Treatment effect estimators and marginal-effects extenders

Two function families live here:

Details

Treatment-effect estimators (legacy)

• estimate_ate() – IPW-weighted OLS ATE.

• estimate_plr() – Partially Linear Regression via DoubleML; falls back to base R cross-fit ridge.

• estimate_pliv() – Partially Linear IV (LATE) via DoubleML; falls back to 2SLS.

• estimate_ate_gcomputation() – G-computation ATE. Thin wrapper over stdReg::stdGlm() when installed; falls back to inline bootstrap standardisation otherwise.

• sensitivity_rosenbaum() – Rosenbaum bounds. Thin wrapper over rbounds::psens() when installed; otherwise normal-approximation Wilcoxon signed-rank bounds.

• e_value() – VanderWeele-Ding E-value. Thin wrapper over EValue::evalues.OLS() when installed; otherwise the closed-form continuous-scale RR proxy.

Marginal-effects extenders (Phase 1.j additions; thin wrappers over Vincent Arel-Bundock's universal API and the emmeans / broom ecosystems):

• morie_effects_emmeans() -> emmeans::emmeans().

• morie_effects_predictions() -> marginaleffects::predictions().

• morie_effects_comparisons() -> marginaleffects::comparisons().

• morie_effects_slopes() -> marginaleffects::slopes().

• morie_effects_tidy() -> broom::tidy() (falls back to a summary()-based tidy frame when broom is unavailable).

Each extender requires the underlying CRAN package and signals a clean stop() when it is missing, leaving the upstream model object untouched. They return the underlying package's native object verbatim so downstream code (e.g. ggplot2 plumbing, rmorie's MRM step) keeps working with the canonical API.

References

Chernozhukov et al. (2018); Robins (1986); VanderWeele & Ding (2017); Rosenbaum (2002); Arel-Bundock (2024, marginaleffects); Lenth (2024, emmeans); Robinson, Hayes & Couch (2024, broom).