Skip to contents

Diagnose robustness to added ("missing") edges

Source: R/add_edges.R
add_edges_robustness.Rd

Tests how the adjustment set and identification respond to edges the root DAG assumes are absent (DAGWOOD exclusion branches). Each added edge is evaluated as its own branch DAG (root + that edge).

Usage

add_edges_robustness(dag, exposure, outcome, add_edges, formula = NULL)

Arguments

dag

A dagitty DAG.

exposure, outcome

Optional; inferred from the DAG when omitted.

add_edges

Character vector like c("Z -> Y", "X <-> Y"). Directed (->, <-) and bidirected (<->, latent common cause) edges are supported.

formula

Optional model formula or engine call; used only to decide whether a role-flipping covariate is in your specification.

Value

A DAGassist_addedge_summary object.

Examples

# What if the DAG is missing an arrow, or has unmeasured confounding?
add_edges_robustness(toy_dag, add_edges = c("A -> X", "X <-> Y"))
#> 
#> Edge-addition (exclusion) robustness:
#> - edges tested: 2
#>   - A -> X: minimal changed: yes; canonical changed: no
#>         new minimal set(s): {A, Z}
#>         role changes: A: nco->confounder
#>   - X <-> Y: effect NOT identifiable if this pathway exists (no adjustment set blocks it)
#> - re-estimation recommended: yes