Monte Carlo power for the PPI++ mean estimator

Usage

simulate_ppi_mean(
  R,
  n,
  N,
  alpha = 0.05,
  delta = NULL,
  sigma_y2 = NULL,
  sigma_f2 = NULL,
  cov_y_f = NULL,
  var_f = NULL,
  var_res = NULL,
  lambda = NULL,
  lambda_type = c("oracle", "plugin", "user"),
  moments = NULL,
  seed = 1,
  family = stats::gaussian(),
  theta0 = 0
)

Arguments

R: Integer, number of Monte Carlo draws.
n: Labeled sample size.
N: Unlabeled sample size.
alpha: Two-sided test level.
delta: Effect size \(\theta - \theta_0\).
sigma_y2: Total variance of \(Y\).
sigma_f2: Variance of \(f(X)\).
cov_y_f: Covariance between \(Y\) and \(f(X)\).
var_f: Variance of \(f(X)\) (optional, overrides sigma_f2).
var_res: Residual variance \(Y - f(X)\).
lambda: Optional user-specified blend weight.
lambda_type: One of "oracle", "plugin", or "user".
moments: Optional list with entries for population moments.
seed: RNG seed.
family: A stats::family() object (gaussian() or binomial()).
theta0: Null hypothesis value \(\theta_0\).

Value

A list with empirical and analytical power, average standard error, average lambda, and simulation details.