FDRCurve: FDR Curve

In rvalues: R-Values for Ranking in High-Dimensional Settings

Description

Estimates the expected proportion of misclassified units when using a given r-value threshold. If plot=TRUE, the curve is plotted before the estimated function is returned.

Usage

FDRCurve(object, q, threshold = 1, plot = TRUE, xlim, ylim, xlab, ylab, main, ...)

Arguments

object	An object of class "rvals"
q	A value in between 0 and 1; the desired level of FDR control.
threshold	The r-value threshold.
plot	logical; if TRUE, the estimated FDR curve is plotted.
xlim,ylim	x and y - axis limits for the plot
xlab,ylab	x and y - axis labels
main	the title of the plot
...	additional arguments to plot.default

Details

Consider parameters of interest (θ_1,...,θ_n) with an effect of size of interest τ. That is, a unit is taken to be "null" if θ_i ≤ τ and taken to be "non-null" if θ_i > τ.

For r-values r_1,...,r_n and a procedure which "rejects" units satisfying r_i ≤ c, the FDR is defined to be

FDR(c) = P(θ_i < τ,r_i ≤ c)/P(r_i ≤ c).

FDRCurve estimates FDR(c) for values of c across (0,1) and plots (if plot=TRUE) the resulting curve.

Value

A list with the following two components

fdrcurve	A function which returns the estimated FDR for each r-value threshold.
Rval.cut	The largest r-value cutoff which still gives an estimated FDR less than q.

Author(s)

Nicholas Henderson and Michael Newton

See Also

OverlapCurve

Examples

n <- 500
theta <- rnorm(n)
ses <- sqrt(rgamma(n,shape=1,scale=1))
XX <- theta + ses*rnorm(n)
dd <- cbind(XX,ses)

rvs <- rvalues(dd, family = gaussian)

FDRCurve(rvs, q = .1, threshold = .3, cex.main = 1.5)

Example output

rvalues documentation built on March 11, 2021, 9:05 a.m.