Simulate Power to Rank the Best Group Using Binomial Outcomes

Estimates the empirical power to rank the most promising group as the best, based on binomial outcomes, via simulation.

Usage

sim_power_best_bin_rank(
  noutcomes,
  p1,
  dif,
  weights,
  ngroups,
  npergroup,
  nsim,
  conf.level = 0.95
)

Arguments

noutcomes: Integer. Number of outcomes to evaluate.
p1: Numeric. Event probability in the first group, i.e. the true best group (scalar or vector of length noutcomes).
dif: Numeric. Amount by which the first (true best) group's probability exceeds the other ngroups - 1 groups (scalar or vector of length noutcomes).
weights: Numeric vector. Weights for each outcome. If scalar, applied equally.
ngroups: Integer. Number of groups.
npergroup: Integer or vector. Sample size per group.
nsim: Integer. Number of simulations.
conf.level: Numeric. Confidence level for the empirical power estimate#'

Value

An S3 object of class empirical_power_result, which contains the estimated empirical power and its confidence interval. The object can be printed, formatted, or further processed using associated S3 methods. See also empirical_power_result.

Details

Each outcome is assumed to follow an independent binomial distribution. The first group is always the true best group: it is simulated with event probability p1, while the other ngroups - 1 groups share probability p1 - dif. The function sums weighted ranks across multiple outcomes to determine the top group, and estimates the empirical power to correctly identify the first group as the best.

If multiple outcomes are defined, weights can be applied to prioritize some outcomes over others. Weights are automatically scaled to sum 1. For each outcome, groups are ranked from lowest (1) to highest (ngroups) observed proportion; the group with the highest total weighted rank across outcomes is considered the best. Power is the proportion of simulations in which that group is the first group.

Examples

  sim_power_best_bin_rank(
  noutcomes = 2,
  p1 = 0.80,
  dif = 0.15,
  weights = 1,
  ngroups = 3,
  npergroup = 30,
  nsim = 1000,
  conf.level = 0.95)
#> Empirical Power Result
#> ----------------------- 
#> Power:       0.8820
#> 95% CI:      [0.8604, 0.9013]
#> Simulations: 1000