Re: [Marsbar-users] Calculating % signal change for block design

[Matthew B. <mat...@gm...>]

Hi,

On Tue, Apr 1, 2014 at 12:32 PM, Talia Raney <tal...@bu...> wrote:
> Hi,
> I figured out the problem with the design path. So here are my beta
> extraction from the raw data
> Based on my previous emails about the issues with exact % signal change
> across English and Spanish conditions:
> Hi,
>
> "On Thu, Jan 30, 2014 at 11:57 AM, Talia Raney <tal...@bu...> wrote:
>> Hi,
>> The beta values are the same across all patients, which is strange because
>> in previous analyses the matrix values were different
>> Here are the dimensions: [61 73 61]
>> These are the values for the matrix that are the same across the 4
>> patients:
>> -3 0 0 93
>> 0 3 0 -129
>> 0 0 3 -75
>> 0 0 0 1
>> The technician in my lab couldn't understand why either and so maybe that
>> is
>> contributing to not statistically different results(For Example: BOLD
>> signal
>> for LIFG tri for English and Spanish were 2.748 and 2.731). Since all the
>> patient vary in the level of bilingualism, it shouldn't happen.
>> I will read up some more about the beta estimates.
> I am confused - that looks like the image affine.  I was thinking of
> the beta values from the regression.   These are the beta* images in
> SPM, and the .beta field in marsbar.  There's an example of beta
> extraction in marsbar here:"
>
> This is the beta values for LIFGtri.
> -0.441987835592497
> -17.6631781046321
> -0.00584444804595380
> -18.1993048739651
> -1.41879647244602
> -13.9670507802562
> -0.764194761758106
> -16.8519803415051
> -2.79227983932455
> -1.60086419296314
> -0.804718943608214
> 4.14987893074880
> 289.420272008739
> 149.868494102840
> 10.2935059115606
> 28.1044882261369
> 10.1211746881779
> 26.6752138395511
> 9.83882094731138
> 28.7118684627458
> 9.78124614493613
> 28.1860180483175
> -5.57499518184095
> -2.65210862167349
> -0.611257093313648
> -79.5111551830495
> 137.309012987820
> 296.308725567405
> 186.455870630854
> 175.695321504058
>   What would be the next step into figuring out the problem for these
> insignificant differences in % signal change?

I think you were worried about the beta values being the same across
patients, but I think we established that this was the affine.

So - with this list of betas - what should I be looking at?  I guess I
would not be surprised if the betas were fairly similar across
conditions and patients, but there's probably a bug if the values are
exactly the same.

Don't forget that "most published research findings are false" [1] -
and neuroimaging studies have low power [2] so it wouldn't be very
surprising if you failed to find the same significant result as one or
even more than one previous study.

Cheers,

Matthew

[1] http://www.plosmedicine.org/article/info%3Adoi%2F10.1371%2Fjournal.pmed.0020124
[2] http://www.nature.com/nrn/journal/v14/n5/full/nrn3475.html