Skip to contents

Build a DirectEffects::sequential_g() formula from a DAG (and optionally a base model formula)

Source: R/seqg_formula.R
seqg_formula.Rd

Returns a formula of the form Y ~ A + X | Z | M required by DirectEffects::sequential_g(), where:

  • A is the treatment (exposure),

  • X are baseline covariates,

  • Z are intermediate (post-treatment) covariates affected by A that affect both M and Y,

  • M are mediators (blip-down model terms).

If formula is provided, X is (by default) constrained to the RHS terms that appear in that base formula (mirroring DAGassist's current ACDE construction).

Usage

seqg_formula(
  dag,
  formula = NULL,
  data = NULL,
  exposure = NULL,
  outcome = NULL,
  m = NULL,
  x = NULL,
  z = NULL,
  include_descendants = FALSE,
  fe = NULL,
  fe_as_factor = TRUE,
  drop_fe_collinear = TRUE,
  strict_rhs = TRUE,
  return_parts = TRUE
)

Arguments

dag

A dagitty object (or a character DAGitty string).

formula

Optional base formula (or an engine call like fixest::feols(...)). If supplied, the function uses its RHS to define candidate covariates and to pick up fixed-effects/grouping variables (which can be added to X).

data

Optional data.frame. Used to (i) drop terms that do not exist in the data, and (ii) optionally drop covariates collinear with fixed effects (important for seqg stability).

exposure, outcome

Optional. If missing, inferred from dagitty::exposures() / dagitty::outcomes().

m, x, z

Optional character vectors overriding inferred mediator (M), baseline (X), and intermediate (Z) terms.

include_descendants

Logical; if TRUE, allow descendants of mediators to be included in M (conservatively restricted to descendants that remain ancestors of Y and descendants of A).

fe

Optional character vector of FE/grouping variables to include in X (merged with any extracted from formula).

fe_as_factor

Logical; if TRUE, wraps bare FE names as factor(FE) (only when FE exists in data).

drop_fe_collinear

Logical; if TRUE and data is provided, drop any single-variable term that is constant within an FE group (DirectEffects is less forgiving than fixest here).

strict_rhs

Logical; if TRUE and formula is provided, X is drawn only from RHS terms in formula. If FALSE, X is drawn from the DAG node set (pre-treatment ancestors) even if not in formula.

return_parts

Logical; if TRUE, returns a list with components and metadata; otherwise returns just the formula.

Value

A formula (or a list with $formula, $x, $z, $m, $fe, $dropped_fe).