Changelog • DAGassist

DAGassist v0.3.0

PDAG robustness features: pdag and uncertain_edges arguments for DAGassist() and a standalone pdag_robustness() function diagnose how adjustment sets and covariate roles change when introducing uncertain edges.
Added automated balance diagnostics: the console report now flags variables whose |SMD| > 0.1 between the Original and DAG-derived models, addressing missingness-induced apples-to-oranges model comparisons.
Additional robustness feature: add_edges_robustness to evaluate robustness to hypothesized missing edges.
Extended support for balance and weight diagnostics under non-console export formats.
Changed estimand terminology for methodological accuracy (SATE -> total effect; SACDE -> direct effect), as DAGassist cannot recover a true SATE or SACDE from continuous treatments.
Bug fixes

Changed valid estimand parameter values (“SATE” -> “total”; “SACDE” -> “direct”).
Phased out support for “ATT” estimand due to estimand variability across binary and continuous treatments. Future updates may reintroduce the ATT estimand.

This is the CRAN version.

CRAN release: 2026-02-20

CRAN release: 2025-11-11

Supports multiple exposures
Detects DAG conditions like M-Bias and Butterfly Bias
Added neutral control (on treatment and outcome) and descendant of confounder on/off back-door path roles
Enabled console LaTeX / markdown output
Updated export functions to support new expanded roles table, and added legend to define abbreviations
Added Canonical -neutral control (on treatment/outcome) comparison column options

Added dotwhisker output functionality
Added support for individually generating and exporting roles and comparison tables
Added bivariate specification to comparison table by default
Significantly enhanced vignette, article, and roxygen documentation

CRAN release: 2025-09-21

Removed compare_specs() and print.compare_specs(), which were unused and causing issues.
Updated DESCRIPTION to reference dagitty and relevant article under methods.

Fixed imply logic. Previously, for the roles table, imply = FALSE would only display RHS variables, but still assessed all DAG-based relationships.
Renamed desc(Y) IO, for intermediate outcome
Dropped desc(X) as a roles column
Added dMed (descendant of mediator) and dCol (descendant of collider)
Updated documentation and added vignettes

Built pkgdown site and several supporting vignettes.
Significantly enhanced package documentation in preparation for CRAN release.
Trimmed exported functions.
Enhanced and documented engine compatibility.
Added omit_factors and omit_intercept parameters.
Reworked imply parameter so it restricts evaluation to variables listed in RHS.
Fixed cluster argument bug.

variable labels across roles + model tables (LaTeX, .docx, .xlsx, .txt).
unevaluated regressors note: list RHS terms not in the DAG, shown in all outputs.
LaTeX + Word + Excel + Text:
- keep factor dummies and fixed effects by default.
- better longtblr conversion/spacing; improved dynamic column width weighting and line breaking for long labels.
- auto-escape labels for neat output.
- cleaner summaries and label support. ## Changed
- consolidated helpers (assist_helpers.R + export_helpers.R).
- improved handling of nuisance vars (FEs/tails).
- significantly improved documentation to prep for CRAN release.
- broader validation against real replication data.

Acyclicity checks and clearer validation messages in validate.R.
Canonical adjustment sets alongside minimal sets.
Support for multiple minimal adjustment sets.
Export to LaTeX, Word (.docx), Excel (.xlsx), and plain text (.txt).
export_report() helper to produce a publication-grade, half-page robustness report from a single call.
New imply parameter to control DAG-implied covariate additions

Fixed-effects / IV parts of formulas (e.g., | region + time) are preserved; validation looks only at the pre-| part.