Skip to contents

Freedman-Lane permutations for DKGE (scaffold)

Source: R/dkge-inference.R
dkge_freedman_lane.Rd

This function orchestrates Freedman-Lane permutations at the *time-series* level: for each subject, fit the reduced model (without the effect of interest), permute residuals, reconstruct surrogate data, refit the full GLM to get B* betas, then re-run DKGE LOSO to obtain a group statistic (e.g., max-|t| over medoid clusters). It requires the caller to provide three adapter functions (or rely on 'fmrireg'/'neuroim2'): - fit_glm(Y_s, X_s, X0_s) -> list(beta = qxP, beta0 = q0xP, resid = TxP) - resample_resid(resid_s) -> resid_s* (TxP) [permute or phase-randomize per run] - transport_and_stat(B_list, X_list, K, c) -> scalar (e.g., max-|t|)

Usage

dkge_freedman_lane(
  Y_list,
  X_list,
  X0_list,
  K,
  c,
  B = 500,
  adapters,
  seed = 123L
)

Arguments

Y_list

list of neuroim2 BrainVectors (or TxP matrices) per subject

X_list

list of Txq design matrices (full)

X0_list

list of Txq0 reduced designs (null space of the contrast)

K

design kernel (qxq)

c

contrast vector (qx1)

B

number of permutations

adapters

list with functions: fit_glm, resample_resid, transport_and_stat

seed

RNG seed for reproducibility

Value

list with fields: stat_obs, stat_null (B-vector), p, details