Re: [Marsbar-users] MarsBaR SPM graph

[Tobias O. <t.o...@ny...>]

Hi Matthew,

Thanks a lot for working through the list and getting back to me on  
that, I do appreciate it.


On Jul 2, 2010, at 8:40 AM, Matthew Brett wrote:

> Hi,
>
> Sorry to be late to reply - it's been a heavy month for traveling.
>
>> I have a 2x4 factorial design and want to plot the contrast  
>> estimates within
>> a given ROI using an F-contrast. For example, I want to use two  
>> kinds of
>> F-contrasts:
>> 1) to look at the effects for each condition individually:
>> eye(8)
>>
>> and
>> 2) to look at one main effect:
>> 1 1 1 1 0 0 0 0
>> 0 0 0 0 1 1 1 1
>>
>>
>> Now, what seems to be the case is that marsbar is plotting the  
>> cbeta values
>> in both cases, while if I do the same plots with SPM (e.g. by  
>> restricting
>> the ROI to be one voxel), then SPM will plot the beta values in the  
>> first
>> case, but the cbeta values in the second. Why do SPM and MarsBaR use
>> different beta values for the first type of contrast? (In fact,  
>> using the
>> cbeta values in the first case yields 'incorrect' results as far as  
>> I can
>> tell, i.e. the profile of the bar plot does not make sense with  
>> respect to
>> what's actually going on in that voxel across the different  
>> conditions.)
>
> I'm very sorry to express ignorance (I don't have matlab on my
> computer at the moment) - but how would you expect the cbetas and the
> betas to differ for the first contrast?

I would also have thought they would be the same in the first case,  
but they aren't. As far as I can tell from investigating this a bit  
further, it is the default 'effects of interest' F-contrast that  
marsbar would use that is 'at fault', since in this case this 'effects  
of interest' isn't a straight eye matrix but some weighted version? If  
I build my own 'effects of interest' F-contrast with the eye matrix,  
then beta and cbeta are obviously identical. Why does marsbar use a  
weighted eye matrix for the 'effects of interest'?

>
>> Also, would it be possible to implement a routine that spits out as  
>> a Matlab
>> variable what is being plotted, in particular the 90% C.I.s  (at  
>> the moment,
>> I have to 'edit' the plot and then get the 90% C.I. value for each  
>> condition
>> separately via the Matlab 'Inspector'). Would standard errors of  
>> the mean
>> also be possible instead of 90% C.I.'s?
>
> You could try adding the CIs to the output structure.  I mean
> something like changing the end of mars_spm_graph.m from:
>
> % Complete return values
> r_st = mars_struct('fillafromb', def_r_st, struct(...
>    'Y', Y, 'y', y, 'beta', beta, 'Bcov', Bcov, ...
>    'cbeta', cbeta, 'PSTH', PSTH));
>
> to:
>
> % Complete return values
> r_st = mars_struct('fillafromb', def_r_st, struct(...
>    'Y', Y, 'y', y, 'beta', beta, 'Bcov', Bcov, ...
>   'CI', CI,
>    'cbeta', cbeta, 'PSTH', PSTH));

Yes, thanks, I have gone into the code and adjusted it to my needs.

>
> Best,
>
> Matthew


--
Tobias Overath, PhD
Department of Psychology
New York University
6 Washington Place, rm 275-276
New York, NY 10003
USA
Telephone: +1 (212) 998-8072  (internal: 88072)
http://homepages.nyu.edu/~to511