Zhe Zhang
Assistant Professor of Business Analytics and Technology
Rady School of Management
Current Research
I am an assistant professor at University of California - San Diego's Rady School of Management, within the Innovation, Technology, and Operations group. I completed my Ph.D. at Carnegie Mellon University at the Heinz College School of Information Systems and Management in Pittsburgh, PA, USA. My research agenda is on the societal impact of information technology. This is represented by two particular projects: one on developing new methods for auditing algorithms for fairness and bias in AI, and another on ridesharing's (Uber, Lyft) impact on increasing not only local consumption, but also spatial diversity in consumption.
This research is highlighted by four ongoing projects described below. My research methods range from machine learning methodology to economic theory to applied microeconomics. I incorporate this range of methods since they are useful to answer a range of related questions. Each project is motivated to provide insights into how changes in information technology may be affecting societal and secondary outcomes:
"Imputation Strategies Under Clinical Presence: Impact on Algorithmic Fairness"
Biases have marked medical history, leading to unequal care affecting marginalised groups. The patterns of missingness in observational data often reflect these group discrepancies, but the algorithmic fairness implications of group-specific missingness are not well understood.
Despite its potential impact, imputation is too often a forgotten preprocessing step. At best, practitioners guide imputation choice by optimising overall performance, ignoring how this preprocessing can reinforce inequities. Our work questions this choice by studying how imputation affects downstream algorithmic fairness.
First, we provide a structured view of the relationship between clinical presence mechanisms and group-specific missingness patterns. Then, through simulations and real-world experiments, we demonstrate that the imputation choice influences marginalised group performance and that no imputation strategy consistently reduces disparities. Importantly, our results show that current practices may endanger health equity as similarly performing imputation strategies at the population level can affect marginalised groups in different ways.
"Has Ridehailing Exacerbated Inequalities in Local Spending? Analysis of Ridehailing Usage and Consumption Patterns in 2012-2016" (joint work with Beibei Li)
This work is motivated by the popularity of peer-to-peer transportation platforms, like Uber and Lyft. We are interested in how the adoption of such services have affected the movement and local consumption of consumers. Particularly, we are interested in the heterogeneity of the effect, such as along spatial, demographic, or temporal dimensions. This is of interest to urban planners and policy makers, to better understand how the transportation system might influence local movement and consumption patterns. This work uses a unique longitudinal panel dataset on consumer transactions.
Initial work presented at WISE 2016 (Dublin) and accepted for oral presentation (video here) in Applied Data Science Track at KDD 2017 (9% acceptance rate).
"Growth versus Competition in Growing Digital Markets: Evidence from Amazon Prime's Effects on Competitors" (with Xiaofeng Liu and Kevin Zhu)
Amazon has been a dominant flagship retailer in online markets. When consumers adopt Amazon's Prime service, it increases their Amazon usage substantially. In this paper, we analyze if these consumers increase or decrease their non-Amazon online spending. Surprisingly, we find that even though these Prime adopters increase their Amazon spending, their spending at non-Amazon online retailers also goes up on average.
This is importantly heterogeneous though. We show that novice e-commerce consumers show positive spillovers from their Amazon Prime adoption, perhaps driven Prime increasing their comfort with e-commerce—while those experience consumers have stronger competitive effects, since they were already comfortable with e-commerce.
A major takeaway of our paper is to consider competition in emerging online markets in a more nuanced manner, cognizant of the potential value of digital agglomeration and market expansion.
"Identifying Significant Predictive Bias in Classifiers" (the work is currently ongoing and joint with Daniel Neill)
The above arXiv link is a 5-page introduction to the project, presented, with a travel award, at the NIPS 2016 Workshop on Interpretable Machine Learning in Complex Systems, and later, the KDD 2017 Workshop on Fairness, Accountability, and Transparency in ML (FAT-ML).
This work is motivated by the increasing use of data-driven classifiers and risk assessment models for decision-making in various public and private sectors. Beyond overall performance assessment of these models, it's important that we identify if there are subpopulations or subgroups where such models may be over- or under-estimating the probability. With exponentially many such groups though, this can be a difficult problem. Our method enables such identification, providing both a way to audit the use of such models and improve them.
"Business Models in the Sharing Economy: Manufacturing durable goods in the presence of Peer-to-Peer rental markets" (joint work with Vibhanshu Abhishek and Jose Guajardo)
Published in Information Systems Research (2021):
https://pubsonline.informs.org/doi/10.1287/isre.2021.1034Presented at CIST 2016 (Nashville), WISE 2016 (Dublin), INFORMS 2016 (Nashville), POMS 2017 (Seattle, finalist for best student paper in supply chain mangement), The Marketing Science Conference 2017, and MSOM 2017.
This work is motivated by the rise of peer-to-peer rental markets for a variety of durable goods (e.g., vehicles, boats, bicycles, condominiums). We seek to provide a simple theoretical model from the perspective of manufacturers: (a) how are manufacturers affected by the entry of such peer-to-peer markets, and (b) what are the optimal business models in light of such markets? We model consumer usage as stochastic and find that consumer heterogeneity in such usage rates is the key interesting parameter that guides our results. When heterogeneity in usage rates is intermediate, we identify a new equalizing effect, a novel effect enabled by peer-to-peer markets that benefits the manufacturer. In such cases, a manufacturer may prefer to embrace peer-to-peer markets even over controlling and operating their own rental markets.
Background / Contact
Carnegie Mellon University, PhD, Heinz College School of Information Systems & Management
Jointly advised by Professors Daniel Neill, Beibei Li, and Vibhanshu Abhishek
Stanford University,
BS, Mathematical and Computational Sciences (MCS)
BA, Economics, Stanford University
In my undergraduate, I studied economics and policy, particularly around climate policy and energy markets. After my undergraduate, I spent some time continuing economics research on coal markets and the behavioral impact of pricing structure in electricity, and also spent time as a researcher at the Natural Resources Defense Council and Union of Concerned Scientists nonprofits. I came to Carnegie Mellon to pursue my PhD using multiple disciplines including computer science and economics. I grew up in the Berkeley, CA area.
During my PhD, I have spent time as a part-time Data Creative staff at DataKind in NYC, NY, providing data science project brainstorming for potential partner organizations. I also was a 2016 Fellow at the Data Science for Social Good Summer Fellowship (DSSG) in Chicago, IL. There I worked on an interpretable prediction project for education support services.