Accepted papers

Authors: Philip Boeken, Joris Mooij

Title: Dynamic Structural Causal Models

Abstract: We study a specific type of SCM, called a Dynamic Structural Causal Model (DSCM), whose endogenous variables represent functions of time, which is possibly cyclic and allows for latent confounding. As a motivating use-case, we show that certain systems of Stochastic Differential Equations (SDEs) can be appropriately represented with DSCMs. An immediate consequence of this construction is a graphical Markov property for systems of SDEs. We define a time-splitting operation, allowing us to analyse the concept of local independence (a notion of continuous-time Granger (non-)causality). We also define a subsampling operation, which returns a discrete-time DSCM, and which can be used for mathematical analysis of subsampled time-series. We give suggestions how DSCMs can be used for identification of the causal effect of time-dependent interventions, and how existing constraint-based causal discovery algorithms can be applied to time-series data.

Link: https://arxiv.org/abs/2406.01161

Authors: Georg Manten, Cecilia Casolo, Emilio Ferrucci, Søren Wengel Mogensen, Cristopher Salvi, Niki Kilbertus

Title: Signature Kernel Conditional Independence Tests in Causal Discovery for Stochastic Processes

Abstract: Inferring the causal structure underlying stochastic dynamical systems from observational data holds great promise in domains ranging from science and health to finance. Such processes can often be accurately modeled via stochastic differential equations (SDEs), which naturally imply causal relationships via 'which variables enter the differential of which other variables'. In this paper, we develop a kernel-based test of conditional independence (CI) on `path-space'---solutions to SDEs---by leveraging recent advances in signature kernels. We demonstrate strictly superior performance of our proposed CI test compared to existing approaches on path-space. Then, we develop constraint-based causal discovery algorithms for acyclic stochastic dynamical systems (allowing for loops) that leverage temporal information to recover the entire directed graph. Assuming faithfulness and a CI oracle, our algorithm is sound and complete. We empirically verify that our developed CI test in conjunction with the causal discovery algorithm reliably outperforms baselines across a range of settings.

Authors: Carles Balsells-Rodas, Yixin Wang, Pedro A. M. Mediano, Yingzhen Li

Title: Identifying Nonstationary Causal Structures with High-Order Markov Switching Models

Abstract: Causal discovery in time series is a rapidly evolving field with a wide variety of applications in other areas such as climate science and neuroscience. Traditional approaches assume a stationary causal graph, which can be adapted to nonstationary time series with time-dependent effects or heterogeneous noise. In this work we address nonstationarity via regime-dependent causal structures. We first establish identifiability for high-order Markov Switching Models, which provide the foundations for identifiable regime-dependent causal discovery. Our empirical studies demonstrate the scalability of our proposed approach for high-order regime-dependent structure estimation, and we illustrate its applicability on brain activity data.

Link: https://arxiv.org/abs/2406.17698

Authors: Simon Ferreira, Charles K. Assaad

Title: Identifying macro conditional independencies and macro total effects in summary causal graphs with latent confounding

Abstract: Understanding causal relationships in dynamic systems is essential for numerous scientific fields, including epidemiology, economics, and biology. While causal inference methods have been extensively studied, they often rely on fully specified causal graphs, which may not always be available or practical in complex dynamic systems. Partially specified causal graphs, such as summary causal graphs (SCGs), provide a simplified representation of causal relationships, omitting temporal information and focusing on high-level causal structures. In this paper, we address the properties of SCGs. Firstly, we show how SCGs differ from other known partially specified graphs. Then, we demonstrate the soundness and completeness of the d-separation to identify macro conditional independencies in SCGs. Furthermore, we establish that the do-calculus is sound and complete for identifying macro total effects in SCGs. Conversely, we also show through various examples that these results do not hold when considering micro conditional independencies and micro total effects. Through this work, we aim to facilitate the analysis of causal relationships in dynamic systems, even when a fully specified causal graph is unavailable.

Link: https://arxiv.org/abs/2407.07934

Authors: Fernanda Almeida, Vânia Guimarães, Vitor Rolla, Luis Pinto-Coelho

Title: Combining Time Series Causal Discovery and Regression for Personalized Heart Rate Prediction in Cardiac Patients

Abstract: Heart rate (HR), a critical cardiovascular health indicator, is commonly monitored in clinical settings, such as intensive care units (ICUs), to detect deterioration and guide treatment decisions. Recent advancements in artificial intelligence and extensive medical datasets have enabled the development of HR forecasting models to anticipate higher cardiovascular and mortality risks. This study investigates whether combining causal discovery with simple machine learning (ML) regression techniques improves personalized HR forecasting in ICU patients compared to using regression alone. A constraint-based causal discovery method operating on multivariate time series was used to identify causal predictors for the regression models. Our study demonstrates that incorporating causal predictors improves HR forecasting performance, surpassing the performance achieved by using regression alone. The best performance was achieved when forecasting a 3-minute horizon, yielding an error of 1.02 ± 0.68 bpm. Forecasting errors increased for longer prediction horizons, with 5.90±3.43 bpm errors at a 60-minute forecast. These preliminary results suggest that causal predictors enhance HR forecasting within actionable timeframes, potentially aiding clinical decisions. The title has slightly changed as requested by the reviewers. As I explained previously, one author was missing.

Link: https://drive.google.com/file/d/1wzWJqDQdO29HxTgCzoUO9DyqnhfOdAUP/view

Authors: Sebastian Hickman, Paul T Griffiths, Peer Nowack

Title: Estimating the ozone-climate penalty with causal inference

Abstract: Ozone air pollution is a considerable threat to human health, contributing to hundreds of thousands of premature deaths annually. Surface ozone concentrations are strongly associated with temperature as a result of a number of interacting physical and chemical processes. While it is known that many factors contribute to the observed ozone-temperature relationship, quantifying their contributions remains a challenge due to the complex and often non-linear physical interactions between variables. The lifetime of ozone, which ranges from hours to weeks in the troposphere, means that understand the drivers of local concentrations is complex. In this preliminary work, we apply double machine learning methods to large observational datasets to provide constraints on the ozone-temperature sensitivity using data from across Europe. Our estimates are compared with chemical sensitivities estimated with a chemical box model simulating conditions observed at a rural station in the United Kingdom. We discuss opportunities to extend our methods to better include the temporal nature of the causal effect of temperature on ozone.

Authors: Qiang Huang, Defu Cao, Biwei Huang, Yi Chang, Yan Liu

Title: Addressing Post-treatment Bias for Reliable Treatment Effect Estimation

Abstract: In recent years, estimating causal effects based on observational data has garnered significant attention, playing a crucial role in decision-making across various fields such as healthcare, economics, and social policy. Most related work assumes that the observed covariates are collected before treatment application. However, in many scenarios, especially those involving time-series data, some covariates are collected after the treatment is applied, known as post-treatment variables. Simply treating them as pre-treatment variables can introduce estimation bias. In this work, we discuss how ignoring or incorrectly handling post-treatment variables can harm causal effect estimation and analyze the reasons behind this bias. We then propose a variable decomposition-based method to model and separate post-treatment variables from observed covariates. Experiments on synthetic data and a real-world dataset demonstrate the effectiveness of our method in eliminating bias caused by post-treatment variables.

Authors: Christopher Lohse, Jonas Wahl

Title: Sortability of Time Series Data

Abstract: Evaluating the performance of causal discovery algorithms that aim to find causal relationships between time-dependent processes remains a challenging topic. In this paper, we show that certain characteristics of datasets, such as varsortability (Reisach et al. 2021) and -sortability (Reisach et al. 2023), also occur in datasets for autocorrelated stationary time series. We illustrate this empirically using four types of data: simulated data based on SVAR models and Erdős-Rényi graphs, the data used in the 2019 causality-for-climate challenge(Runge et al. 2019), real-world river stream datasets, and real-world data generated by the Causal Chamber of Gamella et al. 2024. To do this, we adapt var- and -sortability to time series data. We also investigate the extent to which the performance of score-based causal discovery methods goes hand in hand with high sortability. Arguably, our most surprising finding is that the investigated real-world datasets exhibit high varsortability and low -sortability indicating that scales may carry a significant amount of causal information.

Link: https://drive.google.com/file/d/1lvvBKGm4aBZe7WpdrIrr456Wl01sr1rp/view?usp=drive_link

Authors: Alexandre Trilla, Rajesh Rajendran, Ossee Yiboe, Quentin Possamaï, Nenad Mijatovic, Jordi Vitria

Title: Industrial-Grade Time-Dependent Counterfactual Root Cause Analysis through the Unanticipated Point of Incipient Failure: a Proof of Concept

Abstract: This paper describes the development of a counterfactual Root Cause Analysis diagnosis approach for an industrial multivariate time series environment. It drives the attention toward the Point of Incipient Failure, which is the moment in time when the anomalous behavior is first observed, and where the root cause is assumed to be found before the issue propagates. The paper presents the elementary but essential concepts of the solution and illustrates them experimentally on a simulated setting. Finally, it discusses avenues of improvement for the maturity of the causal technology to meet the robustness challenges of increasingly complex environments in the industry.

Link: https://drive.google.com/file/d/1RTiqv-jWavOuiLOo1xyVitkteevY1dlt/view?usp=drive_link

Authors: Naftali Weinberger

Title: Equilibrium, Dynamics, and Cointegration

Abstract: Graphical causal models provide a systematic framework for determining when, in principle, one can identify a causal quantity from a joint probability distribution, given one’s causal assumptions. Nevertheless, it remains unclear how to generalize this framework to cases involving non-stationary measures of probabilistic independence, such as first-order cointegration. This paper argues that Iwasaki and Simon’s [1994] dynamic causal models allow one to generalize standard causal models in way that conceptually links them to time-series methods. Specifically, we highlight how the features of such models that enable one to disentangle groups of variables influencing one another at different time scales correspond to the lag structure by which first-order cointegration enables one to distinguish longer- and shorter-run relations.