Semiparametric Theory & Method
M.A. Student · Division of Biostatistics · University of California, Berkeley
Weihan Zhang
Looking for 27Fall PhD positions.
News
- Paper ``Estimation and Inference for Causal Explainability'' is accepted by CLeaR 2026!
About
I am an M.A. student at Division of Biostatistics, University of California, Berkeley. I am broadly interested in the minimum-assumption inference, e.g. semiparametric theory & design-based inference, and distributional & conditional treatment effects in causal inference. I spent my college years studying political science at School of Government, Peking University. I am fortunate to work with Prof. Zijun Gao (USC Marshall Data Sciences and Operations) and Prof. Ye Wang (UNC Chapel Hill Political Science).
Research
Design-based and (Quasi) Randomization Inference
Distributional & Conditional Treatment Effects
Publications & Working Papers
Estimation and Inference for Causal Explainability
Abstract: Understanding how much each variable contributes to an outcome is a central question across disciplines. A causal view of explainability is favorable for its ability in uncovering underlying mechanisms and generalizing to new contexts. Based on a family of causal explainability quantities, we develop methods for their estimation and inference. In particular, we construct a one-step correction estimator using semi-parametric efficiency theory, which explicitly leverages the independence structure of variables to reduce the asymptotic variance. For a null hypothesis on the boundary, i.e., zero explainability, we show its equivalence to Fisher's sharp null, which motivates a randomization-based inference procedure. Finally, we illustrate the empirical efficacy of our approach through simulations as well as an immigration experiment dataset, where we investigate how features and their interactions shape public opinion toward admitting immigrants.
A Simple Randomization Test for Interference
Abstract: We propose a randomization test for partially sharp null hypotheses concerning interference in randomized experiments on geographic space or social networks. Existing tests rely on conditional randomization, which achieves exact finite-sample validity but often suffers from low power, since only a subset of units contributes to both the test statistic and the permutation. We exploit the observation that a partially sharp null implies a family of weak nulls on average spillover effects at each proximity level, each of which can be assessed by applying a randomization test to a pivotal statistic constructed from the full sample. The cornerstone of the test is an estimator of the spillover effect at a given proximity level, which we show is equivalent to a variant of the bivariate Moran's I. We generalize it through a regression representation that enables joint testing across multiple proximity levels. We derive the sampling and permutation variances of these statistics and establish their asymptotic validity. Simulation evidence supports our theoretical results, and a replication exercise demonstrates the method's practical value.
Miscellaneous
- Outside research: No anime, no life. One of my favorite works is Saya no Uta (沙耶の唄).
Contact
Division of Biostatistics, University of California, Berkeley
Berkeley, CA