Skip to contents

Monte Carlo vs. analytical power for PPI / PPI++ mean estimation

simulate_power.Rd

[Soft-deprecated]

Prefer power_ppi_mean() for PPI++ power calculations. This function is retained for backward compatibility with power_curve_mean() and type1_error_curve_mean().

Usage

simulate_power(
  delta,
  N,
  n,
  alpha = 0.05,
  R = 1e+05,
  var_f = NULL,
  var_res = NULL,
  sigma_y2 = NULL,
  sigma_f2 = NULL,
  cov_y_f = NULL,
  metrics = NULL,
  metric_type = NULL,
  m_labeled = n,
  correction = TRUE
)

Arguments

delta

Effect size \(\theta - \theta_0\).

N

Unlabeled sample size.

n

Labeled sample size.

alpha

Two-sided significance level.

R

Number of Monte Carlo draws (default 100000).

var_f

Variance of \(f(X)\).

var_res

Variance of residuals \(Y - f(X)\).

sigma_y2

Optional outcome variance (needed for PP++ if cov_y_f is supplied).

sigma_f2

Optional prediction variance (needed for PP++ if cov_y_f is supplied).

cov_y_f

Optional covariance \(\operatorname{Cov}(Y, f(X))\); when present, PP++ power is returned.

metrics

Optional predictive summaries to recover missing moments.

metric_type

Character string describing metrics.

m_labeled

Labeled sample size associated with metrics (defaults to n).

correction

Apply finite-sample corrections when recovering moments.

Value

Named numeric vector with Exact_PP, Empirical_PP, and when possible Exact_PPplus, Empirical_PPplus.