So if two or more covariates are correlated, you may miss the interested
effects by putting them in the same model without orthogonization. For
example, x1=x2+n, without orthogonization, the contrasts for just x1 or x2
won't give you the effects revealed by just regressing images to x1 or x2
independently. You can read Andrade et al 1999, Neuroimage 10:83-86.
---------------------------------------------
Ze Wang, Ph. D
Center for functional neuroimaging,
Dept of Neurology,
Medicine school,
University of Pennsylvania
3400 Spruce Stree,
3W Gates building,
Philadelphia, PA 19104
Tel: 215-662-7341
-----Original Message-----
From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]] On
Behalf Of Jiansong Xu
Sent: Wednesday, January 30, 2008 8:27 AM
To: [log in to unmask]
Subject: Re: [SPM] multiple correlaltion
Thanks.
I have seen the term "orthogonization" many times in different
contexts, but do not understand it. Could you or any one else on the
list explain the meaning of this term?
Best
Jiansong Xu, M.D., Ph.D.
Associate Research Scientist
Dept. of Psychiatry
Yale School of Medicine
Connecticut Mental Health Center, Room S-100
34 Park St.
New Haven, CT 06519
On Jan 29, 2008, at 11:27 PM, Ze Wang wrote:
> HI Jiansong,
> This is right to me. In this model you can test the difference
> between the
> activations of control and patients, and also assess the difference
> between the
> RT correlations with the activations between control and patients.
> Regarding to
> your first email, there is no difference between centered RT and
> original RT for
> checking the correlation. But for multiple covariates, it will be
> different for
> w or w/o orthogonization. You can refer to the newest SPM book
> (Friston et al
> 06) for a detailed description.
>
>
> Ze
> Quoting Jiansong Xu <[log in to unmask]>:
>
>> Dear Friends:
>>
>> I'm trying to assess if there is a difference between control and
>> patients in correlations between RT and changes in BOLD signal. I'm
>> using multiple correlation with constant. Could any one tell me if
>> following procedure is correct:
>>
>> 1. select multiple correlation with constant
>> 2. select the con images of the control and patients
>> 3. select 4 covariates.
>> 4. enter 1 1 1 1 ... 0 0 0 0 ...
>> 5 enter 0 0 0 0 ... 1 1 1 1...
>> 6. enter RT and 0 0 0 0...
>> 7 enter 0 0 0 0 ... and RT
>>
>>
>> Best
>>
>> Jiansong Xu, M.D., Ph.D.
>> Associate Research Scientist
>> Dept. of Psychiatry
>> Yale School of Medicine
>>
>> Connecticut Mental Health Center, Room S-100
>> 34 Park St.
>> New Haven, CT 06519
>>
>
>
> --
> Center for functional neuroimaging,
> Medicine School,
> University of Pennsylvania,
> 3400 Spruce Street,
> 3W Gates Building,
> Philadelphia,
> PA 19104, USA
> tel:215-662-7341
|