Dear Will Penny,
may I trouble you with an additional ANOVA question:
I try to set up a 2x3 within ANOVA with partitioned errors as described in
"ANOVAs and SPM" (Henson and Penny, 2003).
The design matrix at the first level consists of the following regressors:
A1B1 A1B2 A1B3 A2B1 A2B2 A2B3
I used the following contrasts at the first level:
main effect A: [-1 -1 -1 1 1 1]
main effect B:
[-1 1 0 -1 1 0]
and
[0 -1 1 0 -1 1]
interaction AxB:
[1 -1 0 -1 1 0]
and
[1 0 -1 -1 0 1]
On the second level I used a one-sample t-test to test for main effect A and
two one-way ANOVAs (without a constant term) to test main effect B and
interaction AxB (with contrasts: eye(2)).
The question "non-sphericity correction?" I answered with "yes" to model
inhomegeneity of variance (diagonal elements of the variance x covariance
matrix).
The question "correlated repeated measures" I answered with "no" since the
variance x covariance matrix is symmetrical in a 2 by 2 matrix.
My questions:
1. Is the procedure correct?
2. If I intend to compute differential effects on the second level in which
case do I have to use the ANOVA model with constant instead of the ANOVA
model without constant (contrasts at the first level are always differential
effects since each effect will be compared at least with the implicit
baseline)?
Many thanks
Carsten
--
Carsten Giessing
Institut für Medizin
Arbeitsgruppe Kognitive Neurologie
Forschungszentrum Jülich
Tel. +(0) 2461/612483
|