PhD student in Statistics @UCBerkeley @UCBStatistics | Causal inference, Graph-based methods, Econometrics | Previously @UWStat
Joined August 2021
- Tweets35
- Following284
- Followers356
- Likes118
Excited to share a new paper with @pengding00 and Peter Bickel on combining estimators. arxiv.org/abs/2602.16310
TL;DR: When combining estimators, how can we do valid inference when the bias is unknown? [1/2]
We propose a strategy to combine estimators from a sensitivity analysis perspective. We construct a sequence of confidence intervals indexed by the magnitude of bias and propose the b-value to quantify the maximum bias so that combining estimators yields an insignificant result
Excited to share a new work with @pengding00 on causal inference in time-series experiments.
Appreciate any comments or feedback
Unifying regression-based and design-based causal inference in time-series experiments. arxiv.org/abs/2510.22864
Zhexiao Lin retweeted
excited to see the published version of our paper on dealing with missing covariates and outcomes in randomized trials: academic.oup.com/biomet/arti… It is a simple paper with some intriguing results. Slides are here: dropbox.com/scl/fi/jrkfnj0qr… @FanLiDuke
Zhexiao Lin retweeted
On Rosenbaum’s Rank-based Matching Estimator. Matias D Cattaneo, Fang Han, Zhexiao Lin. Biometrika. doi.org/10.1093/biomet/asae0…
Zhexiao Lin retweeted
We give the closed-form expression for the asymptotic variance of the Abadie-Imbens matching estimator.
Songliang Chen, Fang Han: On the limiting variance of matching estimators arxiv.org/abs/2411.05758 arxiv.org/pdf/2411.05758
Zhexiao Lin retweeted
We are thrilled to introduce 🍅 TOMATO - a novel visual temporal reasoning benchmark! Building on top of existing benchmarks that primarily focus on reasoning about the order of different events or retrieving static details among frames, TOMATO emphasizes the necessity of reasoning across all frames as a continuous sequence, where individual frames alone are insufficient or misleading for understanding video contents.
📄 Paper: arxiv.org/abs/2410.23266
🤗 Data: huggingface.co/datasets/yale…
💻 Code: github.com/yale-nlp/TOMATO
🧵1/9
Zhexiao Lin retweeted
We prove that the Abadie-Imbens matching estimator is bootstrap consistent—as long as you scale up the number of matches.
On the consistency of bootstrap for matching estimators. arxiv.org/abs/2410.23525
Zhexiao Lin retweeted
This paper (arxiv.org/abs/2312.07683) will appear in 𝗕𝗶𝗼𝗺𝗲𝘁𝗿𝗶𝗸𝗮. It analyzes NN matching when using rank-based distances.
For people that like matching, rank-based statistics, and nonparametric regression with generated covariates🤓 A collaborative work with the awesome Matias and the amazing Zhexiao @zzzxlin 😍
Zhexiao Lin retweeted
Sharing a new working paper with Anqi Zhao & Peng Ding @pengding00, titled "Factorial Difference-in-Differences." arxiv.org/pdf/2407.11937v2 🧵
Comments and suggestions are welcome!
Zhexiao Lin retweeted
excited to see the physical copy of the book; nervous about the potential errors. Comments are welcome.
I just uploaded the R code and datasets to Harvard Dataverse: doi.org/10.7910/DVN/ZX3VEV I plan to provide Python code as well but I need to learn Python first.
Excited to share my first paper in applications, where we applied the graph neural networks to analyze scRNA-seq data. Very happy to be advised by Wei @weisun2013 on this project!
biorxiv.org/content/10.1101/…
Zhexiao Lin retweeted
My lecture notes on permutation processes
arxiv.org/abs/2407.09664
Zhexiao Lin retweeted
This paper is going to appear in a future issue of the Annals of Applied Probability; it gives the limiting distribution of Chatterjee’s correlation when the data are supported over a manifold.
arxiv.org/abs/2209.11156
Zhexiao Lin retweeted
A simpler example is Kendall's tau, for which the U-statistic version is more efficient than the sample-mean version (centralize using the pop mean). However, the same observation does not apply to Pearson's correlation, for which if you know the pop mean, you should use it.
IPW with the estimated propensity score is another example. The first-stage estimation reduces the asymptotic variance, which surprises many people. A recent paper is arxiv.org/pdf/2303.17102 Also, Newey&McFadden chapter 6 is about "two-step estimation" sciencedirect.com/science/ar…
Zhexiao Lin retweeted
IPW with the estimated propensity score is another example. The first-stage estimation reduces the asymptotic variance, which surprises many people. A recent paper is arxiv.org/pdf/2303.17102 Also, Newey&McFadden chapter 6 is about "two-step estimation" sciencedirect.com/science/ar…
Zhexiao Lin retweeted
On the failure of the bootstrap for Chatterjee's rank correlation. Zhexiao Lin, Fang Han. Biometrika. doi.org/10.1093/biomet/asae0…
Zhexiao Lin retweeted
The paper (arxiv.org/abs/2303.14088) now will appear in a future issue of 𝗕𝗶𝗼𝗺𝗲𝘁𝗿𝗶𝗸𝗮. The message is one-line: Chatterjee's rank correlation is root-n consistent, asymptotically normal, but 𝗶𝗿𝗿𝗲𝗴𝘂𝗹𝗮𝗿, and hence bootstrap inconsistent.
Zhexiao Lin retweeted
For people that like matching, rank-based statistics, and nonparametric regression with generated covariates🤓 A collaborative work with the awesome Matias and the amazing Zhexiao @zzzxlin 😍
Matias D. Cattaneo, Fang Han, Zhexiao Lin: On Rosenbaum's Rank-based Matching Estimator arxiv.org/abs/2312.07683 arxiv.org/pdf/2312.07683