Causal effect identification refined with partial causal information
Causal Effect Identifiability in Causal Abstractions: A Strict Hierarchy of Criteria
Artificial Intelligence
Summary
Figuring out cause and effect from just data usually needs a detailed map of how things influence each other, which is hard to get. The authors study how to identify cause and effect when only partial information about this map is available. They define different ways to tell when cause-effect relationships can be identified across many possible maps and show these ways form a clear ranking. They also find that sometimes an effect can be identified overall even if no single explanation fits all possible maps. This clarifies confusion in previous research.
What this means in practice
- •For data science teams: Use abstract causal knowledge to determine which cause-effect relations can be identified without full causal graphs in high-dimensional data.
- •For business analysts: Assess the reliability of causal conclusions when only partial structural knowledge of systems is available, improving decision-making.
A theory result. No direct application yet.
Authors
Clément Yvernes, Marianne Clausel, Emilie Devijver, Eric Gaussier
Abstract
Identifying causal effects from observational data typically requires a fully specified causal graph, an assumption that is often unrealistic in complex or high-dimensional settings. Causal abstractions address this limitation by retaining only partial causal information, thereby defining equivalence classes of compatible causal graphs. In this paper, we study causal effect identifiability directly at the level of these graph collections, rather than at the level of individual causal diagrams. We formalize several notions of identifiability in this setting and organize them into a hierarchy that clarifies their relationships and differences. In particular, we establish that a causal effect may be identified over a collection of graphs without admitting a single causal derivation that holds uniformly across all compatible graphs. These results help reveal gaps and clarify ambiguities in the existing literature.