Diagnose robustness to added ("missing") edges

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

g <- dagitty::dagitty("dag { Z->X; X->Y }")
dagitty::exposures(g) <- "X"; dagitty::outcomes(g) <- "Y"
add_edges_robustness(g, add_edges = c("Z -> Y", "X <-> Y"))
#> 
#> Edge-addition (exclusion) robustness:
#> - edges tested: 2
#>   - Z -> Y: minimal changed: yes; canonical changed: no
#>         new minimal set(s): {Z}
#>         role changes: Z: nct->confounder
#>   - X <-> Y: effect NOT identifiable if this pathway exists (no adjustment set blocks it)
#> - re-estimation recommended: yes