Identifying Macro Causal Effects in C-DMGs over DMGs
Simon Ferreira, Charles K. Assaad · 2025
Read Identifying Macro Causal Effects in C-DMGs over DMGs by Simon Ferreira, Charles K. Assaad (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-19650.txt.
Introduction
Simon Ferreira and Charles Assaad settle a question about when cause and effect can be read off coarse causal diagrams. Their 2025 paper works with cluster graphs, which group variables into bundles instead of spelling out every link. Such bundles suit real systems where a full diagram is out of reach, but the underlying system may loop back on itself.
The do-calculus is the standard test for whether an effect can be identified from a graph. For cluster graphs built over acyclic systems, earlier work needed size conditions on the clusters. Over the cyclic kind studied here, the authors prove the test works with no conditions at all. They also show that the known signs of non-identifiability carry over to a subset of these graphs.
Catalog record
| CREATOR | Simon Ferreira, Charles K. Assaad |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 133 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2506-19650
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Identifying Macro Causal Effects in C-DMGs over DMGs Simon Ferreira, Charles K. Assaad Abstract The do-calculus is a sound and complete tool for identifying causal effects in acyclic directed mixed graphs (ADMGs) induced by structural causal models (SCMs). However, in many real-world applications, especially in high-dimensional setting, constructing a fully specified ADMG is often infeasible. This limitation has led to growing interest in partially specified causal representations, particularly through cluster-directed mixed graphs (C-DMGs), which group variables into clusters and offer a more abstract yet practical view of causal dependencies. While these representations can include cycles, recent work has shown that the do-calculus remains sound and complete for identifying macro-level causal effects in C-DMGs over ADMGs under the assumption that all clusters size are greater than 1 1 . Nevertheless, real-world systems often exhibit cyclic causal dynamics at the structural level. To account for this, input-output structural causal models (ioSCMs) have been introduced as a generalization of SCMs that allow for cycles. ioSCMs induce another type of graph structure known as a directed mixed graph (DMG). Analogous to the ADMG setting, one can define C-DMGs over DMGs as high-level representations of causal relations among clusters of variables. In this paper, we prove that, unlike in the ADMG setting, the do-calculus is unconditionally sound and complete for identifying macro causal effects in C-DMGs over DMGs. Furthermore, we show that the graphical criteria for non-identifiability of macro causal effects previously established C-DMGs over ADMGs naturally extends to a subset of C-DMGs over DMGs. 1 Introduction Understanding and identifying causal effects is a central goal in many scientific disciplines. In recent years, structural causal models (SCMs) have emerged as a foundational framework for reasoning about causality. These models encode causal assumptions through structural equations and are typically represented by acyclic directed mixed graphs (ADMGs), which capture both causal and confounding relationships. Within this framework, the do-calculus [ Pearl 1995 ] —based on the notion of d-separation [ Pearl 1988 ] —provides a complete and sound set of inference rules for identifying causal effects from observational data, assuming the causal structure is fully specified. However, SCMs do not fully capture systems with cyclic causal dependencies at the structural level, which are common in public health, biology, economics, and engineering systems. For example, there can be a cyclic relation between poor mental health ( e.g. , depression or anxiety) and substance use ( e.g. , alcohol, drugs). The worsening of mental health and increase in substance use can occur in tight time-frames (daily or even hourly), especially in high-risk populations. Over time, they may reach a cyclic equilibrium where both reinforce each other without a clear causal ordering. To address this, the notion of input-output structural causal models (ioSCMs) has been proposed [ Forré and Mooij 2020 ] . These models generalize SCMs by allowing for cycles and induce a new class of graphs known as directed mixed graphs (DMGs) [ Richardson 1997 , Forré and Mooij 2017 , Forré and Mooij 2018 , Forré and Mooij 2020 , Boeken and Mooij 2024 ] , which provide a richer representation of causal structures. Furthermore, Forré and Mooij 2020 introduced an extension of d-separation to DMGs, called \sigma -separation, and showed that the do-calculus, when replacing d-separation by \sigma -separation becomes sound for identifying causal effects in DMGs [ Forré and Mooij 2020 ] . However, in many real-world applications—particularly those involving high-dimensional data or limited domain knowledge—it is often unrealistic to assume a complete specification of the underlying causal graph. This has motivated the development of partially specified graphical models [ Maathuis and Colombo 2013 , Perkovic et al. 2016 , Perkovic 2020 , Jaber et al. 2022 , Wang et al. 2023 , Assaad et al. 2023 , Anand et al. 2023 , Wahl et al. 2024 , Reiter et al. 2024 , Boeken and Mooij 2024 , Ferreira and Assaad 2024 , Ferreira and Assaad 2025a ] , and in particular cluster graphs. Cluster graphs abstract away some of the fine-grained details by grouping variables into clusters, thus offering a more flexible and scalable representation of complex systems. Importantly, cluster graphs allow for cycles, which can arise naturally in feedback systems or time-dependent processes, complicating the analysis compared to traditional ADMGs. In these graphs, causal effects can be separated into two types: a micro causal effect where the interest is the effect of variable within a cluster on another variable in another cluster; and the macro causal effect where the interest in the effect of a set of an entire cluster on another entire cluster. In this work, we focus on the latter. Anand et al. 2023 , Tikka et al. 2023 have shown that do-calculus (the version using d-separation) remains both sound and complete for identifying macro causal effects when the cluster graph representing ADMGs is acyclic. Ferreira and Assaad 2025a , Ferreira and Assaad 2025b showed that the do-calculus (the version using d-separation) is also sound and complete for identifying macro-level causal effects when the cluster graph representing an ADMG is cyclic, denoted here as C-DMG over ADMGs, assuming either that the size of the clusters is unknown, or that each cluster contains more than one variable [ Ferreira and Assaad 2025b ] . Motivated by these developments, we consider the problem of identifying macro-level causal effects in cluster graphs representing DMGs, denoted as C-DMGs over DMGs, a natural generalization of previous work. Our contributions are threefold: • We prove that \sigma -separation [ Forré and Mooij 2018 ] —a fundamental tool in causal reasoning in DMGs—is sound and complete in C-DMGs over DMGs. • We prove that do-calculus (the version using \sigma -separation) [ Forré and Mooij 2020 ] is sound and complete for identifying macro-level causal effects in C-DMGs over DMGs—unconditionally, and without the constraints needed in the case of C-DMGs over ADMGs [ Ferreira and Assaad 2025b , Yvernes 2025 ] . • We show that the graphical characterization of non-identifiability previously developed for C-DMGs over ADMGs [ Ferreira and Assaad 2025a , Ferreira and Assaad 2025b ] also applies for C-DMGs over DMGs under an additional assumption. The remainder of the paper is organized as follows: In Section 2 , we formally presents C-DMGs over DMGs. In Section 3 , we show that \sigma -separation and the do-calculus is sound and complete for macro causal effects in C-DMGs over DMGs and present a graphical characterization for the non-identifiability of these effects. Finally in Section 4 , we conclude the paper while showing its limitations. All proofs are deferred to the appendix. 2 Preliminaries To streamline the presentation and avoid repetitive explanations, we will adopt the unified notation \mathcal{G}^{*}=(\mathbb{V}^{*},\mathbb{E}^{*}) to refer to any type of graph. This notation allows us to generalize results and discussions without redundancy across different graph types. In the remainder, for every vertex V^{*}\in\mathbb{V}^{*} in a graph \mathcal{G}^{*}=(\mathbb{V}^{*},\mathbb{E}^{*}) , we will refer to its parents in graph by, {Pa}\left(V^{*},\mathcal{G}^{*}\right) , its ancestors by {An}\left(V^{*},\mathcal{G}^{*}\right) , and its descendants by {De}\left(V^{*},\mathcal{G}^{*}\right) . We consider that a vertex counts as its own descendant and as its own ancestor. In addition, the strongly connected component of a vertex is defined as {Scc}\left(V,\mathcal{G}^{*}\right)={An}\left(V,\mathcal{G}^{*}\right)\cap{De}\left(V,\mathcal{G}^{*}\right) . In this section, we present