We now assume that a stage two random effects analysis is being done, and
that the data consist of one contrast image per subject.
It's my understanding that you can indeed do the stage two analysis in
PET/SPECT which allows you set up more complex non-time-series designs
including for instance more than one covariate of interest.
Ian
At 09:21 AM 12-11-03 -0500, Heidi Thermenos wrote:
>Thanks, but should I use the PET/SPECT menu for fMRI data?
>HWT
>
>From: Ian Nimmo-Smith <[log in to unmask]>
>To: Heidi Thermenos <[log in to unmask]>,[log in to unmask]
>Subject: Re: ANCOVA for fMRI: Group x Covariate Interaction
>Date: Wed, 12 Nov 2003 13:49:30 +0000
>
>Heidi
>
>To set up your model I think you need to go via the PET/SPECT menu to the
>Multi-Subject Conditions + Covariates design. You will have two covariates
>of interest: Cov (original) and Cov*Group (interaction with dummy variable).
>But someone more SPM-conversant may be able to help you.
>
>Ian
>
>At 07:30 AM 12-11-03 -0500, Heidi Thermenos wrote:
>>Hi all,
>>I am using SPM99 for fMRI data, and would like use a 2group ANCOVA to test
>>the effect of group,
>>removing the effects of
>>(1) a covariate, and
>>(2) covariate x group interaction.
>>
>>How do I enter a 2nd covariate into an fMRI ANCOVA in Basic Models?
>>Should I use t- or F-contrasts?
>>
>>And, can someone help me with these contrasts below?
>>
>>group1 group2 mu covariate
>>-1 1 0 0 (effect of group on signal)
>>
>>For the effect of covariate on signal, indep. of group:
>>0 0 1 1 (is this legal? and yet it seems parallel to the contrast you
>>would
>>specify in 'Correlation'
>>)
>>
>>
>>For the Group x covariate interaction?
>>? ? ? ?
>>-1 1 1 1
>>
>>
>>
>>
>>From: Ian Nimmo-Smith <[log in to unmask]>
>>To: Heidi Thermenos <[log in to unmask]>,[log in to unmask]
>>Subject: Re: ANCOVA: homogeneity of regression lines across groups
>>Date: Mon, 10 Nov 2003 10:53:37 +0000
>>
>>At 02:16 AM 10-11-03 -0500, Heidi Thermenos wrote:
>>>Hi,
>>>I am using SPM99 to analyze fMRI data (aquired using a blocked-design):
>>>
>>>I am conducting a 2-group ANCOVA, to determine whether my findings are
>>>attributable to a potential confounding variable, which I entered as a
>>>covariate.
>>>
>>>I would like to know if my data meet the assumptions necessary for valid
>>>use of ANCOVA,
>>>e.g., homogeneity of regression lines across two groups I am comparing.
>>>
>>>Is there a way to view the regression lines within SPM and determine if
>>>they are homogenous?
>>
>>I will leave methods to 'View' to others. You can determine homogeneity by
>>including a Group x Covariate interaction term. This is an additional
>>covariate which is constant (typically but not necessarily 0) on one group
>>and equal to your confounding variable (possibly mean-corrected) on the
>>other group. The high and low values in the corresponding beta map tell you
>>about non-homogeneity.
>>
>>Ian
>>
>>_________________________________________________________________
>>Send a QuickGreet with MSN Messenger
>>http://www.msnmessenger-download.com/tracking/cdp_games
>
>_________________________________________________________________
>Great deals on high-speed Internet access as low as $26.95.
>https://broadband.msn.com (Prices may vary by service area.)
|