In addition to the code below see also the padjust function in R of form
p.adjust(p, method = p.adjust.methods, n = length(p))
for example
p <- c(0.01,0.04) p.adjust(p,method="fdr",n=2)
which adjusts a set of p-values for a variety of methods including Bonferroni, Hochberg and Holm.
[ JUST CUT AND PASTE AT THE R PROMPT AFTER ADJUSTING INPUTS AS REQUIRED]
Sidak and Holm
This is a stepwise approach which produces the two cutoffs using the Sidak and Holm tests respectively. Only p-values equal or below the cutoff value are statistically significant at alpha=alp when adjusted for the number of p-values entered.
Input p-values in p and Type I error as alp.
p <- c(0.1,0.2,0.002) alp <- 0.05 pos <- rank(p) ps <- sort(p) ncomp <- length(ps) dsidak <- 1 - ((1 - ps)**(ncomp-pos+1)) j.alpha <- rep(alp,length(p)) diff <- dsidak - j.alpha neg.diff <- diff[diff < 0] pos.diff <- neg.diff[length(neg.diff)] index <- diff == pos.diff p.cutoff <- ps[index] print(p.cutoff) p.sig <- p[p <= p.cutoff] holm <- (ncomp - pos + 1)*ps j.alpha <- rep(alp,length(p)) diff <- holm - j.alpha neg.diff <- diff[diff < 0] pos.diff <- neg.diff[length(neg.diff)] index <- diff == pos.diff p.cutoff <- ps[index] print(p.cutoff) p.sig <- p[p <= p.cutoff]
Ryan and Einot-Gabriel
Input p-values in p, Type I error as alp and the two differenced groups in g1 and g2. The adjusted p-values are outputted using the ryan and einot-gabriel methods respectively.
p <- c(0.1,0.2,0.002)
alp <- 0.05
g1 <- c(1,1,2)
g2 <- c(2,3,3)
ncomp <- length(p)
step <- abs(g2-g1)+1
ryan <- p*(ncomp/step)
eingab <- 1 - ( (1 - p)**(ncomp/step) )
unit <- rep(1,length(p))
for (i in 1:length(p)) {
if (ryan[i] > unit[i]) ryan[i] = unit[i]
if (eingab[i] > unit[i]) eingab[i] = unit[i]
}
print(ryan)
print(eingab)