Historical Continuities

This brief history reveals a number of trends that are relevant for forming expectations about compliance with the law of war. First, the laws of war codified in the nineteenth century – and later extended in both substantive content and geographic breadth in the twentieth century – have antecedents hundreds of years earlier in the concrete customary traditions of European political units. Furthermore, these traditions were justified by the religious and political institutions of their eras, particularly in Christianity and the knightly institutions of the ancien régime. When modernity wrought significant changes to these institutions, the code of warfare was not thrown out but was merely adapted, often with very few changes, to the new context.Footnote ²⁴ Common to the rules of warfare in each of these periods, however, was the idea that they applied to some types of enemies but not others. Restrictions applied to fellow Romans and Greeks, respectively, in the classical period, to Christians and particularly fellow knights in the medieval period, to those of one’s own confession in the early modern period, and to other European states in the modern period.

Central to the development of a codified law of war was the recognition among European states that a fellow European state waging war against it was a justi hostes, a lawful enemy one would seek to defeat, but not to annihilate.Footnote ²⁵ However, enemies outside the European community of states were not granted such status and were often classified as savages, barbarians or semi-barbarians by European states: hostis injustus, against whom constraints on warfighting did not apply.Footnote ²⁶ As one historian noted, ‘The rules, objectives, and conduct of war were altogether different once civilization had been left behind’.Footnote ²⁷

An interesting and important coda to this detailing of the Eurocentric development of the law of war is the creation of the 1977 Additional Protocols to the 1949 Geneva Conventions. Unlike previous jus in bello treaties (including the 1949 Geneva Conventions), the Additional Protocols explicitly extend their application to wars of liberation from colonialism as well as, to some degree, civil wars in general (or, in the parlance of international law, ‘non-international armed conflicts’). For the first time, a large proportion of participants in the conference came from the non-European world.Footnote ²⁸ Importantly, Protocol I also contains the most explicit definition of ‘civilian’ in multilateral treaties to date (see Article 50).Footnote ²⁹ The clear extension of the law of war in this way may signal a change in the boundaries of civilization, most critically related to which targets are legitimate and which are not. Thus, as will be detailed further below, this article will seek to demonstrate the relevance of a particular cultural variable and assess its temporal limits.

Eastern Europe and Turkey

Just as the relationship of Eastern Europe to Western Europe continues to be controversial today, there are sound reasons to exclude these states from the European variable explored here. First, the Eastern European states as we know them today were generally not sovereign entities at the time of the initial codification of the law of war in the mid nineteenth-century, and thus did not participate as originators of the law of war. Secondly, there is significant empirical evidence that the Eastern European states were not perceived by their Western European counterparts as members of the European community of states.Footnote ⁴⁶ For instance, an observer of the customs of warfighting in Europe in the seventeenth and eighteenth centuries notes that constraints on warfighting only held insofar as ‘both sides held the same beliefs and agreed to abide by them. When the Teutonic Knights met the peoples of Eastern Europe […] the customary restraints vanished and the squabbles became bitter and vicious’.Footnote ⁴⁷ Because Eastern Europe’s exclusion from the Western European perception of the European community of states separates Eastern Europe from the mechanisms expected to result in increased law of war compliance among European states, the states of Eastern Europe are not coded as European.

Turkey likewise presents some challenges to our coding efforts. Though the Ottoman Empire was a consistent participant in early efforts to codify the law of war, it was not perceived by European states as a member of the European community, and thus we do not code Turkey as European. One scholar writes, ‘the Turks represented all that was negated in the European identity; savage, barbarian, despotic, oppressive, violent, and a threat to European civilization’.Footnote ⁴⁸ Turkey’s Islamic religious traditions undoubtedly contributed to its categorization as ‘Other’ by nineteenth-century Europe; indeed, debates about the extent of Turkey’s ‘Europeanness’ continue to this day.Footnote ⁴⁹

Russia

There is one notable exception to the exclusion of the states of Eastern Europe from the category of European: Russia. Though it is geographically considered part of Eastern Europe and Asia, its political activities during the time of the initial codification of the law of war, particularly its role in the Concert of Europe, provide strong grounds for including it in our category of European states. Russia is coded as European for the very reasons that the other Eastern European countries are not: it participated in the initial codification of the law of war and was perceived by the other leading European powers as being a member of the European community of states.

Russia emerged as a member of the European great power club during the anti-French coalitions of the late eighteenth and early nineteenth centuries.Footnote ⁵⁰ After the final defeat of France in the Napoleonic Wars, Russia joined Britain, Austria and Prussia in a series of bilateral agreements that confirmed each country’s commitment to preventing the re-emergence of the French threat and ensuring the general security of the European continent. The specific text of these treaties provides further indication that Russia was considered a European power; for instance, the Treaty of Chaumont, signed by Austria, Prussia and Russia on 9 March 1814, references ‘their respective States in Europe’.Footnote ⁵¹

In addition to Russia’s general membership in the European community of states, it played a critical role in the early formation of the law of war. An early signatory of the 1864 Geneva Convention, Russia took the lead in convening the 1868 St. Petersburg Convention, at which states agreed to the general principle that the means of war should not cause unnecessary suffering. It was also Russia’s initiative that led to the 1899 Hague Convention. On 24 August 1898, Nicholas II sent the famous ‘czar’s rescript’ suggesting the formation of an international conference to discuss disarmament.Footnote ⁵² While the czar’s goals for an armament agreement were not met, the conference did succeed in ratifying a code of warfare that forms the basis of contemporary jus in bello. Note that including Russia as a European state biases the data against our hypothesis, given that the historical record has established the extensive law of war violations that occurred along the Eastern Front in the Second World War.Footnote ⁵³

Australia, Canada, New Zealand and the United States

Another coding question concerns the status of states formed as settler colonies from European states. Australia, Canada, New Zealand and the United States are distinct from the majority of colonial territories in that the European (British) settlers of these territories became the majority populations of these areas. As a result, these states, particularly Canada, New Zealand and Australia, formed as ‘offshoots’ of Great Britain and maintained close cultural and political ties to the colonial homeland. Inclusion of the United States is further warranted because the United States and Europe exercised joint influence in the early development of the law of war. In 1863, the United States issued the Lieber code, an early statement of restraints on war that would prove foundational to the formation of positive law in Europe. Thus, these states strongly influenced (and were strongly influenced by) the European experiences of the early law of war. Likewise, because these states formed so directly from a European population, they were viewed by other European states, and especially by Great Britain, as extensions of the European community. Thus, from both general historical and European perceptions, these states clearly meet the criteria for European classification. However, they do fall short on the positive law criterion – they were not all, for example, initial signatories of the early law of war treaties – but this was because they remained under the sovereignty of the United Kingdom.

Civilizational Differences in Downes’ Targeting Civilians in War

Alexander Downes argues that wars characterized by territorial aims and attrition are especially likely to lead to civilian targeting. The logic of territorial conflicts can lead to civilian targeting when there is a perceived need to ‘cleanse’ a territory that one of the belligerents intends to resettle. Wars of attrition likewise produce civilian targeting as a measure of desperation; leaders may adopt civilian targeting as a calculated last resort, even as they recognize that it is often ineffective.

Downes’ excellent study captures key elements of the puzzle of civilian targeting. He includes an identity variable to test the hypothesis that ‘mistreatment of civilians is more likely to occur in conflicts in which the belligerents view each other as “barbaric”’.Footnote ⁵⁴ By adopting Huntington’s coding of civilizations to include a variable for whether the two parties at war cross civilizational lines, Downes seems to assume that all belligerents view opponents across civilizational lines as barbaric.

As illustrated by the models that include the variable Cultural Difference in Table 1, Downes consistently finds a negative, statistically insignificant result for this variable. When, however, we substitute our version of this variable – coding for whether a European state is fighting a non-European state (as well as other possible dyad combinations) – we generate the opposite finding. The European v. non-European variable is both positive and statistically significant. The substantive effect of this variable rivals or exceeds that of other key control variables. As shown in Table 2, when a European state fights a non-European state, the probability of civilian targeting increases by over 250 per cent. While the size of this effect remains an order of magnitude lower than Downes’ key explanatory variables of attrition warfare and annexationist war aims, the substantive effect of European v. non-European warfare is greater than, for example, those of reciprocity or relative capabilities. By contrast, Hypothesis 2 receives at best limited support from these results. Although the coefficient on European v. European dyads takes the correct (negative) sign, it is never statistically significant.

Table 1

Determinants of Civilian Victimization by Participants in Interstate Wars, 1816–2003 (Logit Estimates)

Note: replication and reanalysis of Downes (Reference Downes2008, Table 2.9), Models 1 and 2. ***p<0.001, **p<0.05, *p<0.1. Shaded row contains Downes’ original measure of identity. Bolded row contains our measure of identity.

Table 2

Change in the Expected Probability of Civilian Victimization Associated with each Independent Variable (Participants in Interstate Wars, 1816–2003)

Note: extension of replication and reanalysis of Downes (Reference Downes2008). Downes holds variables constant at their means. Because most of his variables are binary, we feel it is more appropriate to hold all binary variables at their modal values, and the one continuous variable (capability) at its mean. Compared to Downes’ Table 2.6, the marginal effects for Downes’ key variables appear much higher for this reason. Shaded row contains Downes’ original measure of identity. Bolded row contains our measure of identity.

Alternative specifications generate similar results that largely support our main claim. Including only one dummy variable that measures identity – for European v. non-European dyads – for example, consistently produces a positive and statistically significant coefficient. The analysis is also robust to dropping the world wars, but is not robust to dropping the United States as a belligerent. In the latter case, depending on the specification of the model, variables ranging from attrition to the dummy for European v. non-European dyads perfectly predicted civilian targeting.

Robustness checks employing alternative codings of European yield interesting results. For this data set, restricting the coding of ‘European’ to all states on the European continent,Footnote ⁵⁵ or to all Western European states, did not generate significant coefficients. But restricting the key independent variable to original signatories of the Second 1899 Hague Convention with Respect to the Laws and Customs of War on LandFootnote ⁵⁶ generated even stronger results on our key independent variable than our preferred coding of European, suggesting that a historical relationship with jus in bello may indeed be a key component of identity in predicting civilian targeting. We further explored this possibility in our test of Hypothesis 3 by interacting the post-1945 era (which largely lines up with the post-1949 Geneva Conventions era) as well as the post-1977 era (which lines up with the post-Additional Protocols era) with European v. non-European dyads, in the expectation that the effect of identity on civilian targeting would ameliorate over time, especially as more non-European states become party to the laws of war. The results offer only mixed support for this claim: compared to pre-1945 cases that include dyads other than European v. non-European belligerents, there may be a slight (although not statistically significant) negative effect on civilian targeting for European v. non-European dyads after 1945. Yet there are no cases of European states fighting non-European states that also target their opponents’ civilians after 1977. That the European v. non-European variable appears to exert such opposing effects in the pre-1977 and post-1977 periods suggests the possibility of a dramatic transformation in the relationship between European identity and compliance over the last several decades. Thus there is ample evidence here of the effect of this particular cultural variable, as well as evidence of possible temporal limits on this variable’s effects.

Race, Religion and Covenants

Benjamin Valentino, Paul Huth and various co-authors have offered an argument similar to Downes’ in their work on civilian targeting. Here we focus on their results in a 2006 World Politics article, co-authored with Sarah Croco, which contrasts the effects of the ratification of international humanitarian law on compliance with those of wars of attrition, adoption of counterinsurgency tactics and war aims. Like Downes, their analysis generates very strong findings in favor of the hypothesis that wars of attrition, counterinsurgencies and annexationism all increase the probability of civilian targeting. Ratification of international humanitarian laws such as the Geneva Conventions, however, does not seem to have the effect of reducing civilian targeting.

Valentino et al. also include a variable for identity conflicts as a control. The hypothesis they mean to test is that ‘States will intentionally kill greater numbers of enemy combatants during war when the majority of their population comprises different major racial or religious [sic] than does their adversary’s population’.Footnote ⁵⁷ They then code for whether ‘more than 50 per cent of the populations of each of the opposing combatant states were of different major races or religions’.Footnote ⁵⁸

In contrast to Downes’ results, Valentino et al.’s identity variable approaches statistical significance.Footnote ⁵⁹ Nonetheless, we argue that their measure of identity misses the mark. Theoretically speaking, and particularly when testing for compliance with jus in bello, we have stronger reasons to believe that a particular difference – European v. non-European – will have a greater effect than the general difference examined in Valentino et al.’s article.

Once again, our findings support Hypothesis 1. When substituting our (European v. non-European) version of the identity variable for that used by Valentino et al., we find a consistently positive and statistically significant effect of this particular configuration of identities on the number of civilians killed during war. As Table 3 illustrates, Valentino et al.’s findings on attrition, counterinsurgency and war aims remain robust. But as displayed in Table 4, European vs. non-European dyads are almost three times as likely to experience civilian targeting as non-European vs. non-European dyads, a substantive effect that is on par with those of relative capabilities and the size of the adversary’s population. Indeed, the number of civilians killed when Europeans fight non-Europeans is significantly higher than when Valentino et al.’s main strategy variable is activated. Also similar to the reanalysis of Downes, however, Hypothesis 2 is not supported by the results; not only is the coefficient on European v. European dyads never statistically significant, it also takes the wrong sign.

Table 3

Regression Models of Civilian Killings During Interstate Wars (1900–2003)

Note: replication and reanalysis of Valentino, Huth and Croco (Reference Valentino, Huth and Croco2006). N=148 for all models. ***p<0.001, **p<0.05, *p<0.1. Shaded row contains Valentino et al.’s original measure of identity. Bolded row contains our measure of identity.

Table 4

Marginal Effect of Variables on the Number of Civilians Killed During Interstate Wars (1900–2003)

Note: extension of replication and reanalysis of Valentino, Huth and Croco (Reference Valentino, Huth and Croco2006). Bolded row contains our measure of identity.

As with our reanalysis of Downes’ data, we ran several robustness tests to verify these results. Including only one identity dummy variable that singles out European v. non-European dyads yields a consistently significant and positive coefficient. Dropping the world wars and the cases in which the United States was a belligerent also produced stable results. The finding that European states are particularly likely to target civilians in non-European states is also robust to a number of specifications of European. Recoding European as a measure of whether a belligerent was located on the European continent consistently generated a significant and positive coefficient for European v. non-European dyads; this result was even stronger when the key independent variable was limited to coding only West European states as European. Recoding the variable to focus on original signatories to the second 1899 Hague Convention yielded similarly robust results.

Interestingly, we found stronger results for the possibility of a temporal effect when reanalyzing the Valentino et al. data compared to the Downes data. Consistent with Hypothesis 3, European v. non-European dyads post-1945 and post-1977 are approximately twenty-five times less likely to see civilian targeting compared to other types of dyads in the pre-1945 and pre-1977 eras.

It is also possible that identity indirectly affects compliance through the variables identified as most important by Valentino et al. For example, non-European states could be especially unlikely to ratify jus in bello treaties, and particularly likely to be the targets of counterinsurgency campaigns and territorial conflicts. Pairwise correlations between the identity variables and both legal and strategic variables produced intriguing, but somewhat inconsistent, results. The correlation between Western European states and ratifications of international humanitarian laws is negative and significant, a finding that runs counter to the logic stated above. European v. non-European dyads are also not necessarily more likely to experience wars of attrition or counterinsurgency, but they are significantly correlated with higher levels of attempted territorial annexation. These data do not provide strong support for the possibility of indirect relationships between identity and compliance mediated by international law or strategy.

Compliance with Jus in Bello

James Morrow’s work on compliance with jus in bello affords an opportunity to extend this analysis to issues beyond civilian targeting.Footnote ⁶⁰ Morrow examines the effects of reciprocal (such as whether the other side has violated jus in bello) and non-reciprocal (such as the regime type of the potentially violating state) variables on compliance with the laws of war regarding aerial bombardment, use of chemical and biological weapons, protection of cultural property, treatment of civilians, observance of armistices, conduct on the high seas, and treatment of prisoners of war and the wounded. He finds that joint ratification of relevant treaties and ratification by democracies are significant predictors of compliance. Including Morrow’s analysis in this article allows us to expand our empirical scope; in one model we restrict the analysis to civilian targeting, but the breadth of his data set invites theoretical and empirical consideration of whether our argument extends to other areas of jus in bello. In particular, we would expect that our argument about the role of identity could apply to the laws of war regarding the use of chemical and biological weapons as well as the treatment of cultural property and prisoners of war.

Unlike Valentino et al. and Downes, Morrow never considers a role for identity in predicting compliance with the laws of war. We suggest that including a variable noting whether the violator was a European state and the victim was a non-European state will also generate a significant effect on the probability of compliance.

The results presented in Tables 5 and 6 are consistent with our predictions. In Model 2, which replicates Morrow’s results but adds our variable for European v. non-European conflict (as well as other dyad configurations) and restricts the analysis to civilian targeting, we find that this particular configuration of identities generates a statistically significant increase in the probability of non-compliance with the laws of war regarding the treatment of civilians on the part of the European state in the dyad.

Table 5

Ordered Probit Analysis of Non-compliance with the Laws of War

Note: replication and reanalysis of Morrow (Reference Morrow2007, Table 1). ***p<0.001, **p<0.05, *p<0.1. Bolded row contains our measure of identity.

Table 6

Change in the Expected Probability of Non-compliance Associated with Each Independent Variable (Participants in Interstate Wars, 1900–2000)

Note: extension of replication and reanalysis of Morrow (Reference Morrow2007). Morrow reports marginal effects graphically. To ease comparison across studies, we instead ran simulations to estimate the effect of changes in the independent variables (all of which are binary) on the probability of reaching the greatest level of non-compliance. All binary variables were held to their modes. These results are based on Model 1a, but in addition to manipulating the variable of interest, we set Civilians to 1 to gauge the effect of changes in values when the issue area was civilian targeting. It is particularly interesting to note that the results on the European variable are the same as the results on Morrow’s key variable of Democracy×Joint Ratification. Bolded row contains our measure of identity.

When we expand our empirical scope to include all jus in bello issue areas as coded by Morrow,Footnote ⁶¹ we find that European states fighting non-European states are 50 per cent more likely to engage in major violations of jus in bello than are non-European states fighting each other. By contrast, we do not find support for Hypothesis 2; the coefficients are neither stable nor significant.

Various additional specifications of these models yielded stable results. Including only one identity variable – a binary variable for whether a European state was fighting a non-European state – produced positive and statistically significant coefficients on this variable. Results with respect to the identity variables were also stable when the world wars were dropped and when US wars were dropped (although in the latter case, the coefficient on identity only approached statistical significance). Robustness checks that used different codings of European are consistent with the findings using other data sets. Although the sign of the coefficient is stable when we alter the coding to include all states that are geographically part of Europe, the coefficient is no longer statistically significant, although it is close. Limiting the variable to all Western European states, or to original signatories of the second 1899 Hague Convention, generates significantly positive coefficients.

Interacting the European v. non-European identity variable with the post-1945 and post-1977 eras produced results that are only weakly consistent with the claim that the effects of identity may be mitigated over time. The interaction terms took the correct (negative) sign, but were not statistically significant. Interestingly, here the correlation between the ratification of international humanitarian law treaties and identity was positive and statistically significant, which contrasts with the analysis of the Valentino et al. data. This result is more consistent with the intuition that ratification itself could signify identity, thus opening up another pathway for identity to influence compliance with the law of war.

As stated above, we would expect our argument about the relationship between identity and compliance with jus in bello to apply beyond the issue area of civilian targeting, but with varying effects. More specifically, the same logic underlying the decision on the part of European states to target civilians in non-European states should also apply in their decision making with respect to the use of chemical and biological weapons as well as their treatment of cultural property and prisoners of war. When we restrict our analysis to these specific issue areas, we find that European v. non-European dyads behave as predicted.Footnote ⁶² This result lends breadth to our findings and suggests additional avenues for research.

Reconsidering Intra-European Compliance

The lack of support for Hypothesis 2 demonstrates the limits of identity in predicting civilian targeting and compliance with jus in bello, especially when compared with the strong support for the claim that European states are especially likely to target civilians and violate the law of war when fighting non-European states. Violation is common. Because compliance with the law of war is not typical of state behavior, we are not surprised that we do not observe strong compliance when European states fight each other. Of the scholars whose work we revisit in this article, only Morrow finds any significant predictors of restraint, and the conditions under which he expects to observe compliance are highly circumscribed. Moreover, it is important to note that many of the deadliest conflicts of the past two centuries took place within European dyads. Germany in particular was wont to target other European civilians in these conflicts, due to a logic that exported its colonial practices back to the continent.Footnote ⁶³ This points to a certain paradox of European identity. For most of Europe’s history, notions of community have coexisted alongside significant and regular bloodshed. For example, although there were strict codes of conduct during the Hundred Years War, these codes allowed for pillage, rape and the treatment of prisoners as property.Footnote ⁶⁴ Because wars between European dyads have historically been both frequent and bloody, to find that European dyads comply more than other dyads would come close to an assertion of greater baseline state compliance than is tenable given what we know about states’ wartime incentives. In other words, because we know state compliance with the law of war is not the norm, and because European dyads have historically engaged in particularly bloody wars, it is not particularly surprising that European dyads do not exhibit more restraint in their ‘internal’ wars. Indeed, the law of war was born of intra-European brutality and designed to regulate it. Instead, it appears that a major effect of these laws was to draw a line between European and non-European states across which egregious violations would be tolerated. That said, it would have been troubling to find that Europeans were significantly more likely to target civilians and violate jus in bello when fighting each other as compared to other dyad types, but this is not the case. Given that European states tend not to be particularly gentle with one another, it is highly significant that they are even more prone to violate the law of war when battling those outside the European community.