 |
 |
Hi Anderson,
Thanks for the quick reply!
Best,
Renske
From: FSL - FMRIB's Software Library [mailto:[log in to unmask]]
On Behalf Of Anderson M. Winkler
Sent: woensdag 6 mei 2015 7:59
To: [log in to unmask]
Subject: Re: [FSL] interaction terms
Hi Renske,
On 5 May 2015 at 16:49, Renske <[log in to unmask]> wrote:
Hi,
I have a question regarding interaction terms in FEAT. I'm analyzing an experiment in which subjects were presented with 3 types of faces preceded by 2 types of primes (primes were presented for 16ms about 150ms before the face; all combinations of faces and
primes were used). I've set-up a model that includes EVs for each face and each prime type, as well as the interaction terms (so 11 EVs: 2 primes, 3 faces, 6 prime*face interaction terms). Testing for main effects of primes and faces is no problem, but I'm
still thinking about what contrasts to include to fully test for a prime*face interaction. Testing beta's for each interaction term against 0 will find areas that show 'special' activation to a specific combination of prime and face (non-additive effects of
being a particular face presented with a particular prime). I was wondering about two things: 1) Would an F-test across these 6 individual contrasts adequately test whether there is ANY prime*face interaction?
Yes, although it may not be informative. Perhaps more informative would be each of the interaction seen individually, and this is probably how I'd analyse.
And 2) Would it make sense to compare beta's for these interaction terms (e.g. a 1-1 contrast for Face1Prime1 Face1Prime2) or would those results be meaningless?
It has a meaning, but I'd say it's too difficult to interpret and to reflect something of biological significance. If you aren't sure about what it would mean, then it's probably not interesting for your hypothesis
anyway. I'd leave aside.
All the best,
Anderson
Thanks!
Renske