Simulation studies

library(logcumulant)

The package includes utilities to reproduce the size and power studies. They are Monte Carlo intensive; the examples below use small replication counts for speed.

Empirical size

Under a true null, a well-calibrated test rejects at approximately the nominal level. The asymptotic reference over-rejects, while the bootstrap restores size.

size_study(sample_sizes = c(30, 50, 100, 200), Nsim = 1000)

Empirical power

Power against a set of alternatives, with optional size-correction for a fair comparison:

power_study(n = 100, Nsim = 1000)

Reproducing the figures

The complete set of diagrams in the accompanying paper can be regenerated with the bundled script:

# from the package source directory
source("reproduce_all_figures.R")