About me

I am an Assistant Professor in the Statistics Department at UC Berkeley. My research aims at understanding causal relationships using large administrative datasets from the medical and social sciences. I develop new ways to form compelling matched or weighted comparison groups in these datasets using tools from optimization. I also study methods for transparent and interpretable inference about causal effects when unobserved confounding variables may be present. I am broadly interested in substantive problems relating to people and institutions, and my methodological work is motivated by collaborations to address such problems in health services research, epidemiology, and education.

Picture

Software

JOINT VARIABLE IMPORTANCE PLOTS

R package jointVIP implementing joint treatment-outcome variable importance plots for diagnostic use in designing observational studies, joint work with Lauren Liao. Read in detail about the construction and use of these plots in our paper with Yeyi Zhu, Amanda Ngo, and Rana Chehab, or try out our Shiny app!

Full downloads (source code and Mac and Windows binaries) are available at CRAN.

SPARSE OPTIMAL MATCHING

R packages rcbalance and rcbsubset for large, sparse optimal matching of treated units to control units in observational studies. Includes support for near-fine balance and refined covariate balance constraints. The underlying network algorithm and its capabilities are discussed in greater detail in the JASA paper with Rachel Kelz, Jeffrey Silber, and Paul Rosenbaum, and the rcbalance package and its use are detailed in the Observational Studies paper.

Downloads (source code and Mac and Windows binaries) are available at the CRAN pages for rcbalance and rcbsubset.

MATCHING AND INFERENCE UNDER ROLLING ENROLLMENT

R package GroupMatch for matching and inference in settings with repeated measures and rolling enrollment (different treated units entering treatment at different times). The package implements optimal matching routines described in our paper with Lauren Forrow, Jonathan Gellar, and Jiaqi Li and a bootstrap procedure and diagnostic test described in our paper with Amanda Glazer.

Source code is available at GitHub.

MULTILEVEL MATCHING

R package matchMulti for optimal multilevel matching designs, joint work with Luke Keele. The algorithm is described in our paper with Lindsay Page and Matt Lenard.

Downloads (source code and Mac and Windows binaries) are available at CRAN.

MULTIOBJECTIVE MATCHING

R package MultiObjMatch for exploring optimal tradeoffs in matched designs with multiple design objectives, joint work with Shichao Han. The algorithm is described in my paper with Rachel Kelz.

Downloads (source code and Mac and Windows binaries) are available at CRAN.

GENERALIZED COMPUTER-AIDED CLUSTERING FOR SOCIAL NETWORKS

Functions written in R for comparing and evaluating candidate clusterings of social network data. The software is described in detail in the paper in the Journal of Social Structure.

R source code.

GENETIC RECORDS FAMILY TREE (GRFT) BROWSER

Web applet in JavaScript for displaying and exploring records of genetic crosses, joint work with John Fernandes. The tool and its use are described in greater detail in our paper with Virginia Walbot. An online example of the tool is available here.

Download. Zip file contains a JavaScript source file, a CSS style sheet, an html file for the host webpage, and a tab-delimited text file with sample genetic records.