[VoxBo] rfx analysis

[marina bedny]

Hi Dan,

I just ran a test group rfx glm to make sure that I get the same  
answer whether I use a contrast vector specified in contrasts.txt vs.  
when I specify the vector myself. I don't. I'm pretty sure that using  
the contrasts.txt file gives you the wrong answer.

As part of my glm's I use derivatives as covariates of no interest,  
in terms of the order in which the derivatives appear in the g matrix  
they are interspersed with the covariates of interest. Each  
derivative follows it's covariate of interest. However, I'm not sure  
how it's treated by the vbmakeglm tool when applied to rfx. I think  
the derivatives (covariates of no interest) are put at the back of  
the line along with the other covariates of no interest. My newest  
guess is that voxbo only screws up by not putting the covariates of  
no interest in the back when the covariate of no interest is the very  
first added before any covariates of interest.

When I look at one of the .map files for running the analysis each  
way here is what I get:

using specified contrast: /jet2/mbedny/PolyHom/group/Inc-ConNTesbeta
contrast_vector: 0 0 0 0 0 0 0 0 0 0 -1 1 -1 1 0 0 0 0 0 0 0 0 0 0 0  
0 0 0 0 0 0 0 0 0 0 0

using contrasts.txt: /jet2/mbedny/PolyHom/group/Amb-ControlNTesbeta_test
contrast_vector: 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -1 0 1 0 -1  
0 1 0 0 0 0 0 0 0 0 0

the reason I think that the second one is incorrect is because all  
the .map files are empty, and because when I specify the contrast  
myself the results make perfect sense.

I think this is something quite important because if the contrast is  
not correct there is no error message. You can get results that have  
nothing to do with what you think you're actually testing.

Marina



On Jun 15, 2006, at 11:17 PM, Daniel Y Kimberg wrote:

> marina bedny wrote:
>> When creating rfx tes files vbprep does not skip over covariates of
>> no interest in applying the contrast vector. So if you happened to
>> add your covariates of no interest first for some subject, the order
>> of the contrast vector will not be right for these subjects. Most
>> people probably know this already, but I thought the info may be
>> useful for someone.
>
> Thanks for posting this.  I can barely keep track of these quirks
> myself, because we've been so inconsistent about which tools require
> complete contrast vectors and which just want the covariates of
> interest.  In part, this is one of the reasons for introducing the
> contrasts.txt file, which should work just fine with rfx scripts (use
> the name of the contrast instead of a vector, no quotes needed if
> there are no spaces in it).  Soon you'll be able to define your
> contrasts of interest at the same time you design your GLM, and if all
> goes well only ever refer to them by name (you can still edit them
> later if needed).
>
> I would definitely encourage everyone to use contrasts.txt in
> conjunction with the rfx scripts, except that I haven't tried it yet.
> I will try to do so tomorrow, but let me know if you have a bad
> experience with named contrasts and rfx scripts.  The contrasts you
> define in a contrasts.txt file should just have the covariates of
> interest in them, and you should get an error otherwise.
>
> dan