See if my statements below are right, Will:
though the test of rp>0 is equivalent to the test of the corresponding
regression coefficient (beta)>0, the test of rp_1 > rp_2' is not
equivalent to the test of 'beta_1 > beta_2'
Kewei
On Wed, 2 Feb 2005, Will Penny wrote:
> Dear Denise,
>
> If your GLM analysis used a design matrix with a single regressor,
> there is a 1 to 1 relationship between the correlation
> coefficient, r, (of regressor and data) and regression coefficient, beta,
> which is
>
> r = beta * s_x/s_y
>
> where s_x is the standard deviation (SD) of the regressor and
> s_y is the SD of the data.
>
> If beta is inferred to be significantly non-zero, using eg. a t/F-test,
> then r is guaranteed to be significantly non-zero.
>
> If your GLM analysis contains a design matrix with multiple
> regressors, as is usually the case, then there exists a one
> to one relationship between each regression coefficient
> and its corresponding *partial* correlation coefficient, rp, ie.
> the correlation between that regressor and the data after having
> subtracted from the data what can be explained by the other regressors.
>
> Again, if the regression coefficient is inferred to be significantly
> non-zero then rp will also be.
>
> But there's no easy way of getting the r/rp values out of SPM2.
> You'll have to matlab it.
>
> Best,
>
> Will.
>
> Denise Dörfel wrote:
>
>> Hi SPM users,
>>
>> I did a correlation analysis between the activation of a contrast and the
>> score of a questionnaire.
>> Is it possible to show the correlation coefficient or to get it out of
>> SPM2?
>>
>> In general, how can I get infos like this out of SPM2?
>>
>> Thanks in advance.
>> Denise Dörfel.
>>
>
> --
> William D. Penny
> Wellcome Department of Imaging Neuroscience
> University College London
> 12 Queen Square
> London WC1N 3BG
>
> Tel: 020 7833 7475
> FAX: 020 7813 1420
> Email: [log in to unmask]
> URL: http://www.fil.ion.ucl.ac.uk/~wpenny/
>
|