nov 2020 11:20:21 +0000, Katerina Pappa (PGR) <[log in to unmask]> wrote:

>Dear SPM and CAT12 experts,
>
>I was hoping you could help with the following please.
>
>As described in Kurth et al. (2015) the TFCE approach (Smith & Nichols, 2009) is recommended in VBM analysis to account for the multiple comparison problem. I used the TFCE toolbox called within CAT12 and run the TFCE estimates for the different contrasts.
>
>When I tried to run the TFCE estimate for the main effect of group I encountered the following error:
>"Please note that permutation is only done within subjects for repeated Anova. ERROR: No contrasts on subjects/block effects allowed."
You are trying to test for group effects at one time point in a flexible factorial model for longitudinal data. This is not supported as a contrast for the TFCE toolbox. In your case I would suggest building a separate model for the baseline data only to test for group effects at baseline. 

>
>Question 1:
>Should I choose the TFCE method to perform corrections in my analysis or use the traditional p < 0.05 FWE at the cluster level (voxel level p < 0.001 uncorrected)?
The cluster statistics is of course an alternative, but often less powerful compared to the TFCE.

>
>Question 2:
>Does the TFCE method work only for within group comparisons?
Yes, but not in a flexible factorial model.

Best,

Christian
>
>Thank you very much for your help!
>
>Best Wishes,
>Katerina
>
>-----Original Message-----
>From: Christian Gaser <[log in to unmask]>
>Sent: 16 October 2020 13:58
>To: Katerina Pappa (PGR) <[log in to unmask]>
>Subject: Re: CAT12 - longitudinal design, flexible factorial model
>
>Dear Katerina,
>
>On 16 Oct 2020, at 10:37, Katerina Pappa (PGR) wrote:
>
>> -----Original Message-----
>> From: Christian Gaser <[log in to unmask]>
>> Sent: 16 October 2020 07:39
>> To: [log in to unmask]; Katerina Pappa (PGR)
>> <[log in to unmask]>
>> Subject: Re: CAT12 - longitudinal design, flexible factorial model
>>
>> Dear Katerina,
>>
>> On Thu, 15 Oct 2020 10:10:02 +0000, Katerina Pappa (PGR)
>> <[log in to unmask]> wrote:
>>
>>> Dear SPM and CAT12 experts,
>>>
>>> I am interested in exploring the structural changes following a
>>> working memory training paradigm in healthy young adults.
>>> Structural MRI was acquired over three time points, i.e. time 1:
>>> pre-training, time 2: early training and time 3: post-training.
>>> Participants were assigned into two groups, i.e. training group and
>>> active control group.
>>> I am using a sub-sample of the full dataset to try out the VBM
>>> analysis.
>>>
>>> I have used the longitudinal option from the CAT12 toolbox to conduct
>>> the pre-processing steps.
>>> Following the CAT12 manual, I used a flexible factorial model with
>>> factors: Subject (Factor 1), Group ( Factor 2, 2 levels) and Time
>>> (Factor 3, 3 levels).
>>>
>>> Main effects and interactions to examine:
>>> Main effect of Factor 2
>>> Main effect of Factor 3
>>> Interaction of Factors 2 by 3
>>> Main effect of Factor 1
>> Here the main effects of Factor 2 and 3 seem unnecessary. It's
>> sufficient to model the main effect of Factor 1. Or are you not
>> referring to the main and interaction effects you have to define to
>> build your model ("Main effects & Interaction" in the flexible
>> factorial batch)?
>>
>> Please follow the example in the CAT12 manual on page 40 for the
>> design and page 44 for the contrasts.
>>
>> F-test
>> For any differences in Time in Group A     eye(k)-1/k
>> For any differences in Time in Group B     [zeros(k) eye(k)-1/k]
>> For main effect Time                              [eye(k)-1/k
>> eye(k)-1/k]
>> For interaction Time by Group                 [eye(k)-1/k 1/k-eye(k)]
>> For main effect Group                            [ones(1,k)/k
>> -ones(1,k)/k ones(1,n1)/n1  -ones(1,n2)/n2]
>> Effects of interest (for plotting)               [eye(k)-1/k zeros(k);
>> zeros(k) eye(k)-1/k]
>>
>> Here n1 and n2 are the number of subjects in Group A and B
>> respectively and k is the number of time points.
>>
>> However, please also see here for a discussion by Martyn McFarquhar
>> about inflating your statistical results by using the contrast for the
>> main effect Group:
>> https://www.jiscmail.ac.uk/cgi-bin/wa-jisc.exe?A2=SPM;ae98f7d5.2005
>> https://www.frontiersin.org/articles/10.3389/fnins.2019.00352/full
>>
>> Best,
>>
>> Christian
>>
>>>
>>> I used the weights as described in
>>> http://www.sbirc.ed.ac.uk/cyril/download/Contrast_Weighting_Glascher_
>>> Gi telman_2008.pdf in order to define the contrasts. The main effects
>>> and interactions work fine.
>>>
>>> However, when I set up the contrast for the main effect of condition
>>> with a single regressor, e.g. time 3 in my study, as described in
>>> design 3, contrast number 5, page 10 (Glascher & Gitelman), it's not
>>> clear to me what this represents. See picture attached for details.
>>>
>>> Could you please assist with this?
>>>
>>> Many thanks for your help in advance!
>>>
>>> Best Wishes,
>>> Katerina
>>>
>>>
>
>