Dear Panda Panda.... ;-)
> We have a 4 condition block-design experiment, in which we would like to
> look at several contrasts. Our questions are as follows:
> 1. Which is the best contrsat in order to see a difference between two
> conditions, e.g., the 1st and 2nd:
> t contrast: 1-100
> or F contrast: 1000;0-100 (Is that a valid contrast for testing a
> directional difference?)
Only your T-contrast tests for the difference between conditions 1 and 2.
And it only tests for condition 1 > condition 2.
Your F-contrast tests for activations in condition 1 "and/or" condition 2
(the +1/-1 sign of the contrast vectors is irrelevant).
If you want a bidirectional test of the difference between conditions 1 and 2,
simply test the F-contrast:
[1 -1 0 0]
(ie same "vector" as your T-contrast, but F rather than T).
> 2. Our experiment is based on a parametric hypothesis, according to which
> cond1<cond2<cond3<cond4. What would be the best way to model this hypothesis
> in a contrast?
Well, the obvious answer is to test the T-contrast:
[-3 -1 1 3]
which looks for an increasing linear trend across conditions 1-4.
If you want to test the more specific hypothesis that cond1<cond2
AND cond2<cond3 AND cond3<cond4, you would have to try
a conjunction of T-contrasts [-1 1 0 0], [0 -1 1 0] and [0 0 -1 1].
Rik
|