Causal & Statistical Inference
Causal Inference
Experimentation
Off-Policy Evaluation
Decision Science
Decision systems for incomplete evidence: off-policy evaluation, experimentation, incrementality, support diagnostics, uncertainty, and launch-readiness.
Focus
My primary research agenda is causal and statistical inference for decisions that cannot rely on clean, complete evidence. This includes offline policy evaluation, experiment design, incrementality measurement, support diagnostics, uncertainty, and guardrails for deciding what is ready to launch, test online, reject, or leave unresolved.
Core Themes
- Support-aware off-policy evaluation: replay estimands, propensity support, conservative lower bounds, subgroup non-harm checks, and out-of-time validation.
- Experiment design under interference: choosing robust designs when treatment effects may propagate through shared budgets, producer exposure, graph spillovers, inventory constraints, or temporal carryover.
- Privacy-robust incrementality: stress-testing lift claims under match loss, attribution loss, aggregation thresholds, reporting noise, and segment suppression.
- Decision-calibrated uncertainty: conformal intervals and uncertainty summaries calibrated to downstream operational decisions rather than generic prediction residuals.
- Launch-readiness framing: separating statistical signal from deployable evidence through decision rules, guardrails, diagnostics, and stakeholder-ready narratives.
Representative Work
- Decision support for marketplace policies under incomplete evidence.
- Support-aware offline policy selection for advertising marketplaces.
- Choosing online experiment designs under interference in ads, recommendations, and member-experience systems.
- Privacy-robust incrementality measurement for advertising systems under signal loss.
- Decision-calibrated conformal uncertainty for pacing decisions in streaming advertising.