Compute the frequentist power at the nominal type I error rate — frequentist_power_at_nominal

Computes the power of the separate and the pooled analysis at the nominal type I error rate, which the plots use as the two baselines every borrowing method is read against.

Both baselines are a property of the design alone, while the results frame holds one row per design and method-parameter combination: the paper's environment repeats each of its 330 designs 56 times. They are therefore computed once per design and copied to the rows that share it. That is exact rather than an approximation, because simulate_test_p_values() reseeds from simulation_config$seed on every call, so the repeats were identical to the last bit anyway.

Usage

frequentist_power_at_nominal_tie(
  results,
  analysis_config,
  simulation_config,
  parallelization = FALSE,
  n_replicates = 1000,
  trial_cache = NULL,
  cluster = NULL
)

Arguments

results: The results data frame.
analysis_config: The analysis configuration.
simulation_config: The simulation configuration.
parallelization: Whether the caller asked for parallelism, as analysis_runs_in_parallel() resolves it. A design costs seconds here - two power computations, either of which may be a simulation - so a run with many of them is worth spreading over a cluster.
n_replicates: Number of Monte Carlo replicates the simulated power estimates are built from, for the case studies analytical_power() cannot be used for.
trial_cache: Optional environment of simulated trials left by frequentist_power_at_equivalent_tie(). Both baselines read the same trials, so a design found there is not simulated again.
cluster: Optional shared cluster from new_analysis_cluster(). NULL starts one for this call if the work warrants it.

Value

The results data frame with the six baseline power columns added.