Thank you Tom. Would you predict that future versions of randomise will be able to accomodate this type of design? I imagine it is a pretty common setup for longitudinal studies. Also, can anyone suggest alternative approaches to determining the group x time interaction for tbss/vbm? This will be an important issue for me in an upcoming study comparing treatment effects.
thanks a lot,
Kyle Kern
________________________________________
From: FSL - FMRIB's Software Library [[log in to unmask]] On Behalf Of Thomas Nichols [[log in to unmask]]
Sent: Tuesday, April 21, 2009 3:09 AM
To: [log in to unmask]
Subject: Re: [FSL] 2x4 ANOVA, 1 repeated measure
Dear Kyle,
Sorry for the delay. The reason that no one has chimed in is that randomise can't deal with this design in general. As you concisely state, you have 2-way repeated measures ANOVA, 1 within-subject factor (time) and 1 between subject factor (group) (and the repeated 'cluster' variable subject).
In this advanced model, the appropriate way to do a permutation test depends on the question asked. If you are asking just about the within-subject factor (time), randomise *could* deal with this, just use the model as you have specified it *but* be sure to specify an exchangeability block file marking the subjects (e.g., transposed it would like like
1 1 1 1 2 2 2 2 3 3 3 3 ...
). Randomise will then know to only permute scans within subject (appropriate for a null hypothesis of no time effect, *and* implicitly assuming exchangeability within subject).
But if you're asking about the between-subject factor, the appropriate way to permute is to shuffle subjects' data as a whole around. Or, equivalently to, permute the design ensuring that only permutations that move around sets of four rows together are used). That currently isn't implemented in randomise.
Hope this helped!
-Tom
On Wed, Mar 25, 2009 at 12:34 AM, Kern, Kyle <[log in to unmask]<mailto:[log in to unmask]>> wrote:
Hi fsl users,
I have a tbss dataset with a control group, a patient group, and 4 timepoints for each subject. I am trying to create a GLM design that essentially models a 2x4 ANOVA with 1 repeated measure (time). I was hoping that anyone who has worked with a similar design could look over my setup and steer me in the right direction. This study has 11 controls and 21 patients, but for simplicity, the following design has 3 controls and 4 patients. I would appreciate any corrections or better approaches to the design below:
.mat file:
cont1_T1 1 0 0 0 0 0 1 0 0 0 0 0 0
cont1_T2 0 1 0 0 0 0 1 0 0 0 0 0 0
cont1_T3 0 0 1 0 0 0 1 0 0 0 0 0 0
cont1_T4 0 0 0 0 0 0 1 0 0 0 0 0 0
cont2_T1 1 0 0 0 0 0 0 1 0 0 0 0 0
cont2_T2 0 1 0 0 0 0 0 1 0 0 0 0 0
cont2_T3 0 0 1 0 0 0 0 1 0 0 0 0 0
cont2_T4 0 0 0 0 0 0 0 1 0 0 0 0 0
cont3_T1 1 0 0 0 0 0 0 0 1 0 0 0 0
cont3_T2 0 1 0 0 0 0 0 0 1 0 0 0 0
cont3_T3 0 0 1 0 0 0 0 0 1 0 0 0 0
cont3_T4 0 0 0 0 0 0 0 0 1 0 0 0 0
pat1_T1 0 0 0 1 0 0 0 0 0 1 0 0 0
pat1_T2 0 0 0 0 1 0 0 0 0 1 0 0 0
pat1_T3 0 0 0 0 0 1 0 0 0 1 0 0 0
pat1_T4 0 0 0 0 0 0 0 0 0 1 0 0 0
pat2_T1 0 0 0 1 0 0 0 0 0 0 1 0 0
pat2_T2 0 0 0 0 1 0 0 0 0 0 1 0 0
pat2_T3 0 0 0 0 0 1 0 0 0 0 1 0 0
pat2_T4 0 0 0 0 0 0 0 0 0 0 1 0 0
pat3_T1 0 0 0 1 0 0 0 0 0 0 0 1 0
pat3_T2 0 0 0 0 1 0 0 0 0 0 0 1 0
pat3_T3 0 0 0 0 0 1 0 0 0 0 0 1 0
pat3_T4 0 0 0 0 0 0 0 0 0 0 0 1 0
pat4_T1 0 0 0 1 0 0 0 0 0 0 0 0 1
pat4_T2 0 0 0 0 1 0 0 0 0 0 0 0 1
pat4_T3 0 0 0 0 0 1 0 0 0 0 0 0 1
pat4_T4 0 0 0 0 0 0 0 0 0 0 0 0 1
EVs 1-3 model controls at each timepoint, 4-6 model patients at each timepoint, and 7-13 model the individual subject means.
contrasts:
Controls-Patients: 1 1 1 -1 -1 -1 0 0 0 0 0 0 0 X Ftest1: Main effect of Group
Time1-Time4: 1 0 0 1 0 0 0 0 0 0 0 0 0 X
Time2-Time4: 0 1 0 0 1 0 0 0 0 0 0 0 0 X Ftest2: Main effect of Time
Time3-Time4: 0 0 1 0 0 1 0 0 0 0 0 0 0 X
C(T1-T4)-P(T1-T4): 1 0 0 -1 0 0 0 0 0 0 0 0 0 X
C(T2-T4)-P(T2-T4): 0 1 0 0 -1 0 0 0 0 0 0 0 0 X Ftest3: Interaction Group x Time
C(T3-T4)-P(T3-T4): 0 0 1 0 0 -1 0 0 0 0 0 0 0 X
And I can also include contrasts for T1-T3, T2-T3, T1-T2, C(T1-T3)-P(T1-T3), C(T2-T3)-P(T2-T3), C(T1-T2)-P(T1-T2), but not include them as ftests right? (since they are linear combinations of the others)
I am not sure how the data should be grouped for the exchangability blocks, so I have them all in one group in the .grp file, is that correct?
Does this setup look appropriate?
attached are the graphical representations.
thanks for the help,
Kyle Kern
________________________________________
From: FSL - FMRIB's Software Library [[log in to unmask]<mailto:[log in to unmask]>] On Behalf Of MCLAREN, Donald [[log in to unmask]<mailto:[log in to unmask]>]
Sent: Wednesday, March 18, 2009 10:06 PM
To: [log in to unmask]<mailto:[log in to unmask]>
Subject: Re: [FSL] modeling slope over time
Instead of estimating the within-subject slopes, I'd do a repeated measures ANOVA. In the ANOVA, you will want subject, group, time, and a group-by-time interaction term. The reason for doing it this way is to avoid the assumption that the change is linear with time.
On Wed, Mar 18, 2009 at 10:01 PM, Kyle Kern <[log in to unmask]<mailto:[log in to unmask]><mailto:[log in to unmask]<mailto:[log in to unmask]>>> wrote:
Hi,
I am trying to set up a longitudinal tbss analysis over 4 time points contrasting a patient
group to a control group. I'm interested in change over time rather than the effect of any
specifc time point. I was planning to model the slope for each subject and then put this
into a group t-test. Would modelling the slope for each subject be the same approach as a
simple correlation where my design matrix might look like this?:
-3
-1
1
3
Would this reflect the magnitude of the slope or does it only reflect the linearity like a
correlation coefficient? I would hypothesize that patients show a steeper slope than
controls though they may show more variability, so a least squares regression slope
might detect differences but a simple correlation coefficient might not.
Also, to perform a 2 level analysis with tbss is it preferable to use randomise and feed
the within subject t-stat into the group comparison or should I use the feat gui and treat
each within subject analysis as a timeseries and feed that into a higher level mixed
effects analysis?
thanks for all the help,
Kyle Kern
--
Best Regards, Donald McLaren
=====================
D.G. McLaren
University of Wisconsin - Madison
Neuroscience Training Program
Office: (608) 265-9672
Lab: (608) 256-1901 ext 12914
=====================
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 (608) 265-9672 or email.
IMPORTANT WARNING: This email (and any attachments) is only intended for the use of the person or entity to which it is addressed, and may contain information that is privileged and confidential. You, the recipient, are obligated to maintain it in a safe, secure and confidential manner. Unauthorized redisclosure or failure to maintain confidentiality may subject you to federal and state penalties. If you are not the intended recipient, please immediately notify us by return email, and delete this message from your computer.
--
____________________________________________
Thomas Nichols, PhD
Director, Modelling & Genetics
GlaxoSmithKline Clinical Imaging Centre
Senior Research Fellow
Oxford University FMRIB Centre
IMPORTANT WARNING: This email (and any attachments) is only intended for the use of the person or entity to which it is addressed, and may contain information that is privileged and confidential. You, the recipient, are obligated to maintain it in a safe, secure and confidential manner. Unauthorized redisclosure or failure to maintain confidentiality may subject you to federal and state penalties. If you are not the intended recipient, please immediately notify us by return email, and delete this message from your computer.
|