[MINC-users] glim_image
Jason Lerch
minc-users@bic.mni.mcgill.ca
Mon Apr 11 16:11:04 2005
--Apple-Mail-16--348434962
Content-Transfer-Encoding: quoted-printable
Content-Type: text/plain;
charset=ISO-8859-1;
format=flowed
Greetings again,
no, you cannot use the results of the regression without an intercept=20
unless you have a good a priori reason of eliminating the intercept=20
from your model - which it doesn't seem like you have. Sorry!
Jason
On Apr 11, 2005, at 3:58 PM, Jamila Ahdidan wrote:
> Dear Jason,
> Thanks for your explanation. It help a lot!
> I have one (or 2) more question. The results from a glim_image with a=20=
> matrix an intercept and the one without intercept are the same, except=20=
> from the range of t values that is bigger without intercept, which=20
> allows me to define highly significant areas (even after the very=20
> conservative bonferroni correction). So, I wonder whether the results=20=
> from the glim without intercept can be used to give an estimate of=20
> what could be found if we had a bigger sample size. Do you think I can=20=
> use the results from the glim without intercept in some way? and if=20
> yes how?
> =A0
> I hope you'll find the time to answer!
> =A0
> Many regards,
> Jamila
>
>
> Jason Lerch <jason@bic.mni.mcgill.ca> wrote:
> Greetings again,
>
> I've attached a graph which shows the difference - the data was
> generated using the following function:
>
> y =3D 2 + 0.5x + Error
>
> where x is a simple sequence between 0 and 10.
>
> There are two regression lines through the graph. The one in blue fits
> a model including the intercept term, the one in red fits a model
> without an intercept. You can see that the red line takes on a value=20=
> of
> 0 when x=3D0 - and that therefore the fit is not as accurate as the =
blue
> line.
>
> None of this stuff is specific to glim_image - this is all standard
> linear model statistics. One of the best online references that I =
know
> about is here:
>
> http://www.itl.nist.gov/div898/handbook/
>
> though there surely are others as well.
>
> Good luck,
>
> Jason
>
>
>
> On Apr 10, 2005, at 3:41 AM, Jamila Ahdidan wrote:
>
> > Hi Jason,
> > Well I have to say that I don't use glim_image to perform a VBM=20
> study,
> > but to perform a t test at each voxel to assess the difference=20
> between
> > my group of patients and my group of controls. So, I don't really=20=
> know
> > whether I want to force to be 0 at x=3D0. (I don't really know what=20=
> that
> > means!).
> > Do you think I'm using the wrong minc command, and if yes do you =
have
> > another idea?
> > =A0
> > Thanks a lot,
> > Jamila
> >
> > Jason Lerch wrote:
> >
> > On Apr 9, 2005, at 6:51 PM, Jamila Ahdidan wrote:
> >
> > > My dilema is whether I should just keep
> > > my good results and forget about the intercept in the
> > > matrix, or I should stick to the intercept and
> > > conclude that nothing is interpretable from my
> > > results.
> >
> > Is there any reason to force the slope to be 0 at x=3D0? If you have
> > standard VBM density data, then that is an invalid assumption, since
> > there is every reason to allow the y to take on an arbitrary value =
at
> > x=3D0, so you would include that column of ones for your intercept=20=
> term.
> > If you have different data then this assumption might be valid -
> > something that is the case, for example, when looking at asymmetry=20=
> VBM.
> > But by and large you will want an intercept.
> >
> > Hope this helps,
> >
> > Jason
> >
> > _______________________________________________
> > MINC-users@bic.mni.mcgill.ca
> > http://www.bic.mni.mcgill.ca/mailman/listinfo/minc-users
> >
> > Do you Yahoo!?
> > Yahoo! Mail - Find what you need with new enhanced search. Learn=20
> more.
>
> Do you Yahoo!?
> Yahoo! Small Business - Try our new resources site!=
--Apple-Mail-16--348434962
Content-Transfer-Encoding: quoted-printable
Content-Type: text/enriched;
charset=ISO-8859-1
Greetings again,
no, you cannot use the results of the regression without an intercept
unless you have a good a priori reason of eliminating the intercept
from your model - which it doesn't seem like you have. Sorry!
Jason
On Apr 11, 2005, at 3:58 PM, Jamila Ahdidan wrote:
<excerpt>Dear Jason,
Thanks for your explanation. It help a lot!
I have one (or 2) more question. The results from a glim_image with a
matrix an intercept and the one without intercept are the same, except
from the range of t values that is bigger without intercept, which
allows me to define highly significant areas (even after the very
conservative bonferroni correction). So, I wonder whether the results
from the glim without intercept can be used to give an estimate of
what could be found if we had a bigger sample size. Do you think I can
use the results from the glim without intercept in some way? and if
yes how?
=A0
I hope you'll find the time to answer!
=A0
Many regards,
Jamila
<bold><italic>Jason Lerch <<jason@bic.mni.mcgill.ca></italic></bold>
wrote:
Greetings again,
I've attached a graph which shows the difference - the data was
generated using the following function:
y =3D 2 + 0.5x + Error
where x is a simple sequence between 0 and 10.
There are two regression lines through the graph. The one in blue fits
a model including the intercept term, the one in red fits a model
without an intercept. You can see that the red line takes on a value
of=20
0 when x=3D0 - and that therefore the fit is not as accurate as the blue
line.
None of this stuff is specific to glim_image - this is all standard
linear model statistics. One of the best online references that I know
about is here:
http://www.itl.nist.gov/div898/handbook/
though there surely are others as well.
Good luck,
Jason
On Apr 10, 2005, at 3:41 AM, Jamila Ahdidan wrote:
> Hi Jason,
> Well I have to say that I don't use glim_image to perform a VBM
study,
> but to perform a t test at each voxel to assess the difference
between
> my group of patients and my group of controls. So, I don't really
know
> whether I want to force to be 0 at x=3D0. (I don't really know what
that
> means!).
> Do you think I'm using the wrong minc command, and if yes do you have
> another idea?
> =A0
> Thanks a lot,
> Jamila
>
> Jason Lerch wrote:
>
> On Apr 9, 2005, at 6:51 PM, Jamila Ahdidan wrote:
>
> > My dilema is whether I should just keep
> > my good results and forget about the intercept in the
> > matrix, or I should stick to the intercept and
> > conclude that nothing is interpretable from my
> > results.
>
> Is there any reason to force the slope to be 0 at x=3D0? If you have
> standard VBM density data, then that is an invalid assumption, since
> there is every reason to allow the y to take on an arbitrary value at
> x=3D0, so you would include that column of ones for your intercept
term.
> If you have different data then this assumption might be valid -
> something that is the case, for example, when looking at asymmetry
VBM.
> But by and large you will want an intercept.
>
> Hope this helps,
>
> Jason
>
> _______________________________________________
> MINC-users@bic.mni.mcgill.ca
> http://www.bic.mni.mcgill.ca/mailman/listinfo/minc-users
>
> Do you Yahoo!?
> Yahoo! Mail - Find what you need with new enhanced search. Learn
more.
Do you Yahoo!?
Yahoo! Small Business - <color><param>0000,0000,EEEE</param>Try our
new resources site!</color></excerpt>=
--Apple-Mail-16--348434962--