Prediction Error Can Mislead Causal Inference
π‘Learn why the best nuisance-function predictor may not produce the best causal estimate.
β‘ 30-Second TL;DR
What Changed
Prediction error did not consistently correlate with causal bias across methods and simulation settings.
Why It Matters
The findings caution practitioners against selecting nuisance models solely by predictive accuracy when the downstream goal is causal estimation. Evaluation should separately assess bias, RMSE, and uncertainty coverage rather than treating nuisance-function prediction error as a proxy for causal performance.
What To Do Next
Benchmark your causal pipeline with bias, RMSE, and confidence-interval coverage in addition to nuisance-model cross-validation error.
Key Points
- β’Prediction error did not consistently correlate with causal bias across methods and simulation settings.
- β’XGBoost achieved the lowest RMSE among non-oracle approaches, while DML-XGBoost generally had stronger 95% confidence-interval coverage.
- β’A joint-error measure based on the absolute cross-product of exposure and outcome nuisance-function errors was only weakly related to causal bias.
Weekly AI Recap
Read this week's curated digest of top AI events β
πRelated Updates
AI-curated news aggregator. All content rights belong to original publishers.
Original source: ArXiv AI β
This is a summary, not the original. Read the source, or get the weekly briefing.
Weekly AI briefing
One email a week. Unsubscribe anytime.