Hi,
sorry to bother you again, but I can't figure it out myself and I also
can't find any in-depth documentation about it...
Consider the following SAS code (A1II2... contain the measurements for
40 subjects):
proc glm;
model
A1II2
A1IN2
A1NI2
A1NN2
= /nouni;
repeated CONTEXT 2, TARGET_SATZ 2;
title "A1 500-900 ms";
This produces not only the univariate ANOVAs, but also a number of
multivariate MANOVAs, including some test statistics (Wilks' Lambda,
Pillai's Trace, etc.) on various hypothesis (no CONTEXT effect, no
TARGET_SATZ effect, no CONTEXT*TARGET_SATZ effect).
Unfortunately, I can't seem to be able to figure out how to reproduce
this in R. There's obviously little sense in treating A1II2..A1NN2 as
separate responses, so I came up with the following (for testing a
single effect):
resp <- cbind(
c(A1II2,A1IN2),
c(A1NI2,A1NN2))
cond <- gl(2,40,80)
subj <- gl(40,1,80)
summary(manova( resp ~ cond + Error(subj/cond) ))
Error: subj
Df Pillai approx F num Df den Df Pr(>F)
Residuals 39
Error: subj:cond
Df Pillai approx F num Df den Df Pr(>F)
cond 1 0.07911 1.63232 2 38 0.2089
Residuals 39
Warning message:
Error model is singular in: aov(resp ~ cond + Error(subj/cond))
Now, this looks halfway reasonable, but it's not quite there. My
questions are:
- why is the Error model singular?
- am I even on the right path, or is this completely wrong?
- what do I do for interaction effects?
I'd be very happy if you could give me any hints of where I should be
going with this...
Thanks again
Bela
