Hi Michael,
I doubt that the loss of 1 DOF is causing the disappearance. In both
models, the test for the difference between groups occurs at a
particular age (age=0). This is not a factor in the DOSS model because
the lines for the two groups are forced to be parallel (so the
difference is the same at all ages). For the DODS, the lines are not
constrained to be parallel, so you might have a large or small
difference at age=0. If you de-mean the ages (ie, subtract the mean
computed over ALL subjects regardless of group), then your test will be
at age=MeanAge. This does not get around the problem, but you are at
least testing at an age where your data are. In expectation, it should
not make a difference, but in practice sometimes it does. If you run the
DODS and do not find a difference in the slopes, then you are justified
in re-running the fit using DOSS.
doug
On 3/9/11 6:55 PM, Christopher Bell wrote:
Michael,
Thanks for your response. It was helpful. As you suggested the
"thickness-age correlation group difference" under DOSS is
meaningless, so I have ignored that output.
I am currently left with a situation where under the DOSS model I see
a large ageXthickness effect (controlling for group) and a nearly
identical map under the DODS model. So far so good.
When I look at the group difference map, I see a large group
difference for the DOSS model, and a LOT less for the DODS model. This
would lead me to believe that the extra DOF in the DODS model is
removing the group difference because my two groups have different
ageXthickness slopes and when this is properly accounted for in the
DODS model the group difference largely goes away. However, when I
look at the " does the thicknessXage correlation differ between
groups" map I see almost nothing significant? I guess my hypothesis is
that the slopes are different enough between my groups to wash away
the group difference, but not large enough to show up as significant?
It would be nice to be able to derive some per group ageXthickness
slopes from clusters or ROIS to investigate this further, is this
possible? I will probably also start looking into other models, just
visually looking at different vertices it appears that most of the
groups difference is in the older subjects (in terms of raw thickness
values) with less/none difference in younger subjects.
One last aside. The demeaning of covariates still slightly confuses
me. In a DOSS model it seems it wouldn't matter where you measure the
intercept since both ageXthickness slopes are equal. In the DODS
model, it doesn't seem to make sense to measure an intercept of two
different slopes, it is more interesting to compare the slopes (in the
case of different slopes you will get a different answer at every
possible intercept). I am obviously a newbie to GLM and QDEC so sorry
for any silly questions.
Chris
On Fri, Mar 4, 2011 at 1:34 PM, Michael Harms <mha...@conte.wustl.edu
<mailto:mha...@conte.wustl.edu>> wrote:
Hi Chris,
There really shouldn't be a "thickness-age correlation group
difference"
result with the DOSS model. I have FS 4.1 (rather than 5.0 on my
system) but running an analogous model, I see that I do indeed get a
verbal "Description" for such a contrast. However, if I compare
that to
the "thickness-age correlation (accounting for group)" result, I
see the
exact same map. And if you look at the .mat files in the contrast
directory generated by qdec, you'll see that those two contrasts are
identical (i.e. [0 0 1]) (or at least they are for qdec with FS 4.1).
So, this appears to be a "bug" in the verbal descriptions that qdec
provides when using a DOSS model.
As to the group difference itself changing between the DOSS and DODS
models, that is totally to be expected. Note that in the DODS model,
whether or not you demean (center) the age variable has a critical
impact on the manner in which you interpret the group contrast.
cheers,
-MH
On Thu, 2011-03-03 at 20:41 -0600, Christopher Bell wrote:
> FreeSurfers,
>
> I have been analyzing my qdec data in version 5.0 and have some
> interesting although somewhat confusing results. Basically I
have run
> a very simple analysis with DODS and DOSS. My discrete factor is
group
> and my covariate is age, about as simple as can be.
>
> When I look at the results for DODS I get:
>
> thickness-age correlation (accounting for group)---result: much of
> brain significant
> group difference (I assume controlling for age, but it doesn't say
> explicitly)--result: one small roi significant
> thickness-age correlation group difference--result: one small roi
> spatially adjacent to group difference roi
>
> When I run DOSS I get:
>
> thickness-age correlation (accounting for group)---result: much of
> brain significant
> group difference (I assume controlling for age, but it doesn't say
> explicitly)--result: much of brain significant
> thickness-age correlation group difference--result: much of brain
> significant
>
> I am mostly surprised by how much larger the (group difference), and
> the (thickness-age correlation group difference) increase with the
> DOSS method. I am also not quite how to interpret the thickness-age
> correlation group difference in DOSS. I was thinking the DOSS method
> constrained both groups to have the "same slope" and so I was
> expecting to get nothing for difference in thickness-age correlation
> difference by group; isn't this suggesting my two groups have
> significantly different ageXthickness slopes even though they are
> constrained to have the same slope by the DOSS method? It would
almost
> make more sense to me if the results were reversed between DOSS and
> DODS. If I had a large thicknessXage correlation group difference
> using the DODS method which allows for different slopes. Thanks for
> any enlightenment.
>
> Chris Bell
> University of Minnesota
> _______________________________________________
> Freesurfer mailing list
> Freesurfer@nmr.mgh.harvard.edu
<mailto:Freesurfer@nmr.mgh.harvard.edu>
> https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
>
>
> The information in this e-mail is intended only for the person
to whom it is
> addressed. If you believe this e-mail was sent to you in error
and the e-mail
> contains patient information, please contact the Partners
Compliance HelpLine at
> http://www.partners.org/complianceline . If the e-mail was sent
to you in error
> but does not contain patient information, please contact the
sender and properly
> dispose of the e-mail.
_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
The information in this e-mail is intended only for the person to whom it is
addressed. If you believe this e-mail was sent to you in error and the e-mail
contains patient information, please contact the Partners Compliance HelpLine at
http://www.partners.org/complianceline . If the e-mail was sent to you in error
but does not contain patient information, please contact the sender and properly
dispose of the e-mail.