R/cgci.R

In musto101/wilcox_R: What the package does (short line)

cgci <-
function(x, alpha = 0.05, ks = 3.5)
{
#gets T_S,n with a coarse grid
# and the corresponding robust  100 (1-alpha)% CI
	n <- length(x)
	up <- 1 - alpha/2
	med <- median(x)
	madd <- mad(x, constant = 1)
	d <- sort(x)	##get robust T_S,n CI
	lo <- sum(x < (med - ks * madd))
	hi <- sum(x > (med + ks * madd))
	tp <- max(hi, lo)/n
	if(tp == 0)
		tp <- 0
	if(tp > 0 && tp <= 0.01)
		tp <- 0.01
	if(tp > 0.01 && tp <= 0.1)
		tp <- 0.1
	if(tp > 0.1 && tp <= 0.25)
		tp <- 0.25
	if(tp > 0.25 && tp <= 0.4)
		tp <- 0.4
	if(tp > 0.4)
		tp <- 0.49
	tstmn <- mean(x, trim = tp)	
	#have obtained the two stage trimmed mean
	ln <- floor(n * tp)
	un <- n - ln
	if(ln > 0) {
		d[1:ln] <- d[(ln + 1)]
		d[(un + 1):n] <- d[un]
	}
	den <- ((un - ln)/n)^2
	swv <- var(d)/den	
	#got the scaled Winsorized variance
	rdf <- un - ln - 1
	rval <- qt(up, rdf) * sqrt(swv/n)
	tslo <- tstmn - rval
	tshi <- tstmn + rval	
	##got low and high endpoints of robust T_S,n CI
	list(int = c(tslo, tshi), tp = tp)
}