[Paper Review] Causal Inference and Data Fusion in Econometrics

Paul Hünermund, Elias Bareinboim|arXiv (Cornell University)|Dec 19, 2019

Advanced Causal Inference TechniquesMathematics18 citations

TL;DR

This paper introduces a unified, nonparametric framework for causal inference in econometrics by integrating do-calculus and data fusion techniques from artificial intelligence. It enables automated identification of causal effects from diverse, heterogeneous data sources—such as observational, experimental, and selection-biased samples—using graphical models, thereby overcoming limitations of traditional econometric methods in handling unobserved confounding, transportability, and data heterogeneity.

ABSTRACT

Learning about cause and effect is arguably the main goal in applied econometrics. In practice, the validity of these causal inferences is contingent on a number of critical assumptions regarding the type of data that has been collected and the substantive knowledge that is available. For instance, unobserved confounding factors threaten the internal validity of estimates, data availability is often limited to non-random, selection-biased samples, causal effects need to be learned from surrogate experiments with imperfect compliance, and causal knowledge has to be extrapolated across structurally heterogeneous populations. A powerful causal inference framework is required to tackle these challenges, which plague most data analysis to varying degrees. Building on the structural approach to causality introduced by Haavelmo (1943) and the graph-theoretic framework proposed by Pearl (1995), the artificial intelligence (AI) literature has developed a wide array of techniques for causal learning that allow to leverage information from various imperfect, heterogeneous, and biased data sources (Bareinboim and Pearl, 2016). In this paper, we discuss recent advances in this literature that have the potential to contribute to econometric methodology along three dimensions. First, they provide a unified and comprehensive framework for causal inference, in which the aforementioned problems can be addressed in full generality. Second, due to their origin in AI, they come together with sound, efficient, and complete algorithmic criteria for automatization of the corresponding identification task. And third, because of the nonparametric description of structural models that graph-theoretic approaches build on, they combine the strengths of both structural econometrics as well as the potential outcomes framework, and thus offer an effective middle ground between these two literature streams.

Motivation & Objective

• To address the persistent challenges in econometric causal inference, including unobserved confounding, selection bias, and population heterogeneity.
• To unify structural econometrics and potential outcomes frameworks through a nonparametric, graph-based approach grounded in causal diagrams.
• To enable automation of the identification task in causal inference by leveraging algorithmic criteria from artificial intelligence.
• To facilitate data fusion across multiple studies and populations by formalizing transportability and extrapolation of causal effects.
• To provide a systematic methodology for combining imperfect, non-random, and structurally heterogeneous data to estimate causal effects

Proposed method

• Uses directed acyclic graphs (DAGs) to represent structural causal models, encoding conditional independencies and causal relationships nonparametrically.
• Applies do-calculus—a set of three inference rules—to symbolically transform causal queries (e.g., P(Y|do(X))) into estimable expressions using observable data distributions.
• Employs the do-operator to formalize interventions by replacing structural equations with constant values, enabling counterfactual reasoning.
• Introduces transportability theory to extrapolate causal effects across different populations by identifying shared structural mechanisms.
• Leverages data fusion techniques to combine observational, experimental, and selection-biased data sources into a single estimable expression.
• Uses algorithmic criteria for completeness and efficiency, enabling automation of the identification process without requiring parametric assumptions.

Experimental results

Research questions

• RQ1 How can causal effects be identified and estimated when data is subject to selection bias or unobserved confounding?
• RQ2 Can a unified framework be developed that integrates structural econometrics and potential outcomes approaches using nonparametric graphical models?
• RQ3 To what extent can causal inference be automated using algorithmic rules like do-calculus, even without full knowledge of structural mechanisms?
• RQ4 How can causal knowledge from one population be validly transported to another when populations differ structurally?
• RQ5 What role does data fusion play in enhancing the robustness and generalizability of causal estimates across diverse data sources?

Key findings

• The do-calculus provides a complete and sound set of rules for transforming causal queries into estimable expressions, enabling automated identification of causal effects.
• Graphical models allow for nonparametric representation of causal relationships, preserving flexibility while ensuring analytical rigor.
• Transportability theory enables valid extrapolation of causal effects across structurally heterogeneous populations by identifying shared mechanisms.
• Data fusion techniques allow researchers to combine observational, experimental, and selection-biased data to estimate causal effects that are not identifiable from any single source alone.
• The integration of AI-based causal inference tools with econometric methodology offers a path toward fully automated, reliable, and generalizable causal estimation.
• The framework supports treatment effect heterogeneity naturally, without requiring parametric functional form assumptions.