The author, Stathis N. Kalyvas (born 1964), is a political scientist known for his analysis of the dynamics of polarization and civil war, ethnic and non-ethnic violence, and the formation of cleavages and identities. He has also researched party politics and political institutions in Europe. He is a professor at Yale University.
Since at least the French revolution, we have collected a frightening register of a good amount of extremely violent events in the context of civil wars. And such apparent massive irrational behaviour among combatants and civilians (a sort of unexpected Hobbesian disease) has been widely examined by the literature. Mainly using anecdotal accounts or cleavage-based analyses, previous works tried to insert violence as a phenomenon directly derived from the (military) logic of civil war, but they overlooked the fact that several manifestations of violence can be exogenous to that logic. In fact, civil war can be an exogenous shock onto some societies that can activate their invisible networks of grievances and feuds among their individuals. Here it is the relevance of that off-the-path work.
Kalyvas’ book is focused in explaining this sort of violence common in civil wars, namely violence that is significant by its number of victims, its ‘barbarism’ (or brutality) and the fact that both victims and executioners have had a peaceful performance in their past interaction (i.e. neighbours). The research is tried to be confined inside that kind of violence, in civil wars, and the work principally provides an explanation to the spatial variance of that dependent variable. Temporal approaches are suggested, but not deeply developed in comparative terms.
As civil wars are frequently battled by means of some sort of irregular warfare by one or both contenders, Kalyvas embodies his explanation in the constraints of irregular wars, namely the ability of competitors to hide themselves behind civilian population, and the uncertainty around who is an enemy and who is neutral in such environment. Enemies can be hidden among the apparently supporters of a community. And contenders can only deal in an efficient way with such informational problems exercising violence against previously selected defectors. The efficient way of doing so is waiting for spontaneous civilian informers, and this is more critical in highly competed areas, where powers are balanced and, consequently, sovereignty is fragmented. On the other hand, the probability of denunciations grows with the level of control that one side has in a given area. The more the control they have, the less the risks of retaliation by the other side informers have to face, but also the less the probability of real defection shifting to the other side and, subsequently, the less the credibility of those numerous and spontaneous informers. Political actors do not search for violence as the first-best option to prevent both shifts and defections, because they generally prefer selective, limited violence (that reduces the probability of shifts and defections by deterrence) to arbitrary, massive violence (that increases the likelihood of those outcomes).
The mechanism follows a tree iterating game where political actors decide if they want to use violence, depending on the degree of control over that territory, but they confront an “identification problem”. Then, individuals decide if the benefits of denunciation (old feuds and grievances) are higher than the risks of doing so, given the level of control of the own side in the territory. A group of institutions to make denunciation easier and more anonymous are required to increase the likelihood of that event, but they perform worse in rural areas, where denouncers are more visible. So, if denunciation and violence take place, victim’s kin can retaliate or not, depending on the risks of doing so given the level of control of the own side in that given territory. When that level is balanced for both sides, the equilibria can be indiscriminate violence or not violence at all, due to mutual deterrence .
This approach to violence as a jointly process, where violence only takes place when civilians and political powers cooperate given their interests and cost-benefits calculi, lets the author to develop a model that predicts on one hand the likelihood of violence in a given area; on the other, it clarifies the moral hazard problem political contenders confront because the abundance of informers is inversely proportional to their usefulness, as we saw above. And, to sum up, that parsimonious model surprise readers because it is able to predict levels (variation) of violence from a micro approach.
To run a convincing empirical test of that model, the author collected regional data of the Greek Civil War, and next to two thirds of violence variation is successfully predicted by the model. And additionally, to test its validity outside the Greek sample, the model is successfully confronted with a large variety of historical and anecdotal accounts about civil wars around the world.
Those findings, both theoretical and empirical, have several significant implications. One of them is that violence in any civil war is a function of civilians’ previous feuds among them and the distribution and degree of control of the contenders over the territory in conflict. The former is given before the war, but it only takes the shape of violence jointly to the latter, namely when sovereignty is fragmented by competing powers.
These findings also explain why contenders will prefer to use selective to indiscriminate violence, and when the latter will take place. The behavioural microfoundations established in the game-theoretical model, and the principal-agent dilemmas the players confront constantly, let to this work to provide an answer to violent outcomes based on incentives and rational behaviour, instead of on gut reactions and irrational impulses.
Leaving apart those implications, the argument of the manuscript arises some questions that are partially answered by the author, specifically if the model is applicable both to ethnic and non-ethnic violence during civil wars, or if the model is efficient predicting the sort and intensity of violence when some complexity is added, namely more than two players (i.e. Colombian conflict), non-unitary players (adding some proxy of ‘internal cohesion’ to the apparatus of both contenders, since soldiers or militias are at the same time agents of one power and individuals with their own private feuds, grievances and family to defend; and they can commit violence or avoid it even when the rational strategy of the unitary contender is the opposite). Additionally, since the temporal dynamics of the civil war are just roughly outlined, one can ask if frontiers between the five defined categories of controlled areas would be blurred depending on the kind of military strategies that take place. Mutual blitzkriegs and positional wars in a vast, underpopulated territory can make very difficult for players to know the real degree of control that both powers have in any given territory, especially in those considered as “3” in Kalyvas’ categorization (see Kalyvas thelogicofviolencei.jpg). In other words, trench wars tend to create stable areas, and thus, concentrated violence in disputed regions. Conversely, military strategies that rely in quick movements and constant interchange of lands tend to spread violence across the territory.
Despite these minor considerations and opened questions, the work is a fruitful approach to violence in civil war. It unusually deals with violence as dependent variable, and it explains variance according to the degree of control that both contenders have over the territory, and how it conditions the behaviour of civilians and authorities. Combining a large variety of social sciences’ tools, from historical accounts to rational choice models, this work also provides a self-assessment around the strength of causality lines (especially a pair of successfully tested potential endogeneity sources), selection biases, explanatory limits and ways to generalize its conclusions.
All this makes this book hard to criticise in methodological terms, but at least one can point out some areas not deeply covered by this explanation. First those referred to conceptual doubts: how to distinguish between such variety of different kinds of violence, especially public versus private violence, since it is indirectly induced in the model (mainly via denunciations)?, or those doubts cited above referred to the features of the war (level of conventionality of the armies, goals that they follow, use of adscriptive discourse and signals), or a more complete explanation for shifts.
Secondly, there are no solid competing explanations for the failed predictions of that parsimonious model. The only reasonable way to increase the level of success predictions is adding complexity to the model, looking for additional independent variables, but perhaps this would rest attractiveness and academic efficiency to that contribution.
And, to finish, perhaps the work can integrate better in the model some significant variables such as elites’ manipulation to shape populations’ own beliefs and interests; as well as those strategies of ‘reputational attacks’, in which contenders pretend to act as their enemies while they commit acts of extreme violence against civilians, confusing the perceptions among population.