Research map / Probabilistic and scientific modeling / Causal discovery

Identifying Macro Conditional Independencies and Macro Total Effects in Summary Causal Graphs with Latent Confounding

AAAI 2025 · Probabilistic and scientific modeling › Causal discovery · approach: Causal effect identification · 0 citations (OpenAlex)

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Shows d-separation is sound and complete for macro conditional independencies in summary causal graphs (SCGs), and that do-calculus is sound and complete for macro total effects, with an SC-Hedge condition for non-identifiability.

Problem

Causal queries in summary causal graphs (SCGs), which are cyclic cluster-level graphs of time series with latent confounding, lack established criteria: which macro conditional independencies and macro total effects can be identified from the graph?

Approach

Shows that SCG cycles and cluster-level abstraction do not break the standard tools: d-separation (via walks and compatible full-time ADMGs) gives sound and complete macro conditional independence criteria, and the do-calculus rules carry over to SCGs via mutilated graphs; adds an SC-projection/SC-Hedge to characterize non-identifiability.

Key results

• Proves soundness and completeness of d-separation for macro conditional independencies in SCGs (via compatible FT-ADMGs).
• Proves soundness and completeness of the three do-calculus rules for macro total effects in SCGs.
• Gives a graphical SC-Hedge condition that implies non-identifiability of macro total effects, though not necessarily a complete characterization.

Surprising findings

• Temporal information within clusters does not improve identifiability of macro queries; macro total effects are always identifiable without hidden confounding, unlike micro queries.

Limitations

• The SC-Hedge characterization of non-identifiability is not necessarily complete.
• Results require full time series from the start of the generative process; otherwise past values act as unobserved confounders.
• Results do not extend to micro queries; do-calculus is not complete for micro queries.

Experimental setup

Builds on: d-separation do-calculus ADMG cluster-ADMG front-door criterion

Closest papers by meaning

• Identifying Macro Causal Effects in C-DMGs over DMGs (NeurIPS 2025 · similarity 0.92)
• Identifiability of Direct Effects from Summary Causal Graphs (AAAI 2024 · similarity 0.88)
• Local Identifying Causal Relations in the Presence of Latent Variables (ICML 2025 · similarity 0.88)
• Coarsening Linear Non-Gaussian Causal Models with Cycles (NeurIPS 2026 · similarity 0.88)
• Discovering Mixtures of Structural Causal Models from Time Series Data (ICML 2024 · similarity 0.88)
• CausalRivers - Scaling up benchmarking of causal discovery for real-world time-series (ICLR 2025 · similarity 0.88)