NEWS

---

marbounds 0.1.0

Initial release of the marbounds package for bounding causal effects under mixed informative and non-informative missingness.

Features

• mar_bounds() - Main function for computing bounds on causal effects
• Support for multiple estimands: ATE, composite ATE (Ψ₁), separable direct effect (Ψ₂)
• Multiple assumption types:
  • General bounds (no assumptions)
  • Bounded proportion of informative missingness (δ)
  • Monotonicity (positive/negative)
  • Bounded outcome risk (τ)
  • Point identification under known sensitivity parameters
• SuperLearner integration for nuisance parameter estimation with V-fold cross-fitting
• Multiplier bootstrap for simultaneous inference over parameter grids
• Influence function-based estimators with asymptotic standard errors

Reference

Rubinstein, M., Agniel, D., Han, L., Horvitz-Lennon, M., & Normand, S.-L. (2026). Bounding causal effects with an unknown mixture of informative and non-informative missingness. Journal of Causal Inference (Accepted). https://arxiv.org/pdf/2411.16902