John Ashburner wrote:
> Dear Barbera,
>
>
>>Thanks for getting back to me, John. I just redid the analyses (yet again)
>>and the results for the simple regression do differ from those for the
>>multiple regression with a single covariate when I use the same design
>>matrices and input values. Significant -- and interesting -- results in
>>one, none in the other.
>
oh dear !
following John's idea though - are the design matrices really the same ?
Perhaps you could get the variables X1=SPM.xX.X for model 1
and X2=SPM.xX.X for model 2, save them in a mat file and email them to me.
>
> I don't really know about this part of SPM, so I've cc'd this email to Will in
> case it is some sort of bug.
>
>
>>I will query the listserv. If you happen to think of a possible explanation
>>later, please do let me know.
>>
>>And just out of curiosity, why are there two options? Is the simple
>>regression left over from spm99?
Don't know.
But in SPM5 there's only multiple regression.
Best,
Will.
>>thanks again,
>
>
> I'm not actually sure.
>
> Best regards,
> -John
>
>
>>At 11:15 AM 3/2/2006 +0100, you wrote:
>>
>>>Dear Barbara,
>>>
>>>
>>>>In SPM2, I am getting significantly different results using "simple
>>>>regression (correlation)" than with "multiple regression" with one
>>>>covariate. Is this a bug? Which method is to be preferred?
>>>
>>>I'm not actually sure of the answer here. Are the design matices and
>>>everything else identical in the two models? If they are, then I would
>>>expect identical results. If not, then I would expect the results to be
>>>different.
>>>
>>>
>>>>Apologies if this is the wrong venue for this question.
>>>
>>>The stats part isn't really my area of expertise (I'm the guy who gets the
>>>[log in to unmask] emails), but you may get some more authoritative
>>>advice if you email the mailing list.
>>>
>>>Best regards,
>>>-John
>
>
>
>
--
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/
|