bootPhi: Properties of bootstrap confidence intervals for the kinship...

In thoree/kinBoot: Bootstrapping and kinship estimation

Properties of bootstrap confidence intervals for the kinship coefficient

Description

The purpose is to compare parametric and nonparametric confidence intervals in kinship applications, currently only for the kinship coefficient.

Usage

bootPhi(
  ped,
  ids = NULL,
  N,
  B,
  CItype = "perc",
  conf.level = 0.95,
  plot = F,
  seed = NULL,
  verbose = F
)

Arguments

ped	ped object with allele frequencies.
ids	Id of pair.
N	Integer. No of simulations.
B	Integer. No of bootstraps.
CItype	Logical. See interval()
conf.level	Double
plot	Logical
seed	Integer.
verbose	Logical.

Details

Marker data are simulated N times giving N estimates of (kappa0, kappa1, kappa2). For each simulation, the realised phi is found. Parametric and nonparametric bootstrapping is done giving boot (the averaged kinship coefficient from B bootstrap simulations) and bias = realised- boot is found. There are various ways to calculate the confidence interval. One option is bca as implemented in coxed::bca. This method fails if the input is a vector of constant values. The default is 'perc', the standard percentile interval. The variable cover is 1 if the confidence interval contains phi and 0 otherwise.

Value

A list with the following sif elements:

• phi: The kinship coefficient of the pedigree.

• averaged : A data frame with two lines, one for parametric and one for nonParametric. The values are averaged over the N simulations. The columns realised, boot, bias, lower, upper, and cover are explained in Details.

• simParametric. The entries are as for averaged but for each parametric simulation.

• simNonparametric The entries are as for averaged but for each nonparametric simulation.

• bootParametric The result of the last parametric bootstrap simulation.

• bootParametric The result of the last nonparametric bootstrap simulation.

Examples

library(forrel)
library(ribd)
library(coxed) # for bca confidence intervals

# Example 0
ped = doubleFirstCousins()
ped = setMarkers(ped, locusAttributes =  NorwegianFrequencies)
ids = leaves(ped)
res1 = bootPhi(ped, ids, N = 1, B = 2)

Example 2
# The example considers the kinship coefficients between
# brothers named `B1` and `B2` using the 35 markers in
# forrel::NorwegianFrequencies.

ids = c("B1", "B2")
ped = nuclearPed(2, children = ids)
ped = setMarkers(ped, locusAttributes = NorwegianFrequencies)
N = 10 # no of confidence intervals. Increase
B = 100 # no of  bootstraps. Increase
res1 = bootPhi(ped, ids, N = N, B = B, seed = 17)

# Basic output
res1[1:2]

# Compare parametric and nonparametric estimates
y1 = res1$simParametric$boot
y2 = res1$simNonparametric$boot
boxplot(y1, y2, names = c("parametric", "nonparametric"),
        main = "Bootstrap estimates of kinship coefficient",
        sub = "Red stapled line: theoretical value")
abline(h = res1$phi, col = 'red', lty = 2)

thoree/kinBoot documentation built on Nov. 22, 2022, 6:31 p.m.