David Airey <[email protected]> has a large repeated
measures ANOVA that he is designing, and he is hitting the 11,000
matrix size limit in Stata/SE (The limit is 800 in Intercooled
Stata). He mentions that he could assume higher order
interactions are not significant and drop them from the model to
fall under the limit. He then asks if anyone has other design
suggestions.
Here is part of a message I have just sent to David:
Another possibility you might want to explore is to abandon
repeated measures ANOVA and move over to using MANOVA. You can
see a discussion of this in "[R] manova" in the section titled
"Multivariate analysis of variance with repeated measures" on
pages 395--407. In this case what you are doing is modeling the
repeated measures as the dependent variables in the MANOVA. The
-manovatest- command after -manova- provides the mechanism for
getting at various tests of terms of interest involving linear
combinations of the various dependent variables and also of the
remaining terms in the model.
I have not analyzed your setup to see if MANOVA will work well
for you. What can happen is that depending on your setup, you
might not have enough observations for the MANOVA to be
estimated. That is one of the possible disadvantages to MANOVA
versus repeated measures ANOVA. On the flip side, there are some
important advantages to MANOVA over repeated measures ANOVA
1. Since all the repeated variables are represented as
dependent variables, you are not forming so many large
interactions and your resulting X'X matrix is smaller.
(You are more likely to be able to fit within the matrix
size limitiation.)
2. The -manova- approach directly models the repeated
measures aspect of the design (it allows correlations
between the dependent variables) instead of trying to
correct for it as is done in repeated measures ANOVA (the
various epsilon correction factors that are applied in
rep. measures ANOVA).
3. Some argue that you should always use MANOVA over
repeated measures ANOVA (because of #2) unless you do not
have enough observations (when the data are wide -- see
discussion below).
The data structure used in MANOVA versus ANOVA is wide versus
long (think -reshape- command). So you would use -reshape-
(maybe multiple times since there are several repeated measures
variables in your design) to get to a wide dataset.
Ken Higbee [email protected]
StataCorp 1-800-STATAPC
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
