bootstrap_experiment: Experimenting with sampling and calculating confidence...

In michaelquinn32/adventureR: Functions for the Big Blog of R Adventures.

Description

When bootstrapping, we are traditionally instructed to generate a sample equal in size to the entire dataset. We can shortcut this method, as long as we are willing to make certain adjustments when calculating confidence intervals. Two alternative sampling methods are tested here, one with replacement and the other without.

Usage

bootstrap_experiment(funs, nrange = c(1000, 10000), rrange = c(0.25, 0.75),
  mus = c(0, 4.5), sigmas = c(1, 5.5), bsi_range = c(1000, 10000),
  length.out = 5, niter = 5)

compare_methods(param, funs)

bs_sim(x, bsi, replace, funs, n, b, ..., probs = c(0.025, 0.975))

red_se(x, probs = c(0.025, 0.975), n, b, xbar)

ss_quantiles(x, probs = c(0.025, 0.975), n, b, xbar)

quantile_ev(x, probs = c(0.025, 0.975), n, b, xbar, int = c(-500, 500))

se_ci(x, probs = c(0.025, 0.975), ...)

qntl(x, probs = c(0.025, 0.975), ...)

Arguments

funs	a list of functions for calculating confidence intervals
nrange	the length of the vector
rrange	the portion of the vector's length used in each subsample
mus	the mean of the sampling distribution
sigmas	the variance of the sampling distribution
bsi_range	the number of bootstrap iterations
length.out	the number of parameters drawn from the range
niter	the number of repetitions for each parameter combination

Details

The primary functions simulate data for estimating bootstrapped CIs for the mean of a normally-distributed vector. They accept ranges for the simulation parameters. These are described in the parameters. The experiment is generic in that you can pass it multiple different functions for calculating confidence intervals in a list. When these functions have different parameters, it is usually necessary to have ... in all of their arguments.

The simulation is split across one function to generate parameters and iterators, and a second function that executes a single iteration. That function also relies on a generic bootstrap simulator.

Value

The function experiment generates the test, which results in a data frame that contains:

• the confidence interval method

• whether replacement was used in sampling

• the upper and lower limits of the confidence interval

• the combination of sampling parmaters

• the bias of the upper and lower CI limits

• the sum of the square biases

Functions

• compare_methods: One iteration of bootstrap experiment

• bs_sim: A generic bootstrap estimator

• red_se: Adjusting standard errors for smaller samples

• ss_quantiles: Subsamping quantiles (using differences)

• quantile_ev: Subsamping quantiles (with empirical distribution function)

• se_ci: Another confidence interval function

• qntl: A curried version of quantile to catch unnecessary arguments

Examples

## Not run: 
 red_se <- function(x, probs = c(.025, .975), n, b, mu) {
     se <- sd(x) * sqrt(b / n)
     qnorm(probs, mu, se)
 }

 funs <- list(se = red_se)
 test <- bootstrap_experiment(funs, length.out = 2, niter = 2)

## End(Not run)

michaelquinn32/adventureR documentation built on May 22, 2019, 9:52 p.m.