gof_power_adaptive: Adaptive power estimation for goodness-of-fit tests

In Rgof: 1d Goodness of Fit Tests

Adaptive power estimation for goodness-of-fit tests

Description

Estimates power in simulation batches and stops when the Monte Carlo standard error of every reported power estimate is no larger than target.mcse, or when B.max is reached.

Usage

gof_power_adaptive(
  pnull,
  vals = NA,
  rnull,
  ralt,
  param_alt,
  w = function(x) -99,
  phat = function(x) -99,
  TS,
  TSextra,
  With.p.value = FALSE,
  alpha = 0.05,
  Range = c(-Inf, Inf),
  nbins = c(50, 10),
  rate = 0,
  maxProcessor,
  minexpcount = 5,
  ChiUsePhat = TRUE,
  SuppressMessages = FALSE,
  target.mcse = 0.01,
  B.min = 500,
  B.max = 10000,
  B.batch = 250,
  B.null = 1000,
  conf.level = 0.95,
  list.with.everything
)

Arguments

pnull	Function to calculate the cdf under the null hypothesis.
vals	=NA values of a discrete random variable, or NA for continuous data.
rnull	Function to generate data under the null hypothesis.
ralt	Function to generate data under the alternative hypothesis.
param_alt	Vector of parameter values under the alternative.
w	Optional weight function; returns -99 if no weights are used.
phat	Function to estimate parameters, or function(x) -99.
TS	Optional user-supplied test statistic or p-value routine.
TSextra	Optional list supplied to TS.
With.p.value	=FALSE; TRUE if a user-supplied TS returns p-values.
alpha	Significance level.
Range	Limits of possible continuous observations.
nbins	Number of bins for chi-square tests.
rate	Poisson rate if sample size is random; 0 for fixed sample size.
maxProcessor	Maximum number of processors used for the statistic-based simulation. P-value and chi-square simulation remains sequential, as in the corresponding existing package routines.
minexpcount	Minimum expected bin count for chi-square tests.
ChiUsePhat	If TRUE, use estimated parameters in chi-square tests.
SuppressMessages	Suppress adaptive progress messages when TRUE. A run-time estimate is still displayed when the estimated maximum run time exceeds 30 seconds.
target.mcse	Target Monte Carlo standard error.
B.min	Minimum number of alternative simulations per parameter.
B.max	Maximum number of alternative simulations per parameter.
B.batch	Number of additional alternative simulations per batch.
B.null	Number of null simulations used once to estimate the fixed critical values for statistic-based tests.
conf.level	Confidence level for Wilson confidence intervals.
list.with.everything	Optional case-study list accepted by the package.

Details

Unlike gof_power(), this routine performs the adaptive simulation directly. For statistic-based tests it estimates the null critical values once using B.null null simulations and holds those critical values fixed while alternative simulations are added in batches.

Statistic-based tests use a two-stage procedure. First, B.null null data sets are simulated and one set of critical values is estimated. Those critical values are then held fixed. Alternative data sets are simulated in batches and exact rejection counts are accumulated.

User tests with With.p.value=TRUE do not require null critical-value simulation; rejection is determined directly from p-value < alpha.

The package's chi-square tests already perform their own bin construction and diagnostics. Their rejection counts are accumulated directly from chi_test_cont() or chi_test_disc(). For discrete data the bin definitions are selected once for each alternative parameter, matching the design of chi_power_disc().

Value

An object of class Rgof_power_adaptive and Rgof_power. It contains power estimates, Monte Carlo standard errors, Wilson confidence intervals, total alternative simulation count, null simulation count, convergence information, and (when applicable) the fixed critical values.

Rgof documentation built on Sept. 13, 2026, 5:06 p.m.