[R] Huynh-Feldt correction

[Wolfgang Pauli]

Dear All,

I tried to help myself by searching in newsgroups etc. I found a test for
epsilon in a newsgroup (see attachement). I wrote a little skript to update
the summary.anova if epsilon < 0.75.
Although I doublechecked my statistic book for the formula, the result is
always the half of the epsilon calculated by SPSS. It is really strange. If I
multiply espilon by two, I get the right results...

regards,

WP

#------------------------------------begin -----------------------------

correctSummary <- function(sum,eps) {
	# eps is the epsilon-value returned by function "hf"
	# sum is the summary.aov object in which the Huynh-Fldt correction has to be
down
	len <- length(sum)
	newSum <- sum
	for (i in 2: len) {
	newSum[[i]][[1]][4,] <- newSum[[i]][[1]][2,]
	newSum[[i]][[1]][3,] <- newSum[[i]][[1]][2,]
	newSum[[i]][[1]][2,] <- newSum[[i]][[1]][1,]
	rownames(newSum[[i]][[1]]) <- c(rownames(newSum[[i]][[1]][1,]),
"Huynh-Feldt", rownames(newSum[[i]][[1]][2,]), "Huynh-Feldt")
	newSum[[i]][[1]][2,][1] <- newSum[[i]][[1]][2,][1]*epsilon*2
	newSum[[i]][[1]][4,][1] <- newSum[[i]][[1]][4,][1]*epsilon*2
	newSum[[i]][[1]][2,][3] <- newSum[[i]][[1]][2,][2]/newSum[[i]][[1]][2,][1]
	newSum[[i]][[1]][4,][3] <- newSum[[i]][[1]][4,][2]/newSum[[i]][[1]][4,][1]
	newSum[[i]][[1]][2,][4] <- newSum[[i]][[1]][2,][3]/newSum[[i]][[1]][4,][3]

	newSum[[i]][[1]][2,][5] <- pf(as.numeric(newSum[[i]][[1]][2,][4]),
as.numeric(newSum[[i]][[1]][2,][1]), as.numeric(newSum[[i]][[1]][4,][1]),
lower.tail = FALSE)
	}
	newSum

}

hf <- function(m){
        # m is a matrix with subjects as rows and conditions as columns
        # note that checking for worst case scenarios F correction first
 might # be a good idea using J/(J-1) as the df correction factor
        n<- length(m[,1])
        J<-length(m[1,])
        X<-cov(m)
        r<- length(X[,1])
        D<-0
        for (i in 1: r) D<- D+ X[i,i]
        D<-D/r
        SPm<- mean(X)
        SPm2<- sum(X^2)
        SSrm<-0
        for (i in 1: r) SSrm<- SSrm + mean(X[i,])^2
        epsilon <- (1/(J-1))*((J^2*(D-SPm)^2)/(SPm2 - 2*J*SSrm + J^2*SPm^2))
	epsilon
}


# ----------------------------------------------------------------------------

On Thursday December 18 2003 12:00, Wolfgang Pauli wrote:
> Dear R-helpers,
>
> Does anybody know, whether there is an option to tell aov/anova, or do
> something similar, to get a Huynh-Feldt correction of dfs in the aov/anova
> function?
>
> Thanks in advance!
>
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://www.stat.math.ethz.ch/mailman/listinfo/r-help