“Constraints, Instrumentality & Signalling: Explaining Sexual Violence by State-Actors in Armed Conflict. An Empirical Analysis, 1989-2009”, M.A. thesis at the University of Stuttgart and SciencesPo Bordeaux, 2015 (Replication data)
Recent research on conflict-related sexual violence has underlined,
that state actors are more likely than non-state actors to perpetrate sexual violence during armed conflicts, and
that sexual violence varies considerably across time, space, involved armed actors and conflicts and is hence far from being inevitable in wartime.
Drawing on the most extensive data set on wartime sexual violence between 1989 and 2009 (Cohen / Nordås 2014), this paper further explores the question under which circumstances armed state actors are more likely to engage in wartime sexual violence and why they perpetrate such violence.
While an abundance of case studies has treated the subject (for a recent overview, see Koos 2015), quantitative research has been scarce so far (for notable, yet limited exceptions, see Butler et al. 2007; Cohen 2013; Cohen / Nordås 2015; Green 2006; Leiby 2011). The present study tries to further narrow this gap by conducting one of the most extensive quantitative analysis on state-perpetrated wartime sexual violence available so far. In total, I tested 14 hypotheses, juxtaposing common explanations with novel approaches.
I find little support for the most common theoretical explanations. For instance, the results suggest that wartime sexual violence is not more likely to be perpetrated by state actors in ethnically motivated conflicts, nor is it in secession wars. Equally, I find no support for the assumption, that wartime sexual violence is a direct reflection of broader societal gender-inequalities. Furthermore, I find little support for the assumption that capable state bureaucracies are able to limit the risk of wartime sexual violence by rogue state agents; however, vertical and horizontal military coherence do. Additionally, genocidal conflicts seem indeed prone to wartime sexual violence by armed state actors, especially if they are characterized by a high degree of lethal violence, underlining an existing nexus of sexual violence with other forms of lethal and non-lethal violence.
While measures of democracy are frequently included in existing statistical analyses on wartime rape and other lethal and non-lethal political violence, no formulation of a democratic peace theory of wartime sexual violence has been proposed yet. Indeed, the empirical test corroborates the assumption, that mature democracies successfully (but certainly not entirely) mitigate the risk of state agents engaging in wartime sexual violence.
Finally, I propose a signalling approach to wartime sexual violence. While assumptions on states using wartime sexual violence as a means of terror to deter civilians from defecting to the enemy found no support in the analysis, armed state actors seem indeed more likely to perpetrate sexual violence in situations of pronounced imbalances of power. One possible explanation points to incentives for armed actors to (mis-) represent their ability and intention to fight through such costly, and thus particularly credible forms of violence.
The paper provides three take-away messages:
Existing explanations of wartime sexual violence need further elaboration. Common explanations (especially the ones referring to ethnical and instrumental motivations) may be biased by particularly outstanding examples (e.g. the Yougoslav conflicts or Rwanda).
Democratic and military institutions and norms matter. Only effective norms and institutions that foster an understanding of sexual violence as counter-productive to any (strategic) end, seem to provide sufficient safeguards to prevent sexual violence in armed conflict.
Prevention matters. The findings indicate, that sexual violence systematically decreases the likelihood of long-lasting peace and becomes more likely in an environment of repeated acts of sexual violence perpetrated by several sides of the conflict.
This website provides the necessary data and code to reproduce the quantitative findings and to plot all the graphs (using R). To reproduce the findings, please download the dataset and the code files you are interested in. Alternativelly, you may download all files as a compressed file (~60Mb).