Re
> Hi
>
> Thank you for your help.
> I did not mean F is a bug. I meant defining t-contrast [-1 -1 2] or [-1 -1
> 1] worked well before. Now it does not. I can define t-contrast, however,
> with two columns (e.g.: [-1 0 1]), but with three I cannot.
> I know these other opportunities (f-test, conjunction). I am just curious
> what is the problem with this kind of t-contrast I want to use.
>
well a t-test [-1 -1 1] isn't valid (doesn't sum up to zero) that's
probably why ..
Best
cyril
>> I have set up an ANOVA-design with three levels (see Design Matrix
>> attached), and I try to calculate a T-contrast to "substract" the
>> fisrt two levels from the third [-1 -1 2].
>> With former revision of spm worked well, but now it says 'trying to
>> define a t that looks like an F'.
>> Is it a bug, or did I something wrong?
>>
> Not sure but I guess it a feature :-)
>
> Unless you have strong hypothesis on the effect size of the col 3 (col3
> > than average col1 and col2) you would perform an F like [-1 0 1; 0 -1
> 1] which tests if col3 is > col1 and/or col2 - or if you want even
> stronger do [-1 0 1] and [0 -1 1] and perform a conjunction null =
> col3>col1 AND col3>col2
>
> Hope this helps
> cyril
>
>
|