Run repeated mediation simulations for a single parameter configuration

Generates n_iter synthetic datasets with mediator-outcome confounding (via mo_confounding), runs all five (or six, when phi > 0) mediation estimators, and summarises bias / RMSE / Type I error for both NDE and NIE.

Usage

run_mediation_sim(
  n_iter = 100,
  n_samples = 500,
  n_features = 20,
  n_mediators = 1,
  beta_X = 0.1,
  alpha_M = 0.5,
  beta_M = 0.3,
  conf_str = 0.8,
  w_signal = 0.7,
  mo_confounding = 0.8,
  phi = 0,
  rho_G1 = 0,
  rho_G2 = 0,
  rho_pop = 0,
  lambda_XM = NULL,
  lambda_MY = NULL,
  omega_1 = NULL,
  omega_2 = NULL,
  feat_cor = 0,
  base_seed = 100,
  n_cores = 1
)

Arguments

n_iter: Number of simulation replicates. Default 100.
n_samples: Observations per replicate. Default 500.
n_features: Number of outcome and negative-control features. Default 20.
n_mediators: Number of independent mediators. When > 1, each mediator has its own genetic instrument Gm and contributes additively to Y. Default 1 (single mediator, backward compatible).
beta_X: Direct effect of X on Y (true NDE). Default 0.10.
alpha_M: Effect of X on mediator. Default 0.50.
beta_M: Effect of mediator on Y (per-mediator true NIE = alpha_M * beta_M). Default 0.30.
conf_str: Confounding strength delta. Default 0.80.
w_signal: Proxy quality omega. Default 0.70.
mo_confounding: Strength of U1 -> M (mediator-outcome confounding). Default 0.80.
phi: Strength of the mediator instrument Gm -> M. 0 = no mediator instrument (five estimators, backward compatible). > 0 = generates Gm and includes the 2-stage MR estimator (IV2SLS2). Default 0.
rho_G1: Correlation of G1 with conf_XM. Default 0.
rho_G2: Correlation of G2 with conf_MY. Default 0.
rho_pop: Shared population structure. Default 0.
lambda_XM: Optional per-path confounder loading vector (X->M path).
lambda_MY: Optional per-path confounder loading vector (M->Y path).
omega_1: Coverage of conf_XM by W1. NULL = use w_signal.
omega_2: Coverage of conf_MY by W2. NULL = use w_signal.
feat_cor: Within-module correlation for block-diagonal co-expression modules in Y and W. 0 = independent features. Default 0.
base_seed: Starting seed; replicate i uses base_seed + i. Default 100.
n_cores: Number of parallel workers. Default 1.

Value

A list with summary (one row per method with NDE/NIE bias, RMSE, Type I), raw (full per-feature results), true_NDE, true_NIE, and params.

Details

When are non-default (rho_G1, rho_G2, rho_pop, lambda_XM, lambda_MY, omega_1, omega_2), the two-stage proximal estimators (PGC2, PGC2Gm) are also included.

Examples

res <- run_mediation_sim(n_iter = 3, n_samples = 100,
                         mo_confounding = 0.8)
res$summary
#>   method   NDE_mean   NDE_bias NDE_pct_bias     NDE_sd  NDE_rmse NDE_mean_se
#> 1  UNADJ -0.4692764 -0.5692764    -5.692764 0.13046612 0.5837922  0.06182445
#> 2 DIRECT -0.4026100 -0.5026100    -5.026100 0.14829814 0.5236818  0.11644642
#> 3   COCA  2.3579413  2.2579413    22.579413 4.42753239 4.9269858  5.55311186
#> 4 IV2SLS -0.1893373 -0.2893373    -2.893373 0.07783502 0.2994552  0.05814444
#> 5    PGC -0.4413775 -0.5413775    -5.413775 0.13261134 0.5571197  0.07335239
#>   NDE_coverage   NIE_mean   NIE_bias NIE_pct_bias    NIE_sd  NIE_rmse
#> 1   0.00000000  1.1309836  0.9809836     6.539891 0.1913857 0.9991730
#> 2   0.01666667  0.8873532  0.7373532     4.915688 0.1684348 0.7560339
#> 3   0.97826087 -2.1607016 -2.3107016   -15.404678 4.4217077 4.9462924
#> 4   0.01666667  0.6710078  0.5210078     3.473386 0.1071212 0.5317263
#> 5   0.00000000  0.7834623  0.6334623     4.223082 0.1400766 0.6485128
#>   NIE_mean_se NIE_coverage  NIE_type1  NDE_type1 n_NDE n_NIE
#> 1  0.07189956    0.0000000 1.00000000 1.00000000    60    60
#> 2  0.11620532    0.0000000 1.00000000 0.86666667    60    60
#> 3  5.53584617    0.9782609 0.02173913 0.06521739    46    46
#> 4  0.06855147    0.0000000 1.00000000 0.81666667    60    60
#> 5  0.07627316    0.0000000 1.00000000 0.98333333    60    60