Print

Print
I have not tested the approach suggested by Simon, but his model seems like it should be similar to my models (with subject/group/condition specified and in the design matrix). I still think you need the subject term in the model though, based on testing I've done.

However, you still can not test the between-subject effects (e.g. G1>G2 or G1>0 or G1C1>0). G1C1-G1C2>G2C1-G2C2 would be a within-subject effect even though you are  comparing groups. The issue with any repeated-measures designs (with multiple groups) is that in order to be able to test the between-subject effect you need the between-subject error term that can't be produced when you have more than one observation per subject in the model in SPM.

If you simply the case to only one 1 group and 3 conditions, testing any one condition against 0 would be a between-subject test and be incorrect in both the flex. factorial or full factorial models.

GLM Flex, a program developed at MGH, produces multiple error terms and can assess both main effects and interactions in the construct of a single model.

Best Regards, Donald McLaren
=================
D.G. McLaren, Ph.D.
Postdoctoral Research Fellow, GRECC, Bedford VA
Research Fellow, Department of Neurology, Massachusetts General Hospital and
Harvard Medical School
Website: http://www.martinos.org/~mclaren
Office: (773) 406-2464
=====================
This e-mail contains CONFIDENTIAL INFORMATION which may contain PROTECTED
HEALTHCARE INFORMATION and may also be LEGALLY PRIVILEGED and which is
intended only for the use of the individual or entity named above. If the
reader of the e-mail is not the intended recipient or the employee or agent
responsible for delivering it to the intended recipient, you are hereby
notified that you are in possession of confidential and privileged
information. Any unauthorized use, disclosure, copying or the taking of any
action in reliance on the contents of this information is strictly
prohibited and may be unlawful. If you have received this e-mail
unintentionally, please immediately notify the sender via telephone at (773)
406-2464 or email.



On Wed, May 16, 2012 at 10:43 AM, Votinov Mikhail <[log in to unmask]> wrote:
Dear SPM list (especially Simon and Donald),
I'm confused by two different responses by Simon and Donald (see below) regarding groups comparisons in flexible factorial model. 
In my case I have 3 groups and 3 conditions, that I want to compare between groups. How should I specify my model correctly? Donald says that you should not model group 
effects in Flex. factorial model, but then Simon suggested to use condition factor as "group" factor. Related question, to Simon, 
why do you specify only the main effect for the condition-factor, but not for subjects factor as well?
Thank you in advance
Best Regards
Mikhail 






Dear Jerome


One possible solution to your problem is to realize, that not every condition needs to be specified in every subject.


Thus, you can set up a flexible factorial model using two factors: Subject and condition. Usually, you would specify unequal variance but independence between subjects and unequal variance and non-independence for conditions. As main-effects & interactions, only specify the main effect for the condition-factor.

For subjects in Group 1, you then specify the scans as belonging to condition 1-3, for subjects in Group 2 as belonging to conditions 4-6. In the ensuing design (with 6 cells corresponding to the three conditions in the two groups) you can then test for all main effects and interactions.


best
Simon





________________________________________
Von: SPM (Statistical Parametric Mapping) [[log in to unmask]]&quot; im Auftrag von &quot;Jérôme Redouté [[log in to unmask]]
Gesendet: Donnerstag, 10. Mai 2012 12:21
An: [log in to unmask]
Betreff: Re: [SPM] Between-groups comparisons

Le 08/05/2012 20:42, MCLAREN, Donald a écrit :
> You stated that you have a contrast between two conditions. This
> implied that you have one measurement per subject. Thus, the
> two-sample t-test would be appropriate.
>
> If you mean that you have two conditions that you want to compare
> between groups, then you have 2 measures per subject. Thus, the
> flexible factorial would be appropriate. Remember that you will only
> be able to compare the two conditions or the group*condition
> interaction. The effect of group is invalid.
>
Dear SPMers,
We are facing a similar problem as we  have to analyse an O15-H20 PET
study with the following specifications:
2 groups: ON medication / OFF medication
3 conditions: Task1 / Task2 / Control (2 or 3 repetition by task)

We defined our design matrix with a flexible factorial design as follows
Factor 1: subjects
Factor 2: Group
Factor 3: Tasks

Our main questions here, are to compare:
     Task 1 vs Control between ON & OFF groups
     Task 2 vs Control between ON & OFF groups

there are different options concerning the effects to include in the Matrix
Main effect of subject (Factor 1)
Main effect of Group ?
Main effect of Tasks ?
interaction Group x Task (Factor 2 x Factor 3)

In the case of interaction Group x Task, the contrast weights we need to
define are different whereas the main effects are modelized or not...
More specifically, if Main Effects are included, it's impossible to test
only the interaction term: the contrast manager seems to wait for
weights in the Main effects columns too.

Any suggestions to help us analyze these data are welcome...
Thanks for your help
J.R