Dear Dr Herath,

You certainly could model B as one regressor, and if your belief is that
the two B sessions are equivalent, then this would be the most appropriate
model.  If you wanted to allow for the possibility that the two B sessions
might be different (eg. if one occurred before A and one after A then there
might be a different response) then it would be worth the degrees of
freedom loss to model them separately.

If the two B sessions do turn out to be equivalent, then in your 'B as two'
model they would be assigned approximately the same parameter estimates, so
that the linear sum of these two columns in the design matrix would be
almost identical to the single column used in your 'B as one' model, and
the 'error' column, at the end of the least-squares fitting procedure,
would therefore also be almost identical in the two models.

I think that the only difference would be a small loss of statistical power
in the 'B as two' model because of the degrees of freedom, but I'm afraid
that I am no statistician,

Best wishes,

Richard Perry.

>Dear all,
>I have a question about t- contrasts. I fear that this might
>be a little naive,  but........
>Say,  I have a condition A with 20 scans and a condition B
>with 40
>scan ( because B was scanned twice within the session).
>        When modelling the data, can I model A  and B as one
>regressor each , or should B modelled as 2 regressors?
>In other words, should the contrast between A and B  be
> 1 -1 or
> 2 -1 -1 ( I understand that this model will allow me to ask
>more questions, but that particulr question is not something
>that I am interested in..)
>Is there a difference in the power of the contrasts?
>Do these contrasts answer the same question mathematically?
>Thank you very much.
>Priyantha Herath, MBBS
>Division of Human Brain Research,		tel:	 +46 (0)8 728 7298
>Department of Neuroscience,			    	 +46 (0)8 728 7297
>Karolinska Institute,
>S 171 77 Stockholm,				fax:	  +46  8 309045

from: Dr Richard Perry,
Clinical Research Fellow, Wellcome Department of Cognitive Neurology,
Darwin Building, University College London, Gower Street, London WC1E 6BT.
Tel: 0171 504 2187;  e mail: [log in to unmask]
Pager: 04325 253 566.