ronggui wrote:
i have try hard to find the answer by google,but i can not find any solution.
so i wan to ask:
1,can we test the if canonical relationship is significant after using cancor?
One reference is T. W. Anderson: "An Introduction to Multivariate Statistical Analysis", second edition, pages 497-498.
2,if it can,how?
Following the reference above:
cancor.test <- function(obj, N){
# obj is object returned from cancor
# N is sample size, which is not contained in the cancor object!
p1 <- NROW(obj$xcoef)
p2 <- NROW(obj$ycoef)
p <- p1 + p2
r <- length(obj$cor)
# Calculating Bartlett modification of minus twice log likelihood:
bartlett <- -(N-0.5*(p+3))*sum( log( 1-obj$cor^2))
# which is approximately chi-squared with p1p2 degrees of freedom:
list(bartlett=bartlett, p.value=pchisq(bartlett, df=p1*p2, lower.tail=FALSE))
}
This tests if ALLl the canonical correlations are zero. Anybody knows how good this approximation is,
and how dependent on multivariate normality?
Kjetil
3,if not,is it under-developed or there is not need to do it?or there is no good way to do it?
i hope my question is not too silly.
______________________________________________ R-help@stat.math.ethz.ch mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
--
Kjetil Halvorsen.
Peace is the most effective weapon of mass construction.
-- Mahdi Elmandjra______________________________________________ R-help@stat.math.ethz.ch mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
