Discussion of"Causal and counterfactual views of missing data models"by Razieh Nabi, Rohit Bhattacharya, Ilya Shpitser,&James M. Robins

2025-06-16

AI Summary

This paper addresses the nonignorable missingness mechanism—Missing Not at Random (MNAR)—which is fundamentally unidentifiable under standard assumptions. Methodologically, it introduces a novel framework integrating causal inference with missing-data modeling, centered on missingness-directed acyclic graphs (m-DAGs) to formally characterize identifiability conditions. By combining counterfactual logic with nonparametric independence constraints, the approach enables rigorous identifiability assessment without parametric assumptions or sensitivity analysis. Theoretical contributions include: (i) precise characterization of identifiability boundaries for broad classes of MNAR models; (ii) establishment of a formal theoretical interface between causal inference tools and missingness mechanism modeling; and (iii) extension of causal graph frameworks to jointly identify both the missing-data distribution and causal effects. The proposed framework delivers a unified solution to MNAR problems that is both theoretically sound and practically implementable.

Technology Category

• Reasoning under Uncertainty: Causality
• Machine Learning: Causal Learning
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning

Application Category

• Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

We congratulate Nabi et al. (2022) on their impressive and insightful paper, which illustrates the benefits of using causal/counterfactual perspectives and tools in missing data problems. This paper represents an important approach to missing-not-at-random (MNAR) problems, exploiting nonparametric independence restrictions for identification, as opposed to parametric/semiparametric models, or resorting to sensitivity analysis. Crucially, the authors represent these restrictions with missing data directed acyclic graphs (m-DAGs), which can be useful to determine identification in complex and interesting MNAR models. In this discussion we consider: (i) how/whether other tools from causal inference could be useful in missing data problems, (ii) problems that combine both missing data and causal inference together, and (iii) some work on estimation in one of the authors' example MNAR models.

Problem

• Explores causal views in missing data problems
• Addresses missing-not-at-random using nonparametric methods
• Uses m-DAGs to identify complex MNAR models

Innovation

• Uses causal/counterfactual tools for missing data
• Employs nonparametric independence restrictions for identification
• Utilizes missing data directed acyclic graphs (m-DAGs)