Stalin’s Terror and the Long-Term Political Effects of Mass Repression
(with Yuri Zhukov)
Repression has a long-term negative effect on political participation. Using millions of arrest records from archival documents, and polling station-level election results, we examine how exposure to Stalin-era repression affects voter turnout in Putin’s Russia. To estimate the effect of repression on voting, we exploit exogenous variation in repression due to the structure of mid-century Soviet railroads, and travel distances to prison camps. We find that communities more heavily repressed under Stalin are less likely to vote today. The electoral legacy of Stalin’s terror – decades after the Soviet collapse, and across multiple election cycles (2003–12) – is systematically lower turnout. To show that our result is not unique to the Putin regime, we replicate our analysis in Ukraine (2004–14), and find similar patterns. These results highlight the negative consequences of repression for political behavior, and challenge the emerging view that exposure to violence increases political engagement. While past research has emphasized the short-term effects of repression over several months or years, we show that these effects may be durable over generations and even changes of political regime. Our findings also demonstrate that repression need not be collective or indiscriminate to have community-level effects.
Zhukov, Yuri M., and Roya Talibova. "Stalin’s terror and the long-term political effects of mass repression." Journal of Peace Research 55, no. 2 (2018): 267-283.
Terrorist Group Abilities, Ideology & Motives (TAIM): Introducing a New Global Database on Terrorist Groups and Activities (with Carly Wayne)
What are the factors that influence the patterns of terrorist violence? Terrorism is often a group-based phenomenon, perpetrated by individuals who claim membership in a larger organization. However, the unit of analysis for many existing terrorism datasets is at the event-level. In this paper, we address this issue, introducing a new cross-national dataset of characteristics of terrorist attacks and groups: the Terrorist Abilities, Ideologies & Motives (TAIM) Metadata. The TAIM Metadata contains a set of three comprehensive data sets of militant groups who have actively engaged in terrorism between 1970 to 2015, drawn from the population of terrorist attacks in the Global Terrorism Database. Designed hierarchically, the TAIM Metadata includes TAIM-event (66,675 obs and 142 variables), TAIM- group-year (4,405 obs and 69 variables), and TAIM-group (935 obs and 47 variables). TAIM is unique in both the breadth of organizations covered — 935 unique militant groups in 117 countries over a 45-year time-span — and the number of original variables coded for each organization, including organizational-level features such as motive, ideology, strength, and birth country, among several others, and dynamically changing event-level attributes such as lethality to date, targeted country, attack frequency to date, and more. The complexity and scope of this dataset open multiple new avenues for the study of terrorism, providing a unique opportunity for scholars to carry out research on multiple levels of analyses in order to uncover relationships between group- and event-level attributes and outcomes over time.
Fighting for Tyranny: How State Repression Shapes Military
Performance (with Arturas Rozenas and Yuri Zhukov)
A state’s military power ultimately rests on the efforts of ordinary citizens in battle. But what if people see the state they are defending as unjust or even tyrannical? To investigate this question, we assemble a novel dataset from over 100 million declassified personnel records of Red Army conscripts in the Second World War, and detailed data on Stalin’s mass repression before the war. Results from three empirical designs show that soldiers from places with more pre-war repression were more likely to fight until death and less likely to flee, but they also displayed less initiative in battle. This finding underscores an overlooked negative externality of repression: past exposure to repression induces conformity, which may help solve some principal-agent problems associated with fighting, but it comes at the expense of military effectiveness and higher wartime casualties.
Political Violence and its Legacies
State Repression and War-Making: Raising an Army of Dissidents
Over the course of the past few centuries, authoritarian states have raised and trained strong armies that have waged deadly conflicts. Despite the ubiquity of such regimes in contemporary times, there is very little research on the role of state repression in shaping individuals' decisions to support the state in its war-making effort. I develop a novel theory of combat motivation that formulates the conditions under which different levels of repression can incite or depress combat motivation. The most common explanation in the literature for how citizens decide on a trade-off between supporting or fighting the state has rested on the logic of coercion, whereby compliance of the citizens can be achieved through only selective violence, which gives citizens an opportunity to avoid it by behaving in a desired manner. I present a model in which this mechanism plays out in the opposite direction, such that indiscriminate repression motivates individuals to support the regime in its war-making efforts to signal loyalty. The model illustrates how, in anticipation of international conflict, an authoritarian has an incentive to target population indiscriminately based on group-level characteristics. The intensity of such repression, however, is qualified by a trade-off between the imminence of external threat and the possibility of a domestic uprising.
The Imperial Army and the Russian Civil War: The Path From Oppression to Insurrection
How does army experience affect a disenfranchised minority group’s post-war behavior? Past research has identified the crucial role of combat experience in improving organizational skills and the ability to engage in collective action. However, little is known about how combat veterans utilize these newly gained skills in a post-conflict society and whether there are meaningful differences among veterans from elite and minority groups. To study the long-run legacies of oppression at wartime and in the aftermath of conflict, I use a set of two unique datasets containing millions of individual records on the Russian Imperial Army conscripts in the WWI and the soldiers of the Red Guard during the Russian Civil War collected from declassified archival documents. I conduct analyses at two different levels – individual and community (based on birth location) – to study how the army experiences of these soldiers from the marginalized and disenfranchised minority groups impacted their behavior toward the Russian state after WWI. I exploit a Bayesian Nonparametric Spatial Regression Discontinuity design at the district borders to empirically identify the post-conflict behavior of former Imperial soldiers during the Russian Civil War. I control for several characteristics of the Russian state and society using district-level covariates constructed from the 1897 Russian Imperial Census. The preliminary findings suggest that soldiers from marginalized minority groups were more likely to join the Red Guard to fight against the crumbling empire. Further, inhabitants of districts with a higher percentage of marginalized groups were more likely to join revolutionary forces.
Terrorist Group Abilities, Ideology & Motives (tAIM): Introducing a New Global Database on Terrorist Groups and Activities (with Carly Wayne)
What explains the tremendous variation in targeting strategies between different terrorist organizations? This article examines the relationship between terrorist group objectives and tactics using a newly created data set of terrorist organizations. We hypothesize that the relative scope of a terrorist group’s motives — whether they are limited or maximalist — will lead groups to choose different types of attack and targeting strategies. Specifically, we argue that groups with limited aims will utilize an attrition strategy, designed to inflict persistent pain that induces a government to concede policy objectives, while groups with maximalist goals will pursue a provocation strategy, designed to achieve important process goals for the group so that the group can eventually take what they want by force. Using a newly built cross-national dataset of 66,675 attacks by 935 different terrorist organizations, we find that, indeed, groups with limited goals are more likely to use conventional weapons to frequently attack mass public targets, while groups with maximalist goals are significantly more likely to utilize sensational weaponry and launch rarer attacks against symbolic government targets. Because of this emphasis on sensational weaponry, attacks by maximalist groups are more lethal, despite their focus on government (rather than mass public) targets. This research has important implications for our understanding of terrorist violence, demonstrating the crucial role motives play in structuring terrorist groups’ strategic in- incentives.
Data and Measurement
Decision Tree Guided Multi-Class Support Vector Machines with Dynamic Class Selections
Large-scale terrorism datasets are inherently different from other political science datasets in that they are mostly incomplete owing to the clandestine nature of the very phenomenon they seek to capture. This problem is compounded by two important factors: event-based terrorism databases generally rely on news reports that rarely provide full facts about these incidents, and perpetrators of these acts usually prefer not to disclose their identity or supply reliable data on their activities. The unclaimed terrorist attacks thus pose some challenging inference problems for researchers interested in using such datasets to test hypotheses related to terrorist groups. In this paper, I attempt to solve this problem of missing data for the most comprehensive event-based terrorism dataset – Global Terrorism Dataset (GTD) –, by using machine learning algorithms. I first use Multivariate Imputation by Chained Equations (MICE) to impute missing event characteristics (such as types of attack, weapon, and target, numbers of killed and wounded, etc.), and then use these complete attributes as predictors in a novel Decision-Tree guided Multi-class Support Vector Machine algorithm to attribute unclaimed terror attacks to known terror groups. I classify all of the events that are listed as having unknown perpetrators, which accounts for half of the entire dataset, by comparing their properties with information obtained from events whose perpetrators are known.
Political Violence and its Legacies
Fealty tested, loyalty forgotten: Does military service exempt veterans from post-war repression?
Human Rights Violations in Turkey: Information Contexts and Reporting Biases
(with Christopher Fariss)
Counting Battle Deaths: Shaping Outcomes in Civil Wars
(with Christian Davenport)