List

(Code corrections made: 19 March 2007)

DESCRIPTION
Functions for kernel estimation of ROC curves. Bandwidth selection based on method of Hall and Hyndman (2003). Confidence intervals constructed using the method of Hall, Hyndman and Fan (2003).
USAGE
ROC.est(x, y, hx, hy, conf=95, intervals=TRUE, …)
REQUIRED ARGUMENTS
 x numeric vector of observations y numeric vector of observations
OPTIONAL ARGUMENTS
 hx and hy hx=bandwidth for x data and hy=bandwidth for y data. Defaults calculated if missing using method specified by method. conf Probability coverage for confidence bands. intervals Indicates if confidence bands required. (Calculation of bands does slow things down.)
VALUE
 x and y (x,y) points on the estimated ROC curve hx and hy value of bandwidths used in calculation lo and hi vectors containing lower and upper limits of the confidence bands. (Same length as x). s2 Vector of standard errors of the curve. (Same length as x).
DETAILS
Details are given in Hall and Hyndman (2003) and Hall, Hyndman and Fan (2003)
EXAMPLES
m <- n <- 50
x <- rnorm(m,0,1)
y <- rnorm(n,1,1)
hopt <- h.norm2(0,1,1,1,n,n)
ROC1 <- ROC.est(x,y,hx=hopt$hx,hy=hopt$hy)plot(ROC1,type=”l”,xlab=”False positive”,ylab=”True positive”)
lines(ROC1$x,ROC1$lo,lty=2)
lines(ROC1$x,ROC1$hi,lty=2)