New findings from conflict archives

Introduction

For decades, scholarship in the field of peace and conflict studies has relied on data gathered from secondary historical accounts and news media reports. While useful for comparative analysis, these data sources have demonstrable biases towards overrepresenting easily observable acts of violence (Davenport & Ball, 2002; Davenport, 2009; Sullivan, 2016a; Weidmann, 2015). One important consequence is that theoretical and empirical analyses are often predisposed towards studying the most visible violent actions and processes (e.g. acts of terror, large demonstrations, pitched battles, fatal shootings, etc.).

To reveal the hidden mechanisms underpinning political conflict, researchers have begun systematically collecting and analyzing thousands of historical records from conflict archives. When we talk about ‘conflict archives’ we refer to catalogs of written material systematically produced as part of a political contest between a government force and an armed challenger (either domestic or international) and later collected and preserved for public access. Materials contained within conflict archives commonly include memos, investigative reports, communiqués, and other documents produced by government or rebel forces. Such records are inherent to conflict; they serve as the primary conduits through which combatants acquire, accumulate, and distribute information. Because combatants produce these documents contemporaneously with violence, the records reveal how actors understood the strategic environment in real time. And because the forces producing these documents were on the ground conducting operations in situations of heightened insecurity, their archives reveal information on violent and nonviolent behaviors that cannot be easily monitored by third-party observers (e.g. NGOs, the media). As a result, data from conflict archives typically contain better coverage (across space and time) and more information (on actors, behaviors, beliefs, and context) than most other systematic collections of material.

This special issue highlights empirical findings and theoretical insights emerging from a research program based around conflict archives. In this introduction, we discuss how material found in archived documents compares to more traditional sources of conflict data. By considering the contributing articles in relation to one another, we hope to provide readers with a comprehensive understanding of field research strategies and analytical techniques for studying original data from conflict archives.

The articles contained in this special issue reveal how researchers use data from conflict archives to challenge widely held conventions and/or reveal insights into aspects of violence not often considered due to lack of credible data. In this way, the common elements unifying this research agenda extend beyond methodological concerns. New findings from conflict archives reveal mechanisms and processes previously buried beneath the fog of war. Using data from conflict archives, research in this special issue connects micro-, meso-, and macro-levels of analyses, demonstrates how identities intersect conflict cleavages in varying ways, and depicts the ‘gray zone of power’ between institutions and violence (cf. Auyero, 2007).

The remaining sections of this article assess the past, present, and future of research on conflict archives. We first review the history of archival research in conflict studies and explain its exponential growth in recent years. We then provide an assessment of the methodological debates emerging from this kind of research and specify a set of best standard practices. Finally, we summarize the contributions to this special issue, documenting how a new generation of research employs conflict archives to shed light on processes of peace and conflict. The final section concludes.

Research from conflict archives: A short review

The convergence of research from conflict archives across the globe reflects two congruent trends: (1) exponential growth in access to conflict archives combined with (2) conflict studies’ increasing attention to data quality and measurement error. With regards to archival access, activists in both the domestic and international arenas continue to push for increased transparency in government. At the same time, practitioners working on peace and reconciliation have progressively come to view the release of conflict archives as key to transitional justice. As access to previously confidential records expands, leading archivists are directing the often years-long processes of diligently cleaning, organizing, and preserving materials into an expanding list of conflict archives, examples of which can be found in Table I of the Online appendix.

For both activists and archivists, the systematic preservation and release of records perform truth telling purposes that are independent of their ability to advance scientific research. ‘When records of past surveillance and repression are made public’, Wisser & Blanco-Rivera (2016: 126) argue, ‘they take on new meaning. Initially used by the state as a means to control citizens and quell opposition, these records can become the evidence used by victims and their relatives for the purposes of accountability, memory making and truth seeking.’ Indeed, in the aftermath of a highly violent conflict, the release of archived documents can constitute an important component of truth and reconciliation processes, revealing to victims’ families what happened to their kin and providing important sources of evidence to prosecute perpetrators.

Simultaneous to the expansion of access to conflict archives, conflict scholars have emerged from the methodological debates of the previous two decades with demands for better data along with a renewed interest in combining quantitative and qualitative methods (cf. Davenport, 2010; Gohdes & Price, 2012; Eck, 2012; Morgan, 2013; Arjona, 2016; Balcells, 2017).

In the social sciences, important work pioneered the path to today’s research, including Rude (1964) and Tilly, Tilly & Tilly (1975), who drew on historic police registers to compile listings of conflict related events, and Weber (1976) and Meyer et al. (1979), who looked into administrative records to examine the expansion of state authority. Like their contemporary counterparts, these studies extracted information methodically from a set of archival materials, so that that information could be compared over time and space. But in many cases, it remains unclear how records were sampled, what the exact coding and measurement procedures were, or what standards were being applied to generate inference.

New research on conflict archives adds rigor to this tradition by systematizing the sampling process, drawing clear and replicable measurement rules, and studying the archival materials with sophisticated analytical techniques. Of particular note is the scholarship of Roger Gould, which serves as an exemplar of how historical archives can be used to solve big questions in social science, and in particular how archives can test theories relevant for contemporary conflict (e.g. Gould, 1995, 1999, 2003). Other noteworthy efforts to systematically employ data from conflict archives to assess theories of peace and conflict include Cunningham (2004), Kalyvas (2006), Lawrence (2013), Davenport (2014), Braun (2016), and Finkel (2017).

To illustrate the transformative effects of this research program, the next section assesses some methodological debates in the study of conflict archives.

Advantages and limitations of data from conflict archives

Archives often provide better coverage across space and time than secondary sources or data produced by third-party observers, such as NGOs or the media. This is true partially because the forces responsible for creating the archived documents typically have more resources than outside monitors. But more importantly, violence systematically distorts both the supply and demand of information gathered by third-party observers (e.g. Cohen & Greene, 2012; Sullivan, 2016a; Weidmann, 2015; Baum & Zhukov, 2015).

Reflecting a famous dictum, if a body falls in a conflict and no one is there to observe it, will it impact our results? The short answer is we rarely know; the longer answer is that it will depend on whether, how severely, and in which direction systematic missingness correlates with our explanatory models (King et al., 1994). Several recent examples reveal the implications for peace and conflict. For instance, biases in media data tend to validate reactionary ‘threat–response’ models of repression, whereas Sullivan’s (2016a,b) archival research reveals a hidden layer of ‘pre-emptive repression’ designed to limit the emergence of future challengers. Similarly, while media sources readily report on observable violence, Zhukov’s (2016) archival analysis reveals that when punishment strategies fail, governments turn to forcible disarmament as a way of denying rebel advances. Further examining situations where violence limited third-party monitoring, Kaplan (2013) uses interviews alongside conflict archives to introduce the role of nonviolent civilian resistance into a body of conflict research that has long been focused narrowly on rebel–government interaction. In each of these instances, scholars deployed conflict archives to overturn prevailing assumptions based largely upon more conventional data sources.

Beyond increasing the scope of available information, conflict archives also commonly provide greater depth of coverage across event types as well as greater subnational disaggregation. This is particularly the case if local archives are used in conjunction with archives at the national level (e.g. Balcells, 2017). In some cases, recent work has combined historical process tracing with conflict archives to study the mechanisms underlying important political phenomena and resolve persistent theoretical debates (e.g. Lawrence, 2013; Schwartz & Straus, 2018). Archives are particularly useful to undertake process tracing, as they can provide information on the wartime processes that are absent in other data sources.

Finally, archives allow for the study of new topics, including the direct examination of the reporting structures combatants rely on to commit acts of violence. Conflict archives can provide first-hand information on the behavior of participants in a conflict along with new meta-data about how the organization producing the documents understood and reported that behavior. For example, surveillance information is passed across a military to direct offensive strategies towards areas where violence is thought to be most effective. Later, the command staff may catalog those reports to evaluate the impacts of violence. When these documents are preserved in conflict archives, they convey information both about what was happening on the ground and how the combatant(s) understood and interpreted those events. In settings when archive materials are based upon government records, conflict archives allow us to literally see like a state (Scott, 1998).

To be clear – our argument is not that archives should be the source of data for studies of peace and conflict, but instead that in many cases conflict archives provide more comprehensive and less biased information on political conflict than any other systematic collection of data. All sources of conflict data are likely to present incomplete representations of political violence. Still, in our assessment raw (i.e. unprocessed) material generated by combatants and collected from conflict archives gets researchers closer to the ground truth of what transpired in a particular event.

Regarding limitations, archival data have political biases, both with regards to their availability (e.g. records can be destroyed or made otherwise unavailable; many are blocked by state secrecy laws) and in the underlying materials. Governments do not commonly make a full listing of their records available. And when documents are released, it is commonly a strategic act as the current government seeks to shame its rivals by releasing documents that are particularly damaging to the opposition. Indeed, this is one way in which victors define ‘history’. The politically motivated release of archival materials may be entirely distinct from their scholarly analytical purpose, and may involve deliberate manipulations of the types of documents that get released and/or the types of information that are collected. State forces can directly affect the flow of materials in an effort to rewrite history, as the Rwandan government did by limiting the collection of information on conflict related violence to acts of genocide (Loyle & Davenport, 2015). In another example, the Causa General in Spain was created specifically to bring only the losers of the civil war to justice (Balcells, 2017).

Archives may also provide information in units of analysis that are not theoretically relevant or useful. For example, governments typically collect their data by administrative districts, and these units may serve as imperfect proxies for testing our theories. Also, units can change over time, and the information on units from period t might not be useful for the understanding of phenomena measured at different units in t+n. Further, the types of information gathered in records and reports may shift over time and space as combatants respond to conflict. For example, Douglass (2016) demonstrates how missingness in the attributes collected by US Army Intelligence during the Vietnam War varied systematically across both the type of operation and the demographics of potential targets. There may also be incentives to report acts of government violence differently from rebel violence. Guberek & Hedstrom (2017) show how the terminology used to describe killings in the Guatemalan National Police Archive shifted across time and across political regimes. Thus, even when an archive provides nearly unprecedented access to repressive institutions, important ‘silences’ remain that may impact subsequent analyses (Guberek & Hedstrom, 2017; cf. Sullivan, 2016a; Weld, 2014). Douglass & Harkness (2018) take this as a starting point before demonstrating how machine learning can be used to construct appropriate units of analysis based on how relevant information varies as a function of conflict.

Acknowledging the challenges discussed above, we now review several additional best standard practices for scholars sampling, coding, and analyzing data from conflict archives. ¹

Best practices for generating and analyzing data from conflict archives

Contributions to the Special Issue

Articles in this issue delve into conflict archives, employing sophisticated methodologies to analyze data and obtain results that speak to central questions in social science. Using data from a wide range of locations, the researchers identify new research questions, challenge conventional wisdom, and present new findings to shape the direction of conflict studies.

A major theme emerging from this research is the organization and operation of state security forces. Many of the articles focus on operations and violence perpetrated by governments. Given their superior resources and typically more established organizational structure, governments collect and preserve documents pertaining to their war efforts far more comprehensively than rebels. And while rebel records do get released, those challenging the state face incentives to minimize any paper trails that could allow the state to target them with repression. But, as we discuss further in the conclusion, the focus on government repression extends beyond the origins of conflict archives. Conflict studies has worked hard to match the study of repression to the study of non-state violence. But such efforts have been hampered both by data availability and by the political motivations of (government) agencies funding large research projects. As a result, there remains much to learn about the operation of coercive governments, and conflict archives offer unique opportunities in which to study their operations.

From this work, we learn how political violence informs and reshapes repressive institutions. Eck (2018) uses detailed archival data on British security inputs to colonial territories to explore an interesting hypothesis: the trajectories of conflict observed prior to independence explain the different paths that states take in developing security and surveillance capacity. States with conflict experience received more colonial security inputs, enabling them to develop police forces that commit fewer violations of physical integrity rights than states that did not have an experience of conflict. In a way, Eck finds that the experience of counterinsurgency made stronger states, which is consistent with the literature on counterinsurgency and civil war that uses conventional sources of data (e.g. Balcells & Kalyvas, 2014). Eck’s (2018) combination of qualitative and quantitative analysis reveals new evidence on previously unobserved mechanisms linking conflict to the expansion of the security apparatus.

Several other contributions in this issue reveal how governments accumulate and deploy intelligence. Hassan & O’Mealia (2018) look at the challenges of managing an effective bureaucracy in a relatively weak state. Given the difficulties of monitoring bureaucratic agents spread out across the state’s territory, central governments look for cues regarding who is loyal or disloyal. Somewhat counter-intuitively, the authors argue that election violence can serve as a signal that influences the career trajectory of individual bureaucrats. Using government records, they show that pro-incumbent election violence signals local bureaucrats’ loyalty to the central government, while opposition initiated election violence demonstrates that local agents are either inept or disloyal. Like Hassan & O’Mealia (2018), Sullivan & Davenport (2018) look at subjects that are difficult for the state to observe. In this case, the authors examine the government’s ability to monitor the clandestine behaviors of political challengers. Employing archival reports from a variety of US intelligence agencies along with documents produced by a black nationalist social movement organization, they show how challengers adapt to overt repression in ways that make subsequent state surveillance more difficult. The results demonstrate how overt repression actually drains state intelligence, challenging conventional wisdom on counterinsurgency and counterterrorism.

Douglass & Harkness’s (2018) study can be read either from a purely methodological point of view or from the view of a theory of state surveillance. In this methods article, the authors provide a new technique for analyzing variance in the information produced within archival sources, which optimizes the spatial units that researchers should use when geo-referencing their data. This minimizes biases produced by the geographic dispersal in document production and the modified area unit problem. As a theoretically relevant piece of conflict scholarship, the article sheds light on how difficult it is for governments to identify challenger activities, particularly those occurring in areas with weak government control and a dense landscape. This finding validates predictions from several dominant models of civil war (e.g. Fearon & Laitin, 2003; Kalyvas, 2006), while simultaneously challenging the assumption that states are capable of precipitating or responding to dissent at will.

Violent conflict has political consequences, and this is often due to the fact that political factors underlie violent dynamics. A second set of articles study dynamics of lethal and non-lethal violence with archival data, and they point to political and ideological factors underlying patterns of violence in different contexts. Scharpf (2018) exploits archival data from Argentina’s Dirty War (1975–81) to study how ideological leanings of military officers shape subnational patterns of state violence. He distinguishes between nationalist officers, who dominated the infantry, artillery, and communications branch, and liberal-conservative officers, who served in the cavalry branch. Because of their ideological convictions, Scharpf (2018) finds evidence that nationalist officers were more prone to use large-scale terror against the broader population, while liberal-conservatives advocated the selective targeting of guerrilla forces and their supporters.

Schwartz & Straus (2018) study a slightly different set of ideological trade-offs in their research focusing on mass atrocities in Guatemala. Following a recent trend in the field (e.g. Kalyvas, 2006; Arjona, 2016; Balcells, 2017), these authors highlight the importance of local-level analysis of patterns of violence. In particular, they emphasize the relevance of local interpretations of military strategy and insurgent threat. They leverage the 359 pages of military plans and communications that comprise Operation Sofía, a one-month-long counterinsurgent campaign waged in the northwestern Maya Ixil region. The authors find some evidence that state actors understood the civilian population as fully dedicated to rebel forces, although their results also suggest that troops on the ground perceived Ixil communities as ‘winnable’, and thus deployed violence for coercive (rather than purely ideological) purposes. Balcells (2018) uses data from the Spanish Civil War (1936–39) to analyze the relationship between violence and displacement during an ideologically polarized civil war. She finds that refugees tended to cluster by ideology when they settled in hosting localities, and that the arrival of refugees sometimes had the side effect of triggering additional violence against local enemies.

Drawing on data from different historical periods and different continents, a final set of studies in this special issue examines the legacies of violent conflict. Osorio, Schubiger & Weintraub (2018) focus on repression carried out by the Mexican government in the 1960s and 1970s, and its long-term consequences for state consolidation. They focus on a particular type of repression: disappearances. Relying on data gathered from a secret governmental report that was subsequently leaked to the press, they find that disappearances have mixed effects on state consolidation: on the one hand, such abuses are positively correlated with contemporary measures of fiscal and bureaucratic capacity; on the other hand, they are also negatively correlated with measures of state monopoly over the use of force. Zhukov & Talibova (2018) examine archival records of Stalinist political imprisonments and identify their effect on contemporary Russian politics. Using proximity to the railroad as an instrumental variable, they find that deportations have a significant impact on contemporary political behavior: they lead to a decrease in both electoral participation and support for Putin. They also find that deportation has influenced political behavior in Ukraine, where turnout is lower in localities where Stalin-era repression was high.

Conclusion

The unique information contained in conflict archives provides new opportunities for conflict scholars to illuminate previously opaque mechanisms and processes motivating violence. Critically, there is a certain subversive element to the study of conflict archives. To prevent their opponents from accessing sensitive information, governments carefully protect and preserve the data and analysis contained in conflict archives. The fact that conflict archives are typically kept confidential even after the cessation of violence implies that they also contain some residual value for those hoping to challenge the state (Art, 2016). ⁶ Finding, releasing, and analyzing materials from conflict archives democratizes knowledge that had previously remained proprietary, bringing into the public sphere formerly secretive information on the causes and consequences of violent resistance and/or repression (Nalepa, 2010).

We hope this issue will stimulate additional research by releasing new data collected from archives across the world, which may later be subjected to comparative analysis. ⁷ Peace and conflict research has moved away from the dichotomy of micro-level case studies and cross-national statistics and towards analyses that include both within-case and across-case analyses (e.g. Christia, 2012; Balcells, 2017). We hope to further this important analytical endeavor. ⁸

Some of the limitations of conventional research are not fully solved in research based on archival data. For example, as discussed above, the selection bias issues related to the data generating process are not always well addressed by researchers. Scholars should do a better job describing how the data generating process might influence their study, and presenting ways to solve the bias. In addition, we still need to attend to the challenges that all scholars working with observational data face with regards to causal inference.

Overall, with its limitations, the unique information contained in conflict archives provides new opportunities for conflict scholars to identify new findings that illuminate the mechanisms and processes motivating violence. This special issue intends to establish a set of standard procedures by which the study of conflict via archives should be executed, and it presents some novel research using conflict archives in a groundbreaking way.

Replication data