A Review of Causal Decision Making

TL;DR

This paper presents a unified framework for causal decision-making (CDM) by decomposing the problem into three tasks—causal structure learning (CSL), causal effect learning (CEL), and causal policy learning (CPL)—across six data paradigms that cover offline and online settings. It surveys state-of-the-art methods for learning causal graphs, estimating average/heterogeneous and mediated effects, and evaluating or optimizing policies, while addressing violations of core causal assumptions and distributional shifts. The authors illustrate practical workflows through real-data case studies (MIMIC-III and MovieLens) and provide a Python-based repository to implement the reviewed methods. The work emphasizes the integration of CSL, CEL, and CPL to enable principled decision-making in domains such as healthcare and recommender systems, and outlines future directions including robustness to assumption violations, fairness, and scalability.

Abstract

To make effective decisions, it is important to have a thorough understanding of the causal relationships among actions, environments, and outcomes. This review aims to surface three crucial aspects of decision-making through a causal lens: 1) the discovery of causal relationships through causal structure learning, 2) understanding the impacts of these relationships through causal effect learning, and 3) applying the knowledge gained from the first two aspects to support decision making via causal policy learning. Moreover, we identify challenges that hinder the broader utilization of causal decision-making and discuss recent advances in overcoming these challenges. Finally, we provide future research directions to address these challenges and to further enhance the implementation of causal decision-making in practice, with real-world applications illustrated based on the proposed causal decision-making. We aim to offer a comprehensive methodology and practical implementation framework by consolidating various methods in this area into a Python-based collection. URL: https://causaldm.github.io/Causal-Decision-Making.

Figures (8)

• Figure 1: Workflow of the CDM. $f_1$, $f_2$, and $f_3$ represent the impact sizes of the directed edges. Variables enclosed in solid circles are observed, while those in dashed circles are actionable.
• Figure 2: Common data dependence structures (paradigms) in CDM. Detailed notations and explanations can be found in Section \ref{['sec:paradigms']}.
• Figure 3: Causal Policy Learning
• Figure 4: An illustration of Simpson's Paradox.
• Figure 5: Left: Illustration of the causal relationship between the customer being a new father or not, beer purchasing, and diaper purchasing, where solid lines represent the true model, and the dashed line corresponds to the spurious correlation between beer purchasing and diaper purchasing. Right: Relationship between various causal structures. Nodes $A$, $B$, and $C$ belong to the necessary and sufficient causal graph for the target outcome $Y$ and are depicted inside the green solid square. Among them, nodes $B$ and $C$ are members of the Markov blanket of $Y$, enclosed by the blue dotted square. Node $S$ is the spurious variable to $Y$, while nodes $N$ and $M$ are unrelated to the target.

Theorems & Definitions (15)

• Definition 3.1
• Definition 3.2
• Definition 3.3
• Definition 3.4
• Definition 3.5
• Definition 3.6
• Definition 3.7
• Definition 3.8
• Definition 3.9
• Definition 3.10