Hi Mike,
attached you find a function that I use for calculating such statistics. It
specifies the designmatrix for 2 groups and more than one image per subject.
Variance components are also specified for each group independently. When I
compared contrasts in such a design with corresponding t-tests I found that
results were quite similar at the same p-value thresholds (whatever this
means....).
But probably you also want to have a look at Rik's recommendation posted last
month regarding a similar problem:
http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0506&L=SPM&P=R23714&I=-3&X=60ECB45C3CDB56D29F&Y=thilo.kellermann%40uni-duesseldorf.de&m=19597
Best wishes,
Thilo
On Thursday 28 July 2005 17:02, Alexa Morcom wrote:
> Mike -
>
> If you do a one-way ANOVA with 4 conditions then strictly speaking you
> won't be able to specify the right variance components for the
> nonsphericity 'correction' in SPM2 (this may not be true of SPM5 but I'm
> not sure if it's 'in' yet-?) - however you could do it by hand and adjust
> SPM.xVi.Vi, or you could simply select the full nonsphericity options as if
> both factors were within subjects, and hope that the estimated covariances
> between the conditions that are from different subjects is near zero (as
> they should be zero!).
>
> Here and in the regression model (in which I don't think there's a
> nonsphericity option) your T-contrast of [-1 1 1 -1] would give you the
> interaction where drug - placebo effects are bigger in the controls than in
> the patients. You'd need to do [1 -1 -1 1] for the other direction. (I'm
> not sure what you mean by 'AND' here)
>
> The other way to do it with no compromise or hacking is simply to do the
> drug-placebo contrasts at the first level as you originally suggested (if
> you have put them into a single design matrix for each subject as 2
> sessions) and take this up to the 2nd level for 1- (for effects within
> group or overall) and 2- (for group differences) sample t-tests . This is
> slightly different from putting everything in a 2nd level mixed ANOVA
> because it effectively tests each effect against its own error term - see
> various postings by Rik Henson, e.g.
> http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0507&L=SPM&P=R11001&I=-3&X=2
>9 B1D5211551440611&Y=amm96%40cam.ac.uk for more detail - but either is
> correct, that's just a matter of preference.
>
> Hope this helps
>
> Alexa
>
> | -----Original Message-----
> | From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]]On
> | Behalf Of Mike Angstadt
> | Sent: 28 July 2005 15:28
> | To: [log in to unmask]
> | Subject: Re: [SPM] mixed between/within subject RFX
> |
> |
> | Okay, so the suggestions made to me included either performing a multiple
> | regression with columns for diagnosis, treatment, and the interaction, or
> | performing a one-way ANOVA with 4 conditions.
> |
> | What do SPMers think is the more appropriate way to get close to a 2-way
> | ANOVA result?
> |
> | In the regression case with 3 columns:
> | first: 1's for Drug, -1's for Placebo
> | second: 1's for Patients, -1's for Controls
> | third: first*second interaction
> | If I do a T-contrast of [1 0 0] that should show activations
> | where Drug was
> | greater than Placebo correct?
> | what about a T-contrast of [0 0 1]? Is that (Drug > Placebo) AND
> | (Patient > Control), or does it just show areas where there was a
> | differential effect of treatment based on diagnosis (without letting me
> | know the direction)?
> |
> | For the One Way ANOVA with 4 conditions:
> | Condition 1: Patients_Drug
> | Condition 2: Patients_Placebo
> | Condition 3: Controls_Drug
> | Condition 4: Controls_Placebo
> | In order to get for example (Patients > Controls) AND (Placebo > Drug)
> | would I use a T-contrast of [-1 1 1 -1]? (basically doing a double
> | subtraction, (-1 +1) - (-1 +1))
> |
> | Thanks
> | -Mike
> |
> | At 10:55 AM 7/27/2005 -0400, Andrew J. Saykin wrote:
> | >Mike,
> | >
> | >You could also use a regression model with main effects for
> |
> | diagnosis and
> |
> | >treatment and include an interaction term for diagnosis X treatment
> | >effects since these are likely of particular interest in your study. As
> | >an example, we used this approach in our donepezil study (Brain 2004;
> | >127:1574-83).
> | >
> | >Andy Saykin
> | >
> | >At 10:23 AM 7/27/2005, Mike Angstadt wrote:
> | >>Hi,
> | >> We're looking for the best way to analyze a design like the
> | >> following:
> | >> - Two groups of subjects (ex: patients and controls)
> | >> - two levels of treatment (ex: drug and placebo)
> | >> Each subject has 2 scans from separate days. One for the drug
> | >> condition and one for placebo. What's the easiest and/or best way to
> | >> look at the group RFX?
> | >> One idea we had was to just model both scans in one
> |
> | large design
> |
> | >> matrix (as in this post
> |
> | http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0502&L=SPM&D=0&I=-3
> | &X=334B9904DDB400BF3B&Y=&P=15311)
> |
> | >> But we weren't sure if that was the best way.
> | >> Since SPM2 doesn't offer a 2-way ANOVA, what's the best option
> | >> to run something like that? Is there a script someone has to
> |
> | make SPM2
> |
> | >> do a 2-way? Or could I fake it with a one-way ANOVA with 4 conditions
> | >> (ex: patient_drug, patient_placebo, control_drug, control_placebo)?
> | >>
> | >> Thanks
> | >>
> | >>-Mike
> | >
> | >========================================================
> | >Andrew J. Saykin, PsyD, ABPP
> | >Professor of Psychiatry and Radiology
> | >Director, Neuropsychology Program and Brain Imaging Laboratory
> | >Department of Psychiatry - DHMC
> | >Dartmouth Medical School
> | >Lebanon, NH 03756
> | >
> | >Tel. (603) 650-5824
> | >Fax (603) 650-5842
> | >
> | >email:[log in to unmask]
> | >
> | >http://synapse.hitchcock.org
> | >
> | >========================================================
> | >
> | >IMPORTANT NOTICE REGARDING THIS ELECTRONIC MESSAGE
> | >This message is intended for the use of the person to whom it is
> |
> | addressed
> |
> | >and may contain information that is privileged, confidential, and
> | >protected from disclosure under applicable law. If you are not the
> | >intended recipient, your use of this message for any purpose is strictly
> | >prohibited. If you have received this communication in error, please
> | >delete the message and notify the sender so that we may correct
> |
> | our records.
--
Thilo Kellermann
RWTH Aachen University
Pauwelstr. 30
52074 Aachen
Tel.: +49 (0)241 / 8089977
Fax.: +49 (0)241 / 8082401
E-Mail: [log in to unmask]
|