Skip to contents

Every registered test includes parity and method metadata from ggpower_tests(). Tests marked approximation use large-sample or proxy formulas. Result objects include notes explaining the kernel used.

tests <- ggpower_tests()
approx <- tests[tests$parity == "approximation", c("id", "method", "module")]
knitr::kable(approx, row.names = FALSE)
id method module
exact_correlation Fisher Z with exact-kernel TODO workspace
z_tetrachoric large-sample tetrachoric z approximation workspace

Workspace approximations

Test Limitation
exact_correlation Fisher-Z; exact small-sample kernel planned
z_tetrachoric Large-sample Fisher-Z on tetrachoric ρ\rho
exact_mcnemar Discordant-pair binomial proxy
z_logistic, z_poisson Wald normal approximation
wilcoxon_signed, wilcoxon_mann_whitney ARE-scaled t tests

Biomarker approximations

Test Limitation
roc_auc_one, roc_auc_two Hanley-McNeil / DeLong-style normal
diagnostic_acc Binomial normal; power = min(sens, spec)
survival_logrank, cox_regression Equal follow-up / events simplification
discovery_fdr Independent tests + simplified BH

Clinical approximations

Test Limitation
rct_noninferiority_binary, rct_equivalence_proportion Normal proportion approximation
simon_two_stage Fixed design; no optimal design search

When exact enumeration is used

exact_fisher and exact_binomial use enumeration when the outcome grid is small enough; large tables fall back to normal approximations.