That is exactly an example of combining main effect with time effect, which - I also think - can be problematic in case of different scaling...
Another example could be an F-test to see whether time effect explains any additional variance.
Similarly, it could be also interesting - in some cases - to examine whether time effect (if there is any) is comparable with main effect (i.e. it may make sense statistically but negligible psychologically/behaviourally).
One may also want to compare different polynomial expansions (i.e. which model for time effect is the most significant/comparable with main effect).
By the way, is there any reason why we cannot scale modulator in a same way as main effect?
Vale,
Auer, Tibor M.D. Ph.D.
MRC Cognition and Brain Sciences Unit
15 Chaucer Road
Cambridge
CB2 7EF
United Kingdom
Phone/Work: +44-(0)1223-273613
Mail: [log in to unmask]
-----Original Message-----
From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]] On Behalf Of H. Nebl
Sent: Tuesday, March 10, 2015 6:28 PM
To: [log in to unmask]
Subject: Re: [SPM] Time modulation
Okay, the scaling issue is was I had assumed then. And thanks for the code. :-)
> As you only compare things alike (eg you wouldn't contrast a main effect (response amplitude) with a time effect)
Yes.
> Let me know if I am overlooking other situations
In some contexts it might be interesting to plot both the amplitude of the average response and the max. change in amplitude over time and possibly, also additively combine them. What I had in mind, say the predictor for average response is a stick function varying between 0 and 1. I had thought (which might well be wrong) that the predictor for a linear time modulation has max. values of 1 and -1 (for the beginning/end of the session) then. Obtaining a beta of 5 for the average and a beta of 3 for the TM would mean the amplitude at the beginning was 8 in beta units and 2 at the end (if onsets exactly matched the beginning/end). If one of the predictors were scaled differently then this would not hold any longer (but could easily be adjusted by multiplying with the required scaling factor, which is a little cumbersome though). But as stated, not sure whether these reflections make sense at all.
Best,
Helmut
|