drmeta_bootstrap_gamma: Parametric-bootstrap test of the constrained scale gradient

In drmeta: Design-Indexed Location-Scale Meta-Analysis

Parametric-bootstrap test of the constrained scale gradient

Description

Tests H0: gamma = 0 against H1: gamma > 0. The null lies on the boundary of the constrained parameter space, so the likelihood-ratio statistic does not have an asymptotic chi-square distribution and a parametric bootstrap is used instead. Data are simulated from the fitted null model, in which between-study heterogeneity is constant in the design-robustness index.

Usage

drmeta_bootstrap_gamma(
  object,
  B = 999,
  seed = NULL,
  parallel = FALSE,
  ncpus = 2L,
  tol = 1e-08
)

Arguments

object	A constrained drmeta fit.
B	Number of bootstrap samples.
seed	Optional random seed.
parallel	Logical; use mclapply on non-Windows platforms.
ncpus	Number of cores when parallel is TRUE.
tol	Tolerance below which the observed likelihood-ratio statistic is treated as exactly zero.

Details

Replicates in which either refit fails to converge are discarded rather than contributing a statistic computed at a non-optimal point. The p-value uses the number of usable replicates as its denominator, and the discarded count is returned so the loss is visible.

When the constrained estimate is already at the boundary, the null and alternative fits coincide and the observed statistic is zero by construction. Bootstrapping in that situation would compare zero against a simulated distribution with a large point mass at zero, and would return a p-value that looks like evidence but only records the proportion of replicates that also reached the boundary. The function therefore short-circuits, returns p.value = 1 with boundary = TRUE, and runs no replicates. A boundary estimate means the data provide no support for a positive scale gradient; it is not a measured degree of evidence. Fit with constrained = FALSE to see whether the unrestricted gradient is negative, which would contradict the substantive constraint rather than merely fail to support it.

When parallel = TRUE, reproducibility across cores requires the L'Ecuyer-CMRG generator, which this function sets and restores when a seed is supplied.

Value

An object of class drmeta_bootstrap_gamma, a list with components statistic (observed likelihood-ratio statistic), p.value, B (requested replicates), B_used (replicates in which both refits converged), n_failed (replicates discarded), simulated (the simulated statistics, with NA for failed replicates), null (the fitted null model), alternative (the supplied fit), and boundary, a logical flag that is TRUE when the constrained estimate already lies at gamma = 0 and no bootstrap was run.

Examples

path <- system.file("extdata", "bcg_design_robustness.csv", package = "drmeta")
bcg <- utils::read.csv(path)
fit <- drmeta(yi = bcg[["yi"]], vi = bcg[["vi"]], dr = bcg[["dr"]])
drmeta_bootstrap_gamma(fit, B = 100, seed = 1)

drmeta documentation built on Aug. 25, 2026, 1:08 a.m.