somebody can correct me if i'm wrong, but here's what i would do.
first, i would just run 2 simple regression models, one for each group.
then, i'd create a multiple regression model with 4 factors - 2 factors
identifying the groups (ie for one regressor, you put 1's for patients
and 0's for controls, and for the other, the opposite) and 2 containing
the RTs for each group (ie one regressor has the RT values for the
patients and 0's for the controls, and the other has the opposite). you
can then show areas in which one group shows a "more positive"
correlation or a "more negative" correlation with RT than the other
group. for instance, if your design matrix was [patients controls
patients_RT controls_RT], then [0 0 1 -1] shows regions in which
patients have a greater positive correlation with RT than do controls.
you can then pull out each subject's betas for any significant ROIs and
run post-hoc tests to determine the actual nature of the correlation
differences (ie positive in one group, negative in the other, or maybe
positive in one group, no correlation in the other).
dani
Daniel Simmonds
Developmental Cognitive Neurology
Kennedy Krieger Institute
[log in to unmask]
>>> Jiansong Xu <[log in to unmask]> 7/23/2007 12:49 PM >>>
Across both groups. How to use multiple regression?
Also, you mentioned "it is consistent with lots of models". Could you
give
me one or two examples?
Thanks
Jiansong
> From: Daniel Simmonds <[log in to unmask]>
> Reply-To: Daniel Simmonds <[log in to unmask]>
> Date: Mon, 23 Jul 2007 12:20:15 -0400
> To: <[log in to unmask]>
> Subject: Re: [SPM] simple regression
>
> hi jiansong,
>
> is the simple regression you are discussing in the patients, in
the
> controls, or across both groups? it may be that there are different
> correlations with RT for each group (ie positive correlation in
> controls/negative correlation in patients and vice versa) and this
could
> potentially indicate that the two groups are using this region
> differently (ie in one group, greater activation in this region leads
to
> faster RT's, while in the other, it leads to slower RT's). this type
of
> question could be answered using a multiple regression as opposed to
a
> simple regression.
>
> dani
>
> Daniel Simmonds
> Developmental Cognitive Neurology
> Kennedy Krieger Institute
> [log in to unmask]
>
>>>> Jiansong Xu <[log in to unmask]> 7/23/2007 12:11 PM >>>
> Thanks. But, the problem is:
>
> Relative to control subjects, patients showed longer RT and less
> activation
> in the lateral prefrontal cortex and less deactivation in the medial
> prefrontal cortex. I¹m glad about this finding and I can claim that
> the
> less activation and deactivation exhibited by patients correlated
with
> their
> longer RT than controls. Now, the RT positively correlated with the
> BOLD in
> the lateral prefrontal cortex and negatively correlated with signal
in
> the
> medial cortex. Such correlation is opposite to my above
interpretation
> of
> ³less activation and deactivation contribute to the longer RT².
>
> One possible interpretation for these ³conflicting² finding is that
> because
> of the ³less activation and deactivation² in some brain areas of
> patients or
> ³slower² performers, the remaining intact brain areas in patients
need
> to
> work longer to compensate for the impaired brain function.
>
>
> I have another study of healthy subjects (different population from
> above
> study) with different task. It also showed greater RT positively
> correlated
> with greater signal changes in the prefrontal and parietal cortex,
and
> subcortical area (thalamus and striatum) and negatively correlated
> with
> signal changes in the ³default brain area² (e.g., medial part of the
> brain).
> Follow your comments, these data suggest that the worse performers
> (i.e.,
> longer RT) showed greater BOLD signal increase in the positive
network
> and
> greater BOLD signal decrease in the negative network, thus showed
less
> functional efficiency in their brain. Is it reasonable?
>
> Best
>
> Jiansong
>
>
>
> From: "Weissman, Daniel" <[log in to unmask]>
> Reply-To: "Weissman, Daniel" <[log in to unmask]>
> Date: Mon, 23 Jul 2007 11:31:14 -0400
> To: <[log in to unmask]>
> Subject: Re: [SPM] simple regression
>
> Dear Jiansong,
>
> If I understand correctly, you've found a positive beta coefficient
in
> a
> simple (across-subjects) regression in which BOLD signal is
regressed
> against RT. In that case, the positive beta coefficient would
indeed
> mean
> that subjects who show larger changes in BOLD signal tend to exhibit
> longer
> RT. Although this finding goes against your prediction, it is
> consistent
> with lots of models. For example, longer RT may indicate greater
time
> on
> task, which results in more activity.
>
> Hope this helps,
> Daniel
>
>
> Dear Friend:
>
> I'm using simple regression to assess the correlation between BOLD
> signal
> changes and reaction time. Several clusters in the prefrontal and
> parietal
> cortex showed significant positive correlation between signal
changes
> and
> RT. Does this positive correlation indicate greater signal changes
> correlated with greater RT? If so, it is opposite to my expectation
of
> greater activity correlated with shorter RT. Any comments are
> appreciated.
>
> Best
>
> Jiansong
>
>
Disclaimer:
The materials in this e-mail are private and may contain
> Protected Health Information. Please note that e-mail is not
necessarily
> confidential or secure. Your use of e-mail constitutes your
acknowledgment of
> these confidentiality and security limitations. If you are not the
intended
> recipient, be advised that any unauthorized use, disclosure,
copying,
> distribution, or the taking of any action in reliance on the contents
of this
> information is strictly prohibited. If you have received this e-mail
in error,
> please immediately notify the sender via telephone or return e-mail.
|