Causal Relational Learning

Abstract

Causal inference is at the heart of empirical research in natural and social sciences and is critical for scientific discovery and informed decision making. The gold standard in causal inference is performing randomized controlled trials; unfortunately these are not always feasible due to ethical, legal, or cost constraints. As an alternative, methodologies for causal inference from observational data have been developed in statistical studies and social sciences. However, existing methods critically rely on restrictive assumptions such as the study population consisting of homogeneous elements that can be represented in a single flat table, where each row is referred to as a unit. In contrast, in many real-world settings, the study domain naturally consists of heterogeneous elements with complex relational structure, where the data is naturally represented in multiple related tables. In this paper, we present a formal framework for causal inference from such relational data. We propose a declarative language called CaRL for capturing causal background knowledge and assumptions, and specifying causal queries using simple Datalog-like rules. CaRL provides a foundation for inferring causality and reasoning about the effect of complex interventions in relational domains. We present an extensive experimental evaluation on real relational data to illustrate the applicability of CaRL in social sciences and healthcare.

Tisane: Authoring Statistical Models via Formal Reasoning from Conceptual and Data Relationships

“…Recent work in the database community helps researchers answer causal questions about multilevel, or hierarchical, data [28,51]. CaRL [51] provides a domain-specific language to express causal relationships between variables and a GUI to show researchers results.…”

“…Recent work in the database community helps researchers answer causal questions about multilevel, or hierarchical, data [28,51]. CaRL [51] provides a domain-specific language to express causal relationships between variables and a GUI to show researchers results. Like CaRL, Tisane leverages the insight that researchers have domain knowledge that a system can use to infer statistical methods.…”

Managing bias and unfairness in data for decision support: a survey of machine learning and data engineering approaches to identify and mitigate bias and unfairness within data management and analytics systems

“…While most works assume all data to be in a single table, Salimi et al [160] also adapt previous works around causality to the context of relational databases since the prior formalisation cannot directly apply there. They propose a declarative language-CaRL: Causal Relational Language-that allows them to represent their relational data into a causal paradigm and specify the potential causal dependencies between attributes.…”

Heterogeneous Treatment Effects in Social Networks

“…We then propose a novel hypothesis testing approach based on the standard I 2 test [29] that accounts for neighbors' covariates and patterns in the social network. Since the number of neighbors of each unit in the network may be different, we use the concept of covariate summary proposed in [54]. Given a covariate of a unit in the social network, we aggregate this covariate across all neighbors into a single summarized number (e.g., the percentage of neighbors who work as Farm Labour).…”

“…A typical assumption made in the causality literature is no-interference or independence among units, meaning that the assigned treatment of one unit should not affect the outcome of another unit (called interference or spill-over effects) [41,54] 2 . However, in many studies in social science, the units may directly or indirectly interact with each other in different ways and as a result, a unit's outcome might be influenced by the behavior, states, or characteristics of its social ties in a network, which leads to violation of the no-interference assumption [65].…”