Reproduce “Coarsening Bias from Variable Discretization in Causal Functionals” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Xiaxian Ou, Razieh Nabi
Researchers from Emory University quantify and correct the coarsening bias that arises when continuous variables are discretized in causal effect estimations, demonstrating a method that reduces this approximation error from first-order to second-order. Their approach yields more accurate causal effect estimates and improves confidence interval coverage in both simulations and a real-world application.
Key insights
- Naive discretization of continuous variables introduces a population-level approximation bias that scales linearly with bin width (O(1/K)), persisting regardless of sample size.
- The proposed debiased functional effectively eliminates the leading first-order bias term, reducing the approximation error to a second-order (O(1/K^2)) dependence on bin width.
- Correcting the causal functional itself for coarsening bias is crucial; sophisticated statistical estimators (like one-step or TMLE) applied to an *already biased* functional will not remove this approximation error.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.