Skip to contents

Benchmark wrapper: Pilar 1 – DAG-adjusted OLS + parametric bootstrap

Source: R/benchmark_wosis.R
benchmark_fit_p1_causal.Rd

Restricts OLS to covariates that appear in the supplied DAG as direct parents of the outcome (falling back to the full covariate set when the DAG has no matching variables) and generates a parametric-bootstrap predictive posterior. Intended for the v3.1.0 WoSIS head-to-head; the exported API is stable enough for use on any regional subset that follows the same schema.

Usage

benchmark_fit_p1_causal(
  train,
  test,
  cov_cols,
  dag = NULL,
  n_boot = 300L,
  seed = 1L,
  calibrate = TRUE
)

Arguments

train, test

Data frames with a response column soc, coord columns lon/lat, and the columns listed in cov_cols.

cov_cols

Character vector of candidate covariate names.

dag

A dagitty DAG object (or NULL to use the full covariate set).

n_boot

Number of bootstrap resamples. Default 300.

seed

Optional RNG seed.

calibrate

Logical (default TRUE). When TRUE, an estimate of the in-sample residual standard deviation is added as iid Gaussian noise to every posterior sample so the predictive posterior carries BOTH epistemic (bootstrap-spread) AND aleatoric (residual) uncertainty. Set to FALSE to reproduce the v3.1.0 epistemic-only behaviour.

Value

An edaphos_posterior with method = "bootstrap", query_type = "map".