Conclusion-flip rules for the fragility index

What counts as "the conclusion changed" is itself a research decision. sign_flip() – the default – flags any estimate whose sign differs from the reference's, which suits signed effects with a null at zero. threshold_flip() flags estimates crossing a substantive threshold from the reference's side, which suits one-sided claims and bounded quantities, where a sign flip may be impossible or meaningless. Any function(estimate, reference) -> logical works in audit_fragility()'s flip argument.

Usage

sign_flip()

threshold_flip(at)

Arguments

at

For threshold_flip(): the substantive threshold.

Value

A function (estimate, reference) -> logical.

Details

Edge worth knowing: a reference estimate of exactly zero has no sign, so sign_flip() flags nothing; choose a threshold rule there.

Examples

rule <- sign_flip()
rule(c(-0.2, 0.1, NA), reference = 0.3)
rule2 <- threshold_flip(at = 0.5)
rule2(c(0.4, 0.6), reference = 0.7)