Exact randomization inference with uncertain record identities

Does an experimental conclusion survive every linkage allowed by the candidate records and an identity-error budget?

For completely randomized experiments whose assignment and outcome files cover the same units once, this code computes the minimum and maximum Fisher p-values over feasible one-to-one linkages, returns the linkages attaining them, and finds the smallest number of changed identities that removes rejection. Every feasible linkage shares one randomization distribution, so the largest p-value comes from the smallest attainable absolute treatment contrast, which an integer program finds.

The test is valid when the candidate set and error budget contain the true linkage. The maximum p-value is the robust test; picking the minimum after seeing outcomes is not a valid test. Method, proofs and simulations are in the paper.

Example

source("R/core.R")
source("R/graph.R")
source("R/information.R")

z <- c(1, 1, 1, 1, 0, 0, 0, 0)
y <- z
accepted <- c(2L, 1L, 3:8)
groups <- c(1, 1, 2, 3, 2, 4, 3, 4)
graph <- outer(groups, groups, "==")

fit <- graph_information_test(y, z, graph, accepted, trusted_error_budget = 2)
fit[c("lower", "upper", "p_min", "p_max")]
fit$least_favorable_map

graph_breakdown_budget(y, z, graph, accepted, alpha = .05)

graph_breakdown_budget() returns the first identity budget at which the robust test fails to reject, with a linkage that attains it.

Reproduce

Requires R, the packages in renv.lock (renv::restore()), and LaTeX.

make methods   # tests, lint, both simulations, paper
make report    # rebuild the paper from committed results

• R/: p-value extrema and witnesses (information.R), linkage bounds (graph.R), the binary accuracy-only comparator (accuracy.R).
• scripts/: the controlled design, the identifier simulation, and table rendering.
• results/: outputs the paper reads.
• tests/: checks against exhaustive enumeration of linkages and assignments.