Paired management contrasts with within-chain block resampling

Differences are alternative minus benchmark within each HCR and scenario. Whole matched trajectories are resampled, never annual rows. Blocks wrap within chains, never across chain boundaries. No scenario probability weights or scientific equivalence margins are assumed. Missing paired metrics are reported and excluded consistently from the point estimate and resampling.

Usage

compare_mse(
  summaries,
  benchmark,
  draw_map,
  metrics = c("mean_tacc", "aav", "terminal_soft", "terminal_hard"),
  block_length = 10L,
  n_boot = 1000L,
  seed = 271828L
)

Arguments

summaries

Output from summarise_mse().

benchmark

The EM_label to compare with, present once per HCR/scenario.

draw_map

Data frame with unique Iteration, chain, and draw (within-chain order) for every evaluation iteration.

metrics

Numeric trajectory metrics to compare.

block_length

Requested circular block length within each chain.

n_boot

Number of bootstrap replicates.

seed

Reproducible seed; the caller's RNG is restored.

Value

Data frame of paired differences and percentile intervals.