Skip to contents

Benchmark estimators across confounding scenarios on synthetic data

Source: R/sensitivity.R
gan_sensitivity.Rd

For each scenario in the grid, generates n_iter synthetic datasets with run_single_iteration() (using the trained generator for texture), runs every estimator, and summarises bias / RMSE / power. This is the multi-confounder, negative-control-aware generalisation of the package's parameter sweeps.

Usage

gan_sensitivity(
  trained_gan = NULL,
  conf_grid = c(0.2, 0.5, 0.8),
  coverage_grid = c(0.3, 0.7, 1),
  k_grid = 1,
  nc_model = "proxy",
  n_iter = 50,
  n_samples = 500,
  n_features = 20,
  beta_X = 0.1,
  alpha_M = 0.5,
  beta_M = 0.3,
  effect_size = NULL,
  base_seed = 700,
  n_cores = 1,
  outcome_type = c("continuous", "survival"),
  effect_scale = c("loghr", "rmst"),
  surv_h0 = 0.1,
  surv_event_frac = 0.6,
  surv_censor_rate = NULL
)

Arguments

trained_gan

An iconic_gan (or NULL to use default texture).

conf_grid

Confounding-strength values to sweep. Default c(0.2, 0.5, 0.8).

coverage_grid

Negative-control coverage values in [0,1]. Default c(0.3, 0.7, 1).

k_grid

Numbers of latent confounders to sweep. Default 1.

nc_model

Negative-control model (function or name). Default "proxy".

n_iter

Replicates per scenario. Default 50.

n_samples

Samples per replicate. Default 500.

n_features

Features per replicate. Default 20.

beta_X, alpha_M, beta_M

Causal paths (ground truth). Defaults 0.10 / 0.50 / 0.30.

effect_size

Optional pure-direct total effect override (see run_single_iteration()).

base_seed

Base RNG seed. Default 700.

n_cores

Parallel workers across replicates. Default 1.

outcome_type

"continuous" (default) or "survival" When survival, the DGP generates time-to-event outcomes and estimation uses the Cox / RMST survival drivers via iconic_estimate().

effect_scale

"loghr" (default) or "rmst". Only used when outcome_type = "survival".

surv_h0

Baseline hazard for the survival DGP. Default 0.1.

surv_event_frac

Target fraction of observed events. Default 0.6.

surv_censor_rate

Explicit censoring rate. Default NULL.

Value

A list with summary (one row per scenario x method, with conf_strength, coverage, k, true_total and the columns from summarise_results) and grid (the scenario grid).

Examples

sens <- gan_sensitivity(NULL, conf_grid = 0.8, coverage_grid = 0.7,
  n_iter = 2, n_samples = 100, n_features = 5)
head(sens$summary)
#>   conf_strength coverage k true_total method       mean     median         sd
#> 1           0.8      0.7 1       0.25  UNADJ  0.6288594  0.6377314 0.07535704
#> 2           0.8      0.7 1       0.25 DIRECT  0.3671867  0.3533574 0.11724910
#> 3           0.8      0.7 1       0.25   COCA -0.6149253 -0.5515430 0.22603691
#> 4           0.8      0.7 1       0.25 IV2SLS  0.2203744  0.2251649 0.09778570
#> 5           0.8      0.7 1       0.25    PGC  0.2784006  0.2934452 0.09297978
#>          bias   abs_bias       rmse power  n
#> 1  0.37885941 0.37885941 0.38554541   1.0 10
#> 2  0.11718666 0.11718666 0.16157144   0.9 10
#> 3 -0.86492528 0.86492528 0.89111120   1.0 10
#> 4 -0.02962563 0.02962563 0.09738335   0.5 10
#> 5  0.02840064 0.02840064 0.09266774   1.0 10