[VoxBo] vbdumpstats broken

[Evan Chen]

Hello Geoff,

I think this may be related to my vbdumpstats question from a few days ago.
In my case, the mean scaling and detrending options were somehow inserted
multiple times (as both on and off) in the .prm file, resulting in
vbdumpstats confusion.  Deleting the appropriate lines gave more appropriate
results.  Beyond that, Dan has indicated that he is taking a closer look at
the code for the cause of this issue and I'm including a portion of the
correspondence below.  I hope this is helpful.

Best,
Evan

>>>> Evan Chen wrote:
>>>> contrast=mycontrast scale=t effdf=13.00 stat=2804.147468434274
>>>> one-tailed_p=0.0000 two-tailed_p=0.0000
>>> ...
>> Thanks a lot for your suggestions!  Sadly, neither defining a contrasts.txt
>> file nor setting the contrast flag to "foo t vec 1" worked and both produced
>> the same t-value of 2804.  I did run the same ROI and contrast in IDL and
>> got a much more reasonable result:
>> 
>> Contrast: 1.0     * Effect
>> t:       -3.7184586
>> effdf:        13.000000
>> p(one tail):       0.99871113
>> 
>> Is there something simple that I could be missing here?
> 
> For some reason, your parameter file has multiple lines indicating
> whether or not it's mean scaled and/or detrended.  The first such
> lines indicate that it is (both) and the second such lines indicate
> that it isn't.  Here are the lines I mean:
> 
> DataScale:     mean
> Option: detrend
> ...
> DataScale:      none
> Option: none
> 
> The second is what you want for this kind of group (second tier)
> analysis.  From the script file, it looks like you had the GLM set up
> correctly, so I will have to take a look at the code and see how the
> file got that way.  Apparently vbdumpstats believes the first such
> lines, while the IDL voxel surfer believes the last.  Or something
> along those lines.  You can get vbdumpstats to do the right thing by
> editing the file to remove the first two lines, leaving intact the
> "none" lines.  Once you do that, it produces the same results as IDL
> (at least for the example you gave).  I will take a closer look at the
> code to see how things got this way, and to make sure it isn't causing
> other problems.
> 
> dan



On 8/31/06 2:43 PM, "Geoffrey K Aguirre" <aguirreg at mail.med.upenn.edu>
wrote:

> Dear Voxars ---
> 
> The vbdumpstats command line routine has a bug. It does not respect
> the settings in the .glm file regarding application of mean centering
> and drift correction, and defaults to "on". As a consequence, the
> values returned for group (random effects) analyses are messed up.
> 
> Any help would be appreciated,
> 
> Geoff
>

-- 
Evan Chen, PhD

Post Doctoral Research Fellow
University of Pennsylvania
3 West Gates Building
3400 Spruce Street
Philadelphia, PA 19104-4283

echen2 at mail.med.upenn.edu
Office: 215-614-0177