Hi there!
Yifan Liu
Incoming Postdoctoral Fellow · Harvard University
I am an incoming Postdoctoral Fellow at the Climate and Sustainability Impact Lab, Harvard Business School. My research focuses on energy and environmental policy, with a particular interest in transportation systems and emerging technologies. I integrate public policy and management theories with computational methods, including machine learning and causal inference, applied to novel datasets to study pressing policy questions. My current work examines how market competition and policy design can incentivize improvements in energy infrastructure quality and public service provision. I received my PhD in Public Policy and MS in Computer Science from Georgia Tech, and my MPA in Environmental Policy from Cornell University.
Research
Working papers
-
Why is electric vehicle (EV) charging so unreliable? The effects of competition on supplementary service quality Job market paperSupplementary services — such as coffee shops providing free wi-fi and airlines offering airport lounges — are increasingly important to operational performance and customer experience. Yet little is known about what drives variation in their quality. While prior research shows that competition improves core service quality, we theorize that its effects on supplementary service quality depend on the source of competition. Competition from establishments offering the same core service enhances supplementary service quality, whereas competition from those offering different core services diminishes it. We test our hypotheses using data on the reliability of U.S. public EV charging stations, a widely-offered supplementary service across hotels, retailers, parking facilities, and other establishments. Analyzing 49,902 charging stations with 805,639 consumer reviews from 2011 to 2024, we apply machine learning methods to measure charging reliability. We find that competition among establishments offering the same core service is associated with higher reliability, suggesting that the positive effects of core service competition spill over to the quality of supplementary services. In contrast, competition from charging establishments with different core services is associated with lower station reliability. This latter effect is more consequential, as most nearby charging competitors provide different core services. Our results indicate that, within our sample period, a one-standard-deviation increase in such competition is associated with a 10.4% to 32.2% decline in service reliability scores over the subsequent five years. These findings have important implications for firms' demand management and for public policy aimed at improving EV charging infrastructure performance.
-
Promoting sustainable travel decisions through health and active lifestyle messagingU.S. travelers heavily rely on personal vehicles for daily travel, contributing to an estimated $260 billion annually in social costs from tailpipe emissions. Most people seldom realize how much air pollution they are exposed to during their daily travel or how switching modes could benefit their health. Here we examine whether messages about health and active lifestyle benefits delivered through mobile apps can encourage travelers to walk, bike, or take the bus, overcoming present bias. Using randomized A/B testing in a nationally representative sample of 4,840 U.S. car owners, we test whether mobile app-based messages about the health and active lifestyle impacts of their daily travel decisions can shift preferences towards sustainable modes. We find that travelers are more likely to choose sustainable modes over personal vehicles when provided with credible, health-focused messages at the time of decision-making. These strategies were particularly effective for urban residents and individuals with pre-existing health conditions, but they backfired for long-distance commuters, underscoring a key tension between travel time and perceived health risks. When paired with existing mobility apps, such behavioral interventions can potentially reduce 1 billion car miles at a cost of 2–4 cents per trip, offering a complementary alternative to existing approaches such as reduced fares, free fares, or direct cash.
-
Ranking charger reliability at citiesA growing community of researchers has been expanding the role of machine learning in behavioral strategies for mitigating and responding to climate change. In the areas of transportation and mobility, one of the largest emitting sectors, unreliable charging infrastructure remains a persistent behavioral barrier to electrification. Consumer reports and news coverage have highlighted widespread reliability concerns at many electric vehicle (EV) charging stations worldwide. However, there is often a mismatch between consumers' actual charging experiences and uptime metrics tracked by station managers, reflecting fundamental data and measurement challenges to assess operational performance. To automate the detection of user-oriented station reliability from citizen-generated data, we propose a context-aware machine learning pipeline that combines few-shot learning and counterfactual reasoning to classify reliability failures, while preserving the auditability of expert knowledge. Our open-source implementation supports spatial analysis of reliability performance that can be scaled and adapted across markets at low cost. To support long-range planning, we develop a ranking system for charging reliability across U.S. metropolitan and micropolitan areas. Using this approach, we reveal higher disparities in charger reliability experienced by consumers along federally-designated highway corridors and the most populous cities where reliable charging is most critical.
-
Do non-targeted incentives shape energy infrastructure? Evidence from Opportunity ZonesTargeted clean energy subsidies, which restrict eligibility to specific technologies or sectors, are often vulnerable to political disruption. This challenge is especially salient as federal policy authority contracts and state-level variation in clean energy policy grows in importance. These dynamics raise the question of whether non-targeted incentives can sustain infrastructure deployment under policy uncertainty. This study addresses that question by examining the Opportunity Zone (OZ) program as a key case of a non-targeted investment incentive without technology or sector restrictions. Using a matched sample of designated and eligible-but-not-designated census tracts, we employ a difference-in-differences event study design that exploits the 2018 OZ designation to compare clean energy infrastructure deployment across the two groups from 2014 to 2025. We find that OZ-designated tracts have 35.0% more EV charging stations than matched control tracts seven years after designation. The effect is larger in areas without active state-level EV charging incentives, consistent with substitution rather than complementarity between targeted and non-targeted policy instruments. We also find evidence of spatial displacement, with designated tracts gaining energy infrastructure at the expense of adjacent low-income, non-designated communities.
-
Kinetic externalities of electric vehiclesAbstract coming soon.
-
Engagement effects of gamified health and environmental metrics on travel mode choiceHigh reliance on personal vehicles in the US imposes economic, environmental, and public health costs, yet adoption of sustainable alternatives such as public transit, walking, or biking remains limited due to deeply ingrained travel habits. Drawing on customer engagement theory, this study examines whether gamification can strengthen travelers' behavioral, affective, and cognitive engagement with sustainable modes when integrated with salient health and environmental benefits. Results show that gamification increases the likelihood of choosing sustainable modes in the stated choice scenarios, indicating stronger behavioral engagement. In particular, emissions-based badges and leaderboards increased the likelihood of selecting bus transit by more than 18 and 13 times, respectively. Health-based metrics increased the likelihood of walking, with calories-burned badges and leaderboards raising it by 5 and 8 times, respectively. Overall, the findings suggest that gamification paired with personalized health and environmental benefits may strengthen travelers' engagement with sustainable modes.
Publications
-
Reliability of electric vehicle charging infrastructure: A cross-lingual deep learning approachThis paper develops a cross-lingual deep learning framework to analyze user-generated reviews of EV charging stations across multiple languages. Applied to a large dataset of charger reviews, the model accurately classifies reliability failures and reveals systemic patterns in charging infrastructure quality, with implications for policy and investment in EV infrastructure.
-
Machine learning discovery of regional and social disparities in electric vehicle charging reliability with GPT-5There is growing interest in studying charger reliability to address persistent barriers to electric vehicle (EV) adoption and advance the decarbonization of transportation, one of the largest emitting sectors globally. This study introduces a machine learning pipeline that detects spatial disparities in charger reliability based on 838,785 U.S. consumer reviews of their experiences. We document new performance benchmarks in reliability detection using zero and few shot learning capabilities and expert counterfactual reasoning (F1 score: 0.97, SD: 0.02), outperforming previous models in the domain of electric mobility, such as ClimateBERT. Using this approach, we find evidence of widespread charging reliability issues in about half of all U.S. counties (1,653 of 3,244 counties), especially in the most populated areas. Disparities in charger reliability are most pronounced in metropolitan areas and along federally-designated EV corridors, raising concerns about inconsistent user experiences in high-traffic zones.
-
Effectiveness of health and environmental information provision in promoting sustainable travel modesCommunity-wide adoption of sustainable travel modes such as transit, walking, and biking can alleviate congestion and emissions while improving air quality and public health. However, promoting these modes in the U.S. is challenging due to the high reliance on personal vehicles, which contribute $260 billion annually in social costs. This study explores whether mobile app-based information provision about the health and environmental benefits of sustainable modes can meaningfully change traveler preferences. In a sample of 3,470 U.S. car users aged 55 years and under, balanced by gender, income, and census regions, this study tested the effectiveness of information provision over a 90-day summer season, targeting bus transit, walking, and biking. Results show that participants who received information on environmental benefits related to emission reductions were four times more likely to choose bus transit, while those informed about active health benefits related to calories burned were nearly seven times more likely to choose walking, compared to the control group. However, due to barriers such as safety concerns and lack of infrastructure, health and environmental information was not effective at promoting biking.
-
"Good" XAI design: For what? In which ways?The rapid proliferation of Explainable Artificial Intelligence (XAI) has led to the development of numerous evaluation frameworks aimed at guiding and optimizing its design. However, these frameworks often emphasize the technical properties of XAI artifacts, overlooking the nuanced perceptions and values of end-users. Recognizing that XAI impacts society and individuals in non-neutral ways, this study adopts a human-centered approach to systematically examine the effects of recommended XAI properties on the general public in everyday scenarios through a formative study involving 87 end-users. The findings reveal that comprehensibility is the most valued XAI property, while frequently advocated properties like contrastivity may have overall negative effects. These results highlight the necessity of a goal-driven reverse engineering approach that integrates human values into XAI design to ensure positive user outcomes.
-
Explainable AI for daily scenarios from end-users' perspective: Non-use, concerns, and ideal designCentering humans in explainable artificial intelligence (XAI) research has primarily focused on AI model development and high-stake scenarios. However, as AI becomes increasingly integrated into everyday applications in often opaque ways, the need for explainability tailored to end-users has grown more urgent. To address this gap, we explore end-users' perspectives on embedding XAI into daily AI application scenarios. Our findings reveal that XAI is not naturally accepted by end-users in their daily lives. When users seek explanations, they envision XAI design that promotes contextualized understanding, empowers adoption and adaption to AI systems, and considers multistakeholders' values.
-
Generative AI and electric vehicle service operations in urban and remote areasFor the first time in nearly three decades, the transportation sector is now the largest source of U.S. greenhouse gas emissions. To accelerate climate action, governments are promoting zero emission vehicles (ZEV) policies to accelerate the electrification of cars and trucks, as well as increase equity in access to public charging facilities. However, given the decentralized models of charging station growth, individual station operators set prices and access policies, which have created data interoperability challenges for large-scale analysis of service operations. By guiding context learning with chain-of-thought prompting, we significantly reduce research evaluation costs with GPT-4, compared with conventional methods of analysis. Using this approach, we evaluate the state of the U.S. electric vehicle charging infrastructure from 2011–2022. The analysis covers 31,527 chargers nationwide, with special emphasis on reliability and distributive-equity issues that impact climate-disadvantaged communities.
-
Citizen generated intelligence for transport electrification policiesZero-emissions vehicle policies have been increasingly focused on the deployment and development of electric vehicle (EV) charging infrastructure. One fundamental barrier is the lack of common technology standards which has led to poor data interoperability between networks. In this article, we use machine learning to aggregate citizen-generated data to identify large-scale consumer issues related to infrastructure service provision in the context of electrification in the global transportation sector. Our dataset includes 407,051 publicly-accessible user reviews at 56,291 EV charging stations across the U.S., Europe, and East and Southeast Asia from 2010 to 2020. We employ transformer-based natural language processing (NLP) methods to classify topics of discussion from EV users in charging stations globally. We find evidence that the largest closed network Tesla provides EV consumers with better charging service reliability across regions than non-networked stations.
Publication list on Google Scholar →
Teaching
Teaching
Teaching Under a Lead Faculty
PUBP 3020 · Spring 2026 · Georgia Tech
Applied Political Economy
PUBP 3042 · Spring 2025 · Georgia Tech
Data Science for Public Policy
PUBP 3042 · Fall 2024 · Georgia Tech
Data Science for Public Policy
Guest Lectures
MGT 445 · Spring 2025 · University of Rhode Island
Strategic Management
PUBP 8751 · Spring 2023 · Georgia Tech
Big Data and Policy
Teaching Cases
Spatial Data Visualizations
Python and Tableau applied to big data and electric vehicle charging infrastructure.
ML for Electricity Demand
Using machine learning techniques to analyze electricity demand with Microsoft Azure.
ML for Air Quality Prediction
Using machine learning techniques to predict air quality with Microsoft Azure.
CV
Meet Udon, my dog friend and CV guardian.
Pet Udon 4 times to unlock my CV.
Click Udon to pet! 🐾
News
-
Apr 2026 I am deeply honored to receive the 2026 William H. Read Award — the highest distinction bestowed upon graduate students at the Carter School of Public Policy at Georgia Tech. It is a meaningful and humbling way to close my doctoral chapter, and I am grateful to the mentors, colleagues, and community that made this journey so rewarding.
