Perturbation family: re-randomisation (permutation) test

Permutes treatment labels across subjects, optionally within strata, refits the pre-specified model each time and returns the permutation p-value for the observed contrast. The returned row carries the observed estimate and the permutation p; the null distribution is attached as attr(, "null").

Usage

stress_permute(
  spec,
  n = 1000,
  strata = NULL,
  statistic = c("estimate", "t"),
  seed = 1,
  cores = 1L
)

Arguments

spec

A stress_spec().

n

Number of permutations.

strata

Name of a subject-level column to permute within; NULL permutes freely.

statistic

"estimate" (the contrast; the permutation refits skip the adjusted degrees of freedom, so this is the fast choice) or "t" (estimate over its standard error, refitted with the spec's inference).

seed

Random seed.

cores

Number of worker processes; 1 fits sequentially.

Value

A one-row stress_rows data frame with p equal to (1 + number of |null| >= |observed|) / (n + 1).

Examples

small <- mmrm::fev_data[mmrm::fev_data$USUBJID %in% paste0("PT", 1:60), ]
spec <- stress_spec(small, "FEV1", "USUBJID", "AVISIT", "ARMCD", "PBO", covariates = ~ FEV1_BL)
stress_permute(spec, n = 19)
#>                  family           label        kind        role is_base
#> 1 re-randomisation test 19 permutations alternative sensitivity   FALSE
#>   converged                                                          note
#> 1      TRUE statistic = estimate; 19 of 19 permutations converged; seed 1
#>   estimate   stderr df lcl ucl   p n_subj n_obs aicc param covariance
#> 1 3.597563 3.026508 NA  NA  NA 0.1     59   166   NA    NA         us
#>          method                 vcov covariates alpha
#> 1 Kenward-Roger Kenward-Roger-Linear    FEV1_BL  0.05