Derive bootstrap method for a statistic

Infers whether a user-supplied indicator function (fun) behaves as a group-specific statistic (computed independently within each category) or a whole-cube statistic (requiring pooled information across all categories).

Usage

derive_bootstrap_method(
  df,
  fun,
  ...,
  cat_var,
  min_cat = 2,
  max_cat = 5,
  index = -1
)

Arguments

df

A dataframe.

fun

A function which, when applied to df returns the statistic(s) of interest. This function must return a dataframe with a column diversity_val containing the statistic of interest.

...

Additional arguments passed on to fun.

cat_var

A character vector specifying the grouping variable(s) used by fun. The function fun(df, ...) should return a row per group.

min_cat

Integer. The minimum number of categories to include in the "short" dataset used for comparison. Defaults to 2.

max_cat

Integer. The maximum number of categories to include in the "long" dataset. Defaults to 5.

index

Integer. Position from which the max_cat categories will be selected for processing. Use -1 (default) use the final min_cat and max_cat categories in the dataset.

Value

A single character string: either

• "group_specific" or

• "whole_cube"

indicating the appropriate bootstrapping strategy.

Details

The function evaluates bootstrap method by comparing the statistic computed on two filtered subsets of the data:

• Short subset: contains min_cat consecutive categories.

• Long subset: contains max_cat consecutive categories.

Both subsets end at the same category index to ensure overlap.

The indicator function is evaluated on both subsets. If the indicator values for the overlapping categories match exactly, the function is considered group-specific. Otherwise, it is assumed to require whole-cube bootstrapping.

See also

Other indicator_uncertainty_helper: boot_list_to_dataframe(), bootstrap_cube_raw(), calculate_boot_ci_from_boot(), resolve_bootstrap_method()

Examples

# 1. Calculate mean per group
sepal_length_per_species <- function(x, f) {
  out_df <- aggregate(Sepal.Length ~ Species, x, f)
  names(out_df) <- c("Species", "diversity_val") # Rename columns
  return(out_df)
}
sepal_length_per_species(iris, mean)
#>      Species diversity_val
#> 1     setosa         5.006
#> 2 versicolor         5.936
#> 3  virginica         6.588

# The mean is calculated per species, so we expect the group-specific method
derive_bootstrap_method(
  df = iris,
  fun = sepal_length_per_species,
  cat_var = "Species",
  min_cat = 1,
  max_cat = 3,
  mean
)
#> [1] "group_specific"

# 2. Calculate indicator based on whole dataset
sepal_length_per_species2 <- function(x, f) {
  out_df <- aggregate(Sepal.Length ~ Species, x, f)
  out_df$Sepal.Length <- out_df$Sepal.Length / nrow(out_df)
  names(out_df) <- c("Species", "diversity_val") # Rename columns
  return(out_df)
}
sepal_length_per_species2(iris, mean)
#>      Species diversity_val
#> 1     setosa      1.668667
#> 2 versicolor      1.978667
#> 3  virginica      2.196000

# Now we expect the whole-cube method
derive_bootstrap_method(
  df = iris,
  fun = sepal_length_per_species2,
  cat_var = "Species",
  min_cat = 1,
  max_cat = 3,
  mean
)
#> [1] "whole_cube"