Abstract:Discovering causal relationships from observational data is a challenging task that relies on assumptions connecting statistical quantities to graphical or algebraic causal models. In this work, we focus on widely employed assumptions for causal discovery when objects of interest are (multivariate) groups of random variables rather than individual (univariate) random variables, as is the case in a variety of problems in scientific domains such as climate science or neuroscience. If the group-level causal models are derived from partitioning a micro-level model into groups, we explore the relationship between micro and group-level causal discovery assumptions. We investigate the conditions under which assumptions like Causal Faithfulness hold or fail to hold. Our analysis encompasses graphical causal models that contain cycles and bidirected edges. We also discuss grouped time series causal graphs and variants thereof as special cases of our general theoretical framework. Thereby, we aim to provide researchers with a solid theoretical foundation for the development and application of causal discovery methods for variable groups.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the challenges faced in causal discovery in multivariate groups (i.e., groups composed of multiple random variables). Specifically, the paper focuses on how to ensure that the causal assumptions at the group level remain valid after grouping the micro - level causal models. The paper explores the following points: 1. **Relationship between micro - and macro - level causal assumptions**: When transitioning from micro - level causal models (such as causal relationships between individual random variables) to macro - level causal models (such as causal relationships between multivariate groups), the paper studies whether micro - level causal assumptions (such as the causal faithfulness assumption) can be directly transferred to the macro - level. 2. **Validity of the causal faithfulness assumption**: The paper specifically analyzes the applicable conditions of the causal faithfulness assumption in multivariate groups. The causal faithfulness assumption is an important assumption in causal inference, which requires that the conditional independence in the observed data can reflect the graphical separation in the causal graph. However, the paper points out that this assumption may not hold in multivariate groups and proposes the concept of non - local faithfulness violation, that is, even if the weaker faithfulness assumptions (such as adjacency faithfulness and orientation faithfulness) are satisfied at the macro - level, the causal faithfulness may still be violated. 3. **Conditions for ensuring causal faithfulness**: To overcome the above problems, the paper provides two criteria that can ensure that the causal faithfulness assumption still holds at the macro - level after simplifying the micro - graph to the macro - graph. These criteria are related to the connectivity within the variable groups, for example, the connectivity achieved through loops or directed / bidirectional paths. 4. **Application of time - series causal graphs**: The paper also discusses how to extend time - series causal graphs to time - series causal graphs of multivariate groups and proposes a condition called "causal mixing" that can lead to stronger causal conclusions at the group level of time - series. In summary, this paper aims to provide researchers with a solid theoretical foundation so that they can develop and apply more effective causal discovery methods when dealing with the causal discovery problems of multivariate groups. This is not only of great significance for research in fields such as climate science and neuroscience, but also provides new ideas for complex system modeling in fields such as economics.

Foundations of Causal Discovery on Groups of Variables

An Introduction to Causal Discovery

Causal Discovery over High-Dimensional Structured Hypothesis Spaces with Causal Graph Partitioning

Causal Discovery under Latent Class Confounding

Reinterpreting causal discovery as the task of predicting unobserved joint statistics

A Survey on Causal Discovery: Theory and Practice

Causal discovery and inference: concepts and recent methodological advances

Directed Cyclic Graph for Causal Discovery from Multivariate Functional Data

Confidence in Causal Discovery with Linear Causal Models

Causal discovery with endogenous context variables

Causal Discovery via Causal Star Graphs

Amortized Causal Discovery: Learning to Infer Causal Graphs from Time-Series Data

Deep Learning-based Group Causal Inference in Multivariate Time-series

Causal Discovery from Multiple Data Sets with Non-Identical Variable Sets

Review of Causal Discovery Methods Based on Graphical Models

Causal Discovery of Linear Non-Gaussian Causal Models with Unobserved Confounding

Causal Discovery from Temporally Aggregated Time Series

Structural restrictions in local causal discovery: identifying direct causes of a target variable

High-dimensional causal discovery based on heuristic causal partitioning

Survey and Evaluation of Causal Discovery Methods for Time Series

Partial Homoscedasticity in Causal Discovery with Linear Models