Rosenbaum sensitivity bound for matched pairs — morie_causal_rosenbaum

Sweeps the odds-ratio bound Gamma and reports the largest value at which the Wilcoxon signed-rank result remains significant.

Usage

morie_causal_rosenbaum_bound(
  paired_diff,
  gamma_max = 3,
  n_gamma = 25,
  alpha = 0.05
)

Arguments

paired_diff: within-pair outcome differences; zeros dropped.
gamma_max: largest Gamma to try.
n_gamma: grid size.
alpha: significance level.

Value

list with gamma_critical, gamma_grid, p_upper, significant_at_gamma_1, interpretation.

Details

What this does NOT do is test whether hidden bias exists – nothing can, from observational data. It converts a result into a statement of the form "an unmeasured confounder would have to change treatment odds by a factor of Gamma* within matched pairs to overturn this", and that statement is only informative against the confounders that are plausible in the application.

References

Rosenbaum, P. R. (2002). Observational Studies, 2nd ed., Ch. 4. Springer.

Examples

set.seed(1)
morie_causal_rosenbaum_bound(rnorm(60, 0.8))$gamma_critical
#> [1] 3