Time-varying causal effect beta(t) over a sliding window — causal_effect_time

Closes the Pilar 1 x Pilar 3 loop: the v1.4.0 backdoor-adjusted estimator is applied within non-overlapping (or overlapping) windows of the temporal frame, producing a beta_hat(t) trajectory with bootstrap CIs. Mann-Kendall tests for a significant trend.

Usage

causal_effect_time_varying(
  frame,
  dag,
  exposure,
  outcome,
  window = 24L,
  step = 6L,
  adjustment = NULL,
  B = 200L,
  min_n = 30L,
  seed = NULL
)

Arguments

frame: A data frame with columns t, lon, lat, the exposure, outcome and any adjustment columns.
dag: A dagitty DAG. Used only to derive the adjustment set when adjustment = NULL.
exposure, outcome: Character column names.
window: Integer; number of distinct t values per window.
step: Integer; how many t values to advance the window.
adjustment: Optional character vector of adjustment columns.
B: Integer bootstrap replicates per window. 0 disables CI estimation (just a point estimate per window).
min_n: Minimum in-window sample size to fit a window. Windows smaller than this yield NA estimates.
seed: Optional RNG seed for bootstrap reproducibility.

Value

A data.frame of class edaphos_causal_4d with columns t_start, t_end, t_centre, n, beta_hat, se, ci_lo, ci_hi.