Dear Dorian and Jason,
> If you take the contrast:
> c = [3 2 1 2 3]
> and then subtract the mean you get the contrast:
> c = c - mean(c);
> c = [0.8 -0.2 -1.2 -0.2 0.8]
> This contrast tests for an equal difference between each adjacent
> condition and has a sum of zero.
this approach would identify a 'V' shaped contrast. For a
neurobiological system this might not be quite what you want.
If you wanted to capture a 'U'-shaped function, indicative of a typical
Yerkes-Dodson type neuropsychological or psychopharmacological function,
then I would suggest you stick to the mean corrected quadratic instead,
eg. [2 -1 -2 -1 2], as you originally had it.
In practice, for five levels, the corelation coefficient between [2
-1 -2 -1 2] and [0.8 -0.2 -1.2 -0.2 0.8] is ~0.95 , so it is
unlikely to make much difference.
best wishes,
James
>
>
> I want to catch the brain areas that are activated like a U-shape in 5
> conditions. In practice the two extremes would have a value of 1, and
> the one in the midle -1, However the two remaining values would be
> -0.5 each:
>
> [1 -0.5 -1 -0.5 1]
>
> However this doesn't reflect the change I want to catch, because the
> second condition changes from the first one much more (1.5) then the
> second from the third (0.5).
>
> Anybody had this problem before? What numbers would catch the signal
> change in the attached example?
>
> Regards.
> Dorian
> Ruhr University of Bochum
>
>
--
--------------------------------------
Dr James Rowe
Senior Clinical Research Associate and
Consultant Neurologist,
Cambridge University Department of Clinical Neurosciences,
Box 83, R3 Neurosciences,
Addenbrooke's Hospital,
Cambridge,
CB2 2QQ
UK
Tel: +44 (0)1223 273630
|