Skip to contents

Direction stability of a result that misses significance

Source: R/direction.R
stress_direction.Rd

A small trial can carry a real effect and still miss significance for want of power. For such a result the stress test reports how stable what was seen is: how many pre-specified re-analyses reverse the direction of the estimate, and which; how many move it outside the filed confidence interval; and how many participants the greedy deletion search removes before the direction reverses.

Usage

stress_direction(
  spec,
  census = NULL,
  max_k = 10L,
  method = c("ranked", "greedy"),
  cores = 1L
)

Arguments

spec

A stress_spec().

census

Optional stress_census over the pre-specified re-analyses. If given, its counted rows supply the reversal and interval counts; without it only the deletion count is computed.

max_k

Most participants to remove in the deletion search (stress_fragility() with target = "direction"). The default is 10.

method

"ranked" (default) or "greedy", as in stress_fragility().

cores

Passed to stress_fragility().

Value

A list of class stress_direction: base (the pre-specified fit), z, n, significant, alpha; from the census n_counted, n_reversed, n_outside_ci and reversed_rows (NA and NULL without one); from the search k_rev (participants removed to reverse, NA if not reached within max_k), k_rev_reached, subjects, share_of_n and search, the rows of the search.

See also

stress_verdict() for the summary label, stress_fragility() for the deletion searches.

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,
                    covariance = "cs", method = "Satterthwaite", vcov = "Asymptotic")
cen <- stress_census(spec, stress_covariance(spec, c("cs", "ar1h")), stress_covariates(spec))
stress_direction(spec, cen, max_k = 6)
#> Direction stability
#> Base: estimate 3.48, 95% CI -0.796 to 7.75, p = 0.11 (z = 1.61); not significant at alpha 0.05
#> Pre-specified re-analyses: 0 of 2 reverse the direction; 0 move the estimate outside the filed interval
#> Participants removed to reverse the direction (ranked search): 5 of 59 (8%): PT3, PT19, PT43, PT5, PT26