I am a researcher in international relations, focusing on economic statecraft, as well as influence and information operations. I also develop machine learning methods for estimating heterogeneous treatment effects in experimental and observational data.
My book project Alms and Influence examines when economic inducements – such as foreign aid, large-scale investment initiatives, and discounted sales of natural resources – can buy influence abroad. I argue that inducements can be lucrative to not just the recipient but also the sender. This leads to what I call the inducement dilemma: when a state profits from giving inducements, it will be less likely to cut them off, even when it does not receive any concessions. Consequently, the recipient has less incentive to concede, rendering the inducement ineffective in extracting concessions. For an inducement to be useful in extracting concessions, its credibility must come from the punishment for reneging, rather than the profit from delivering. I test these dynamics in China’s Belt and Road Initiative, the Marshall Plan, and Russia’s energy diplomacy in the 21st century.
I hold a PhD in Politics from Princeton and have taught at Oxford, Georgetown, and the Institute for Qualitative and Multi-Method Research (IQMR). Before academia, I led an analytics team at Google and worked at the United Nations, World Bank, and OECD on development assistance, infrastructure financing, and industrial policy.
Publications
Behind the Screen: Understanding National Support for an Investment Screening Mechanism in the European Union
(with Sophie Meunier) Review of International Organizations, 2022, 17(3): 513–41.
What determines national preferences for institutionalizing FDI screening? Over the past decade, advanced economies worldwide have tightened their national investment screening mechanisms (ISMs) for foreign direct investment (FDI). In March 2019, the European Union (EU) adopted its first common FDI screening framework. Based on extensive interviews with high-level EU and country officials involved in the negotiation process, and using a unique measure of national support for the EU-wide ISM created through the first-ever elite survey on this subject matter, we find that countries with higher technological levels were more supportive of FDI screening due to concerns over unreciprocated technological transfer. We also find sector-dependent effects of Chinese FDI on country-level support for FDI screening: Countries with high levels of Chinese FDI in strategic sectors are more likely to support the EU ISM, while those with high levels of Chinese investment in low-tech sectors tend to oppose screening. Our overall findings suggest that EU investment screening, and national-level screening in general, might become more restrictive in the future, especially in light of the COVID-19 pandemic.
Testing Concerns about Technology’s Behavioral Impacts with N-of-one Trials
(with J. Nathan Matias and Eric Pennington) Proceedings of the ACM Conference on Fairness, Accountability, and Transparency, 2022: 1722–32.
As public trust in technology companies has declined, people are questioning the effects of digital technologies in their lives. In this context, many evidence-free claims from corporations and tech critics are widely circulated. How can members of the public make evidence-based decisions about digital technology in their lives? In clinical fields, N -of-one trials enable participant-investigators to make personalized causal discoveries about managing health, improving fitness, and improving their education. Similar methods could help community scientists understand and manage how they use digital technologies. In this paper, we introduce Conjecture, a system for coordinating N-of-one trials that can guide personal decisions about technology use and contribute to science. We describe N-of-one trials as a design challenge and present the design of the Conjecture system. We evaluate the system with a field experiment that tests folk theories about the influence of colorful screens on alleged phone addiction. We present findings on the design of N-of-one-trial systems based on submitted data, interviews, and surveys with 14 participants. Taken together, this paper introduces N-of-one trials as a fruitful direction for computer scientists designing industry-independent systems for evidence-based technology governance and accountability.
Audience Costs and the Credibility of Commitments
(with Samuel Liu and Kai Quek) Oxford Bibliographies in International Relations, 2021.
Working Papers
Affluence without Influence? The Inducement Dilemma in Economic Statecraft (under review)
When can economic inducements——such as foreign aid, investment, and especially large-scale development initiatives——buy influence abroad? Countries often use financial favors to induce foreign policy concessions from other countries. The effectiveness of such inducements hinges on whether the sender can credibly threaten to halt or withdraw the inducements when the target does not concede. I examine a substantial set of development initiatives that are lucrative not just for the target but also for the sender. I argue that when the sender profits from the inducement it gives, it will not cut off the inducement, even if the target does not concede. I test this inducement dilemma in China’s Belt and Road Initiative (BRI). Using over 200 elite interviews, official documents published by the Chinese government, and original datasets on China’s overseas project contracts, I show that Beijing’s dual goals of the BRI are to (1) tackle domestic economic and environmental problems by encouraging Chinese companies to implement infrastructure projects and invest abroad, and (2) gain international acceptance of China’s development and governance models. Consistent with my argument, the profit motive undercuts the foreign policy goal. These infrastructure projects promote international support for China’s governance and development models only when these projects do not serve China’s economic motive of promoting outward direct investment.
Divide and Conquer: Russian Information Operations and Polarization Through One-Sided Narratives (with Noel Foster, under review)
Growing ideological polarization now threatens democracies worldwide. This article addresses the inconclusive findings in political science regarding the role of counterattitudinal information in fueling such polarization. We argue that exposure to factual information embedded in one-sided narratives triggers reactance among those perceiving it as counterattitudinal, leading to backfiring, while simultaneously inducing confirmation bias among those who align with the information, hardening pre-existing beliefs. These two psychological phenomena combined lead to heightened polarization. We test our arguments using original, pre-registered survey experiments based on Russian information operations in Estonia, a NATO frontline state and prominent Kremlin target. Consistent with our theory, we find that one-sided factual content on Soviet legacy and migration polarizes voters along ethnolinguistic and pre-existing ideological cleavages. These findings advance scholarly understanding of the sources of ideological polarization and effects of information operations, particularly in the context of malign foreign influences.
The Indirect Effect: How Hidden "Relay Stations" Amplify Russian Information Operations (with Noel Foster, under review)
Why do states like Russia invest heavily in foreign-facing media with small audiences? We argue that these states spread information operations through "relay stations"-credible, independent local media in the target country-enabling indirect consumption by audiences. Relay stations work via two mechanisms: direct citations and appropriation of content from information operations. Focusing on Taiwan's coverage of Russia's invasion of Ukraine, we find that journalists rarely attribute negative connotations to Russian state-sponsored outlets and instead incorporate Russian content into reporting. Using an original multilingual dictionary and argument similarity measure, we analyzed 17,258 news articles, finding systematic evidence that reputable, major news outlets in Taiwan cited Russian government-sponsored sources and echoed Russian narratives. This indirect spread of information operations through relay stations suggests that prior research may have underestimated exposure to such operations, making countermeasures more difficult than previously anticipated.
Tying-Hands Versus Bluster: Authoritativeness, Words, and Deeds in Crisis Communication (with Noel Foster and Jackie S.H. Wong, under review)
When do public statements reveal a state's true intentions? While much of the international relations literature argues that states issue public threats during crises to tie their own hands, recent research suggests that public signals are often noisier than private ones and states bluster to de-escalate. We contend that a state can simultaneously tie its own hands and bluster during a crisis, albeit through communication channels of different authoritativeness. Less authoritative channels offer plausible deniability, making bluster more likely, whereas states are more inclined to tie their hands through more authoritative channels. Examining China's public threats to Taiwan in 3,635 newspaper articles from 2016–2022, we find systematic evidence supporting our argument. We show that only military threats published in People’s Daily — the most authoritative newspaper of the Chinese Communist Party — forecast incursions by the People’s Liberation Army into Taiwan’s air defense identification zone. In contrast, threats published in less authoritative outlets like the Global Times do not. Our findings reconcile the seemingly contradictory mechanisms of tying hands and bluster, and offer implications for crisis de-escalation.
Attributions and Deescalation: The Public Dynamics of U.S.–China Crisis Deescalation (with Kai Quek, under review)
Policymakers and scholars alike have expressed grave concerns over the risks of an inadvertent crisis between the United States and China. What could be done to deescalate such a crisis and reduce the risk of war? Direct theoretical and empirical guidance on this question is scarce. Existing literature suggests that publics in both countries are nationalistic and reward confrontation and displays of strength. We challenge this conclusion and argue that handling a crisis with restraint through responsibility sharing can facilitate deescalation. First, we construct an interactive theory of crisis deescalation based on (1) attribution, where one side can send a public signal to attribute blame for the crisis to self, to other, or to neither parties; and (2) response, where the other side can either accept or reject the attribution. Then, we design parallel and interactive survey experiments in the United States and China to map our theory to real-world empirics. We find that blaming neither parties by attributing the crisis as an accident received the strongest approval from both American and Chinese citizens, especially when the other side cooperates in accepting the attribution. Our findings illuminate the public feasibilities of different policy pathways of deescalating an inadvertent crisis and reducing the risk of war.
Transparent Machine Learning through Improved Variable Importance Measures (with Marc Ratkovic)
Boosting and random forests are among the best off-the-shelf prediction tools. These methods offer a variable importance measure (VIM), which is a cumulative measure of the improvement in accuracy over the algorithm. We show existing variable importance measures, as implemented, are biased, returning positive scores on irrelevant variables. Intuitively, if a variable is irrelevant but correlates with a relevant variable, this correlation may lead to an improvement in performance may be misattributed to the irrelevant variable. We introduce a method that removes this bias. The method works by separating each predictor into a component explained by other predictors (a "predicted variable"), and a component not (a "partialed out variable"). We assess variable importance only through any improvement attributable to the latter. We prove the method returns a valid VIM, meaning it is mean-zero and asymptotically normal for irrelevant variables. Simulation evidence and applications to UCI data suggest the method also performs favorably relative to several existing machine learning methods in terms of predictive accuracy.
