Propagate input uncertainty through a footprint.

Monte Carlo propagation of extension uncertainty: perturb the extension vector with multiplicative lognormal noise (so values stay non-negative and the expected factor is one), re-run the footprint for each draw, and summarise the spread of each output cell. A point estimate with no interval is not a trustworthy result; this turns one into a distribution.

Usage

propagate_fp_uncertainty(run_fn, extensions, cov = 0.1, options = list())

Arguments

run_fn

Function taking a perturbed extension vector and returning a footprint tibble with the grouping columns named in options$by plus a value column. Wrap compute_footprint() with the other arguments fixed.

extensions

Numeric vector of base extensions per sector.

cov

Coefficient of variation of the extensions: one number, or one per sector. Zero means no uncertainty.

options

Named list overriding n (draws, default 200), probs (lower/median/upper quantiles), by (grouping columns) and seed (for reproducible draws).

Value

A tibble with the by columns plus mean, sd, cv, q_low, q_med and q_high per output cell.

Examples

run_fn <- function(ext) {
  tibble::tibble(
    target_area = 1L, target_item = 10L, value = sum(ext)
  )
}
propagate_fp_uncertainty(
  run_fn,
  extensions = c(60, 40),
  cov = 0.1,
  options = list(n = 100, seed = 1, by = c("target_area", "target_item"))
)
#> # A tibble: 1 × 8
#>   target_area target_item  mean    sd     cv q_low q_med q_high
#>         <int>       <int> <dbl> <dbl>  <dbl> <dbl> <dbl>  <dbl>
#> 1           1          10  100.  6.76 0.0673  88.3  99.8   115.