Run every within-condition reliability metric

Convenience orchestrator that runs the reliability metrics proper (split-half with Spearman-Brown correction and ICC) on a single condition's signal matrix and wraps the two results in an rcisignal_rel_report for joint printing and plotting.

Use this when you want the full within-condition reliability report in one call. Pass each metric individually if you need to tune arguments per metric.

Usage

run_reliability(
  signal_matrix,
  n_permutations = 2000L,
  null = c("none", "permutation", "random_responders"),
  noise_matrix = NULL,
  icc_variants = c("3_1", "3_k"),
  mask = NULL,
  seed = NULL,
  progress = TRUE,
  acknowledge_scaling = FALSE
)

Arguments

signal_matrix: Pixels x participants, base-subtracted.
n_permutations: Passed to rel_split_half(). Default 2000.
null: Forwarded to rel_split_half(). Default "none".
noise_matrix: Required when null = "random_responders"; forwarded to rel_split_half().
icc_variants: Passed to rel_icc().
mask: Optional logical vector of length nrow(signal_matrix) (column-major) threaded through to both metrics. Build with make_face_mask() (parametric oval and sub-regions) or read_face_mask() (PNG/JPEG mask).
seed: Optional integer; used for the split-half permutations.
progress: Show cli progress bars.
acknowledge_scaling: Logical. Forwarded to rel_icc().

Value

Object of class rcisignal_rel_report with $results = named list of two result objects (split_half, icc) and $method = "reliability".

What is included (and what is not)

Returns the two metrics that quantify the reliability of the group-level classification image proper: split-half (a permutation-based estimate of group-CI stability with Spearman-Brown projection to the full sample) and ICC(3,*) (the psychometric variance decomposition). These are non-redundant.

Leave-one-out influence screening lives in rel_loo() and is not bundled here. Its output is an influence diagnostic, not a reliability statistic, and mixing it into the reliability report invites mis-reading r_loo values (which are near 1 by construction) as reliability.

Reading the result

$results$split_half, $results$icc, one result object each, with the same fields as the standalone functions. $method = "reliability".

Reliability metrics expect raw masks

Both downstream metrics expect the raw mask. Inputs with attr(., "source") == "rendered" (set automatically by Mode 1 readers like extract_signal()) error in rel_icc() unless acknowledge_scaling = TRUE.

Examples

if (FALSE) { # \dontrun{
# End-to-end demo: simulate -> compute CIs -> assess reliability.
sim <- simulate_briefrc_data(n_per_condition = 10, n_trials = 60, seed = 1)
cis <- ci_from_responses_briefrc(sim$data, noise_matrix = sim$noise_matrix)
rel <- run_reliability(cis$signal_matrix,
                       n_permutations = 200L, seed = 1)
print(rel)
} # }