Skip to contents

Block-bootstrap the backdoor-adjusted direct effect

Source: R/causal_posterior.R
causal_effect_bootstrap.Rd

Resamples clusters (not rows) with replacement and refits the adjusted OLS on each resample, returning the vector of direct-effect coefficients. When the clustering structure is k-means on (lon, lat) (as in the Cerrado pipeline from v1.4.0 onward), this produces spatial-clustering-aware confidence intervals that the asymptotic confint(lm) interval ignores.

Usage

causal_effect_bootstrap(
  data,
  dag,
  exposure,
  outcome,
  adjustment = NULL,
  cluster = "kmeans_cluster",
  B = 500L,
  effect = c("direct", "total"),
  seed = NULL
)

Arguments

data

A data frame with the exposure, outcome, adjustment columns and a cluster-id column.

dag

A dagitty DAG (only used when adjustment is NULL).

exposure, outcome

Character; column names.

adjustment

Character vector of adjustment-set column names (defaults to the minimal set from the DAG).

cluster

Character; name of the cluster-id column in data.

B

Integer; number of bootstrap resamples. Defaults to 500L.

effect

One of "direct", "total" (passed to the adjustment set resolver when adjustment = NULL).

seed

Optional integer for reproducibility.

Value

Numeric vector of length B with one direct-effect estimate per bootstrap resample.