summarise_estimator {SimNPH}

Generic Summarise function for esitmators

Description

Generic Summarise function for esitmators

Usage

summarise_estimator(
  est,
  real,
  lower = NULL,
  upper = NULL,
  null = NULL,
  est_sd = NULL,
  name = NULL
)

Arguments

est

estimator, expression evaluated in results

real

real summary statistic, expression evaluated in condition

lower

lower CI, expression evaluated in results

upper

upper CI, expression evaluated in results

null

parameter value under the null hypothesis

est_sd

standard deviation estimated by the method, evaluated in results

name

name for the summarise function, appended to the name of the analysis method in the final results

Details

The different parameters are evaluated in different envionments, est, lower, upper, est_sd refer to output of the method and are evaluated in the results dataset. real refers to a real value of a summary statistic in this scenario and is therefore evaluated in the condition dataset. null and name are constants and directly evaluated when the function is defined. The argument null, the parameter value under the null hypothesis is used to output the rejection rate based on the confidence intervall. Which is output in the column null_cover

Value

A function that can be used in Summarise that returns a data frame with summary statistics of the performance measures in the columns.

Examples

# generate the design matrix and append the true summary statistics
condition <- merge(
  assumptions_delayed_effect(),
  design_fixed_followup(),
  by=NULL
) |>
  tail(4) |>
  head(1) |>
  true_summary_statistics_delayed_effect(cutoff_stats = 15)

# create some summarise functions
summarise_all <- create_summarise_function(
  coxph=summarise_estimator(hr, gAHR_15, hr_lower, hr_upper, name="gAHR"),
  coxph=summarise_estimator(hr, hazard_trt/hazard_ctrl, hr_lower, hr_upper, name="HR"),
  coxph=summarise_estimator(hr, NA_real_, name="NA")
)

# runs simulations
sim_results <- runSimulation(
  design=condition,
  replications=10,
  generate=generate_delayed_effect,
  analyse=list(
    coxph=analyse_coxph()
  ),
  summarise = summarise_all
)

# mse is missing for the summarise function in which the real value was NA
sim_results[, names(sim_results) |> grepl(pattern="\\.mse$")]
# but the standard deviation can be estimated in all cases
sim_results[, names(sim_results) |> grepl(pattern="\\.sd_est$")]