CovrgPlot: Coverage probabilities of the confidence intervals,...

In CooccurrenceAffinity: Affinity in Co-Occurrence Data

Coverage probabilities of the confidence intervals, calculated and plotted

Description

This function calculates and plots the coverage probabilities of the four confidence intervals CI.CP, CI.Blaker, CI.midQ and CI.midP produced by functions AlphInts() and ML.Alpha(). The coverage plots are useful in showing how the conservative confidence intervals (CI.CP and CI.Blaker) tend to have above-nominal coverage probabilities versus the intervals CI.midQ and CI.midP that have coverage much closer to the nominal confidence level lev.

Usage

CovrgPlot(
  marg,
  scal = log(2 * marg[3]^2),
  lev = 0.95,
  alim = 4,
  intpts = 1,
  plotCI = 1:4
)

Arguments

marg	a 3-entry integer vector containing (mA,mB,N)
scal	an integer parameter (default 2*N^2, capped at 10 within the function) that should be 2 or greater
lev	a confidence level, generally somewhere from 0.8 to 0.95 (default 0.95)
alim	the absolute value of alpha at which the plotted figure is cutoff; so the x-axis limits are (-alim, alim)
intpts	an integer controlling how many points are used to interpolate between values alpha for which the Extended Hypergeometric distribution function is equal exactly to (1+lev)/2 or (1-lev)/2 at some integer co-occurrence value x
plotCI	a vector of up to 4 numbers out of the set 1,2,3,4, corresponding to which Confidence Intervals (1=CP, 2=Blaker, 3=midQ, 4=midP) should have their coverage probability values plotted simultaneously (with different colors) in the figure

Details

This function calculates and plots the coverage probabilities of the confidence intervals CI.CP, CI.Blaker, CI.midQ and CI.midQ produced by function AlphInts(), for fixed marginals "marg" = (mA,mB,N) and confidence level "lev" as a function of alpha. The plots serve as a useful exhibit to show the range of always conservative (and sometimes quite conservative) coverage probabilities for CI1, and the much closer-to-nominal coverage probabilities for the recommended Confidence Interval CI2. The outputted vector "avec" of alpha values is the set of all discontinuities in the slope of the coverage-probability function together with "intpts: equally spaced values between them. The plotted coverage probabilities are the corresponding exact (not simulated) coverage probabilities of each of the selected (up to 4) confidence intervals at the alphvec points in the four-column array "covPrb".

Value

This function plots a figure of overlaid coverage probabilities with differently colored lines, explained in the figure legend. In addition, it returns "avec", a vector of alpha values at which coverage probabilities are calculated, an xn x 4 x 2 array "arrCI" where xn is the length of "avec" and for each index i in 1:xn the array arrCI[i,,] is the 4x2 matrix of 4 Confidence Interval lower and upper alpha-value endpoints; and "covPrb", the xn x 4 matrix of coverage probabilities for the four CI's at each of the alpha values avec[i] for i =1,...,xn.

variable

Column name of the dataframe produced by reshape::melt

value

Column name of the dataframe produced by reshape::melt

abovethr

Column name of the dataframe produced by reshape::melt

belowthr

Column name of the dataframe produced by reshape::melt

aboveperc

Column name of the dataframe produced by plyr::ddply

belowperc

Column name of the dataframe produced by plyr::ddply

legendtext

Column name of the dataframe

Author(s)

Eric Slud and Kumar Mainali

References

to be added

Examples

# True coverage probability by the 95% Confidence Interval
CovrgPlot(marg = c(50,70,150), lev = 0.95)

# the remainder of the script has been enclosed under \donttest{}
# to bypass the CRAN's 5 second limit on example files
# --------------------------------------------------------------



# If confidence level is not specified, it shows results for 95% CI
CovrgPlot(marg = c(50,70,150))

# True coverage probability by the 90% Confidence Interval
CovrgPlot(marg = c(50,70,150), lev = 0.90)

# Select some of the confidence intervals to plot
CovrgPlot(marg = c(50,70,150), lev = 0.95, plotCI=1:3)
CovrgPlot(marg = c(50,70,150), lev = 0.95, plotCI=c(1,3,4))
CovrgPlot(marg = c(50,70,150), lev = 0.95, plotCI=1)

 #end of \donttest{}

CooccurrenceAffinity documentation built on May 4, 2023, 1:07 a.m.