Develop causal-graph-based adjustment for causal-forest signals

Develop a causal-graph-based procedure for selecting valid adjustment sets when estimating the causal-forest treatment contrast for each candidate driver, thereby avoiding conditioning on mediators, colliders, or post-treatment variables.

Background

In the MCES implementation, each continuous driver is dichotomized at its median and treated in turn, while the remaining pre-selected drivers are used as controls. Without a stated causal graph, this procedure can condition on variables that bias the estimated treatment contrast.

The paper therefore treats the causal-forest output as one evidence signal rather than an identified average treatment effect. A principled extension would use causal-graph information to determine admissible adjustment sets before estimating and pooling these effects.

References

It is the reason the question taken up below --- a premium of 7 or of 10 percent? --- cannot be closed with these data, and the reason no claim in this section is robust to every structure we considered: under W5 the premium is not identified by any control set in a cross-section.

— A Multiverse of Good and Bad Controls: Candidate Causal Graphs for Interpreting Model Robustness Analysis  (2609.16618 - Okubo, 15 Sep 2026) in Section 'A world not identified by adjustment: the time-indexed mixed mechanism' (within Section 5.3, 'The Union Wage Premium: A Fragility Verdict Reversed')

More importantly, treating each driver in turn as the treatment while using all remaining drivers as controls can, without a stated causal graph, condition on mediators, colliders, or post-treatment variables and thereby bias $\hat{\tau}$; we therefore read $\tilde{e}_{11}$ as one heterogeneity-sensitive evidence signal among many, not as an identified average treatment effect. Supplying a causal graph to choose valid adjustment sets is the principled fix and is left to future work.

— Multi-Method Causal Evidence Synthesis: Ranking Candidate Drivers by Convergent Cross-Method Evidence from Observational Data  (2608.20187 - Gupta et al., 20 Aug 2026) in Section 3.2.11, “Method 11: Causal Forest”

Third, while the Planetary Prediction Engine’s anti-leakage protocols (Feature Gate, Split-Isolated Imputation) provide strong safeguards, formal verification of causal direction filters remains an open challenge, particularly for targets with complex, bidirectional relationships to candidate covariates.

— Planetary Prediction Engine: Autonomous Geospatial Prediction via Intelligent Data Selection and Foundation Model Embeddings  (2608.26088 - Ma et al., 26 Aug 2026) in Section 3.5, “Limitations and Future Work,” p. 13