Exactness of the permutation p-value over a finite design

For statistic values over every assignment of a finite design, the permutation p-value of an assignment is the share of assignments with a statistic at least as large; the share of assignments with p-value at most alpha never exceeds alpha (Research.P3.Exposure.permutation_valid). Under interference this holds whenever the statistic is imputable under the null on the focal units.

Usage

morie_permutation_valid(stats, alpha = c(0.01, 0.05, 0.1))

Arguments

stats

Statistic values, one per assignment in the design.

alpha

Levels to check.

Value

A list: pvalues (one per assignment) and a data frame levels with alpha, rejection_share and valid.

References

Wager S (2024). Causal Inference: A Statistical Learning Approach, ch. 11.

Examples

morie_permutation_valid(stats = c(3, 1, 4, 1, 5, 9, 2, 6), alpha = c(0.1, 0.25, 0.5))
#> $pvalues
#> [1] 0.625 1.000 0.500 1.000 0.375 0.125 0.750 0.250
#> 
#> $levels
#>   alpha rejection_share valid
#> 1  0.10            0.00  TRUE
#> 2  0.25            0.25  TRUE
#> 3  0.50            0.50  TRUE
#> 
#> attr(,"theorem")
#> [1] "Research.P3.Exposure.permutation_valid"