Job Market Paper: Who Rescues? Who Betrays? The Fate of Cross-Group Friendships in Wartime

Do cross-ethnic social ties completely cease to matter once the bullets start flying? Based on a new survey on survivors of the 1992-5 Bosnian War, I find that cross-ethnic assistance was not only widespread but highly correlated with the number of outgroup friendships pre-war. In contrast, help from complete strangers and active betrayal by close friends appear to have been rare. Drawing on 160 interviews, I show how cross-ethnic social networks were used to channel assistance to those in need from those with the resources and position to provide it. Most people, it appears, were willing to help not only friends but even friends-of-friends and acquaintances, provided the risk was not too great. The result was widespread assistance that was helpful but generally not life-saving. Finally, integrating census data and a comprehensive database of wartime fatalities with my survey results, I argue that intermarriage and residential integration led to more assistance but had little effect on the overall death toll. Policymakers should, therefore, continue to promote cross-ethnic social networks as a means to mitigate non-fatal suffering during mass violence, but should not rely on them as a means to prevent violence outright.

Download Working Paper

Online Appendices

Black Networks Matter

A historic tide of black-led protests made history in 2020 not only for their political impact, but for their ability to mobilize millions of people outside of a formal organizational framework. What explains this success? Although this stunning turnout may be partially attributable to impersonal forms of media such as YouTube videos and social media influencers with millions of followers, prior research suggests that peer recruitment among friends, family, and acquaintances is likely to have had a substantial impact. In this paper, we explore the extent to which black protesters have harnessed the power of their own personal networks, circumventing slower-moving hierarchical organizations, to build a true grassroots movement. To do so, we employ a novel instrument of unparalleled scope: a 25,000-respondent multi-wave panel survey including thousands of protesters and a targeted oversample of black adults. Our data allow us to address a wide array of questions about how black protesters use peer recruitment to drive turnout. 

Causal Estimation in Weighted and Directed Networks

Recent advances in causal inference have opened a new chapter in network research. Whereas past network experiments frequently used entire networks as their units of observation, requiring researchers to collect data on dozens if not hundreds of villages or schools, researchers can now intervene on individuals within a single network and watch their effects spillover onto untreated neighbors, classmates, and friends. Our ability to model these peer effects is still highly constrained, however, in part because we have yet to fully take advantage of the rich data networks have to offer. Social ties, for instance, can have varying strengths ("weights") which can be indirectly measured through the frequency of contact or survey responses. Other ties, such as who passes on information to whom, may also be directed—one way—rather than mutual. Building on Aronow (2012) and Athey, Eckles, and Imbens (2018), I develop a new method for using permutation inference on directed and weighted networks. I show that by leveraging such data, when available, we can achieve considerably more statistical power than existing methods, opening the door to smaller, lower-cost network experiments and quasi-experimental frameworks. I demonstrate the effectiveness of this technique first, using data from a conflict-intervention experiment, and second, from a natural experiment in a conflict setting.


Do Terrorist Attacks Always Affect Political Attitudes and Participation?

(with William Minozzi, Michael Neblo, and David Lazer)

Terrorists target an audience beyond their victims, and many scholars claim that terrorism often succeeds in shifting public opinion and behavior. People exposed to violence are said to exhibit a backlash against the perpetrator’s identity group, a rightward political shift, and heightened political engagement. We test these theories through a mixed-methods analysis of a natural experiment: during a multiyear panel survey on young Americans’ political opinions and behavior, a subset of respondents experienced a terrorist attack in their community. Difference-in-differences estimates show no meaningful causal effects on terror-exposed respondents. Furthermore, field interviews suggest that these individuals did not interpret this emotionally-jarring attack through a political lens, despite their own interest in politics and political leaders framing the attack in a highly politicized way. Our results call into question the scope of existing theories, suggesting that even close encounters with terrorism can fail to sway public opinion.

Download Working Paper

Online Appendices

Photo: Aftermath of terror attack at Ohio State University. Credit: Blervis (Wikipedia)

Modeling the Intrahousehold Spread of COVID-19

The COVID-19 pandemic has disrupted the lives of most people around the world. Understanding the extent to which intrahousehold spread is driving infection rates and the risk factors associated with such spread represents an important step toward containing the pandemic. Using an online panel of 66,072 unique survey-takers across the United States, we identify 1,432 intrahousehold networks of households with more than one resident and at least one infection. Within each household, we examine multiple networks based on parent-child ties, roommate ties, and other commonalities to determine which types of ties are associated with COVID-19 contagion. We find that larger households and households with more people per bedroom are more likely to report at least one case of COVID-19. While African-American and Latinx, low-income, and urban respondents stand a greater risk of having at least one household member become infected with COVID-19, these characteristics are not associated with likelihood of transmission within the household, which is primarily driven by household density irrespective of race or geographic context. Among members of an infected household, sharing a bedroom is significantly associated with COVID-19 transmission, while age and gender similarity among a pair of individuals is not. Our findings demonstrate that large-scale panel surveys can be applied for epidemiological surveillance, and suggest that individuals in households with enough room to isolate sick members—across all demographic groups—are at substantially lower risk of infection.


Stuck in the Mud: Roads, Rain and Repression

(with Carl Müller-Crepon)

Does government access to remote regions encourage or hinder armed conflict? Road networks have long been an essential tool of state-building for projecting state power from the capital into the hinterlands.  Such power could potentially deter would-be rebels and stamp out clashes between local ethnic groups or it could lead to increased rebel capacity and more state-initiated violence. To answer these questions, we use quasi-random variation in heavy rainfall as an instrument for whether certain roads are temporarily removed from the network due to flooding. We then examine whether outlying areas rendered-temporarily inaccessible experience an uptick in violence. Our evidence comes from a newly-digitized historical dataset of African road networks extracted from Michelin road maps using deep-learning neural networks, satellite-based weather data, and a high-resolution conflict event dataset. Based on our results, we expect to be able to highlight those parts of the African road network where improvements would foster state reach the most.


Segregation and Coexistance in Informal Settlements

In the summer of 2016 I spent eight weeks in Kampala, Uganda interviewing and surveying residents in slums about their friendship, secret-sharing, and money-borrowing networks.  My preliminary results suggest that migrants from the North of Uganda are more likely to segregate themselves along ethnic lines, while those hailing from the center are more likely to form cross-ethnic ties.  I am currently exploring to what degree this homophily (self-segregation) can be attributed to war-related trauma from internally displaced persons (IDPs) who fled the conflict between the Ugandan military and the Lord's Resistance Army.  In addition, there is evidence to suggest that residents who have spent more time in the diverse environment of Kampala are more likely to be open to friendships with other tribes. Hence, a slum, in spite of crime, unemployment, and conflict over property rights, can potentially serve as a breeding ground for coexistence.

​Other Projects

  • An online randomized control trial to increase voter turnout through student friendship networks

  • An analysis of racial and gender bias in the interdisciplinary field of network science

  • A randomized control trial in rural India looking at how family planning interventions can multiply their impact by taking advantage of pre-existing friendships and kinship ties


Below: Connections among my Facebook friends as of Fall 2015. Purple nodes are Williams College alumni, dark green are high school classmates, light green are former high school athletes I've coached, blue are miscellaneous Boston friends,  and yellow are campers and counselors from Seeds of Peace International Camp.