A. Survey Data

The data used in this study to evaluate the relationship between international law and individual attitudes come from a cross-national survey, “The People on War Report” (POW) conducted at the behest of the ICRC (1999). The ICRC commissioned the survey in 1999 to mark the fiftieth anniversary of the signing of the Geneva Conventions, and it was intended to gauge general attitudes toward the laws of war writ large and the conduct these rules were designed to regulate. Many surveys contain at most a handful of items related to wartime conduct, such as a number of polls looking at public support for the use of torture against detainees (Gronke et al. Reference Gronke, Rejali, Drenguisa, Hicksa, Millera and Nakayama2010). Other surveys examining a broader range of behavior tend to focus more on government repression than on the direct wielding of violence by belligerents in the midst of armed conflict, which can be perpetrated by either state or non-state actors (Huddy et al. Reference Huddy, Feldman, Taber and Lahav2005; Merolla and Zechmeister Reference Merolla and Zechmeister2009). By contrast, the ICRC POW survey is unique in its wide coverage of questions related to many aspects of violence, as well as the legal rules and principles intended to regulate and restrain such abuses.

Beyond a battery of items on attitudes toward wartime conduct, another strength of the ICRC survey is its cross-national coverage. While many studies are limited to a single country,Footnote ⁸ the ICRC survey provides a useful comparative element by including participants from eleven different countries that had recently experienced, or were still experiencing, armed conflict. The countries surveyed are Afghanistan, Bosnia-Herzegovina, Cambodia, El Salvador, Georgia and the breakaway region of Abkhazia, Israel and the Palestinian occupied territories, Lebanon, Nigeria, Philippines, Somalia, and South Africa.Footnote ⁹ These countries also offer variation along several relevant dimensions, including geographic region, ethnicity and religion, conflict duration and severity, whether fighting was still ongoing, and length of time since hostilities ceased.Footnote ¹⁰

The survey was designed and fielded by the ICRC in conjunction with the survey firm Greenberg Research. A stratified multistage cluster sampling method was employed. The sample was first stratified to obtain an equal representation of the main geographic locations and groups involved in each conflict. For instance, in the case of Georgia-Abkhazia, around five-hundred respondents were selected from Georgia and another five-hundred from the Abkhazia region. In Bosnia-Herzegovina, the sample contained approximately five-hundred Muslims, five-hundred Croats, and five-hundred Serbs. The number of respondents surveyed in each country thus ranged between approximately one-thousand and fifteen-hundred depending on the number of parties to the relevant conflict. Each subsample was further stratified by geographic region. To ensure a representative sample to the greatest extent possible, respondents were then randomly selected within the resulting strata.

One further question concerns the timing of the survey, which was initially fielded around two decades ago. In the intervening years, other notable developments that have taken place include the emerging norm of the responsibility to protect (RtoP) and the formation of the International Criminal Court (ICC). There are several reasons for which potential insights gleaned from the ICRC survey remain relevant, especially in the case of evaluating the relationship between IHL and attitudes toward wartime violence. Although the ICC did not officially begin functioning until 2002, the Rome Statute establishing the court opened for ratification in July 1998, well before the time the survey was fielded. Precedents over possible prosecutions for violations had also been in place beforehand, including the special tribunals investigating and trying crimes committed during previous conflicts in Rwanda and the former Yugoslavia.Footnote ¹¹ Similarly, the origins of the discourse surrounding RtoP developed in the middle and late 1990s in the aftermath of the Rwandan genocide and Balkan atrocities (Bellamy Reference Bellamy2009, 3–4), which also began before the survey was fielded. Although it is difficult to know a priori whether findings based on the ICRC survey would be stronger than, weaker than, or the same as similar polling conducted in the present day, the data nonetheless provide a useful reference point for thinking about the relationship between international law and individual attitudes toward wartime violence.Footnote ¹²

Of course, the ideal would be a cross-national panel survey of most countries of the world conducted across multiple waves in the manner of the World Values Survey, or similar regional projects such as the Afro- and Eurobarometers.Footnote ¹³ Yet the sheer breadth of coverage of questions asked about both wartime conduct and IHL makes the ICRC survey the most promising resource currently available for examining the sources of public attitudes toward the use of violence across a number of countries. The ICRC survey has also been successfully employed in other scholarly research looking into related questions on wartime violence. For instance, Meernik and King (Reference Meernik and King2014) examine differences in public support for transitional justice mechanisms, such as trials and truth commissions. Spini, Elcheroth, and Fasel (Reference Spini, Elcheroth and Fasel2008) look at attitudes toward wartime violence using a more limited set of items on support for abuse, though focusing specifically on the role of individual, group, and national context with little attention to the role of international law. Elcheroth (Reference Elcheroth2006) similarly looks at attitudes toward wartime violence using a more limited set of indicators, but he centers more on the role of individual and community-level trauma, while combining understandings of international law with support for abuse. My study thus seeks to build on this prior work by specifically examining the ways in which knowledge of IHL may directly influence attitudes toward wartime violence using a more systematic and disaggregated conceptualization of abuse.

The survey instrument contained a core set of questions measuring attitudes toward a wide range of types of wartime violence, as well as a variety of items dealing with individual attributes and experiences. Following common procedures used in many other cross-national surveys, question wording was sometimes modified slightly to ensure consistency in meaning. A full description of the sampling methods and questionnaire can be found in the final report for the survey (ICRC 1999, 87–88). The organization’s report also gave some indication of a possible relationship between a greater understanding of IHL and more restrained attitudes to what should be permitted during warfare (ICRC 1999, 19–20). Yet the original report’s discussion was in many respects preliminary and suggestive, given that it looked at only a few select abuses in isolation and did not take into account alternative factors or potential threats to validity. Most importantly, there still remains little sense in the existing literature of the relative explanatory power of specific mechanisms accounting for how awareness of international law may relate to individual attitudes toward abuses. The research design and analysis that follows thus seeks to address each of these issues to provide a fuller account of the relationship between international law and support for wartime violence.

B. Dependent Variable—Attitudes Toward Wartime Abuse

The main dependent variable involves the construction of a series of indices measuring each respondent’s general attitudes toward wartime abuse. The indices were created using a principal factor analysis of respondents’ answers to twelve separate items detailing various types of violations. Factor analysis allows for uncovering a common dimension, or set of dimensions, underlying a given set of variables (Kim and Mueller Reference Kim and Mueller1978b, 9–10). The component items fall into two main categories based on the identity of the victimized group—civilians or prisoners.Footnote ¹⁴ Table 1 summarizes the full set of items, which were rescaled so that higher values indicate greater levels of support for abuse. The six civilian measures focus on preferences toward attacks against civilian versus military targets, as well as strategies that disproportionately harm noncombatants, such as the use of land mines. For instance, the phrasing for the first civilian item is as follows: “When combatants attack to weaken the enemy, should they ….” Respondents were then given the following options to complete the sentence in declining order of willingness to accept violence against civilians: “Attack enemy combatants and civilians” / “Attack enemy combatants and avoid civilians as much as possible” / “Attack only enemy combatants and leave the civilians alone.”

Table 1.

Questions Used to Construct Support for Wartime Abuse Indices

The six component prisoner abuse measures correspondingly cover a variety of common violations associated with the capture and detention of enemy combatants, ranging from torture to denying access to outside monitors. For example, in the first prisoner item listed in Table 1, respondents were asked: “Would you save the life of a surrendering enemy combatant who killed a person close to you?” Respondents could then answer either yes, they “Would save,” or no, they “Would not save.” Importantly, across all of the items used to construct the indices, the phrasing does not make any direct reference to international law, but focuses on the nature of the physical acts directed toward either civilians or prisoners. The resulting outcome variables thus seek to capture general attitudes toward wartime violence, while remaining distinct from the main explanatory variable centered on individual knowledge of IHL. The full text for these and other relevant questions is included in Supplementary Appendix A.

Analyzing these twelve items revealed one common underlying factor, having an eigenvalue of 2.04. Additional tests confirmed that a single dimension of support for wartime violence best described the relationship among the component variables.Footnote ¹⁵ The loadings generated from the factor analysis were then used to create an index of general attitudes toward wartime violence, Overall Abuse. To facilitate subsequent estimation and interpretation, the measure was rescaled to have a range between 0 and 1, where higher values indicate greater support for wartime violence.Footnote ¹⁶ Because past research suggests there may be important differences in the treatment of various groups during war (Morrow Reference Morrow2007; Downes Reference Downes2008; Wallace Reference Wallace2012), separate indices were also created for violence toward civilians and prisoners. Civilian Abuse and Prisoner Abuse are corresponding indices generated by separate factor analyses on the six specific items for each type of abuse, respectively, and their scales were then normalized in a similar manner.

Figure 1 presents a histogram of the distribution of values across the three indices, overlaid with a smoothed kernel density plot. Figure 1a indicates that attitudes are weighted toward the lower end of support for wartime violence, which should offer some relief to activists and policymakers worried over the general hardening of attitudes that can occur in conflict and post-conflict societies (Kaufmann Reference Kaufmann1996). Nevertheless, only around 10 percent of respondents professed the lowest levels of objection to the resort to abuse, meaning that the vast majority of those surveyed felt at least some situations were warranted where protections could be denied to civilians and captured combatants. When distinguishing between the particular victim of abuse, lower end values are more prevalent for the separate civilian and prisoner indices. Figures 1b and 1c also show that attitudes toward prisoner abuse tend to be more concentrated, not only at the lower end of the scale, but also sometimes at the upper ends, while opinions toward the treatment of civilians are relatively more dispersed across the full range of values.

Figure 1.

Summary of Support for Wartime Abuse Indices

Note: Histogram for each relevant abuse index is overlaid with a kernel density plot (black line).

Those respondents indicating the greatest levels of abhorrence for wartime abuse are not necessarily the same when it comes to the treatment of civilians versus combatants. Although the two victim-specific indices are positively correlated overall (r = 0.39), they are actually negatively related (r = –0.29) at the lower ends for values less than or equal to 0.1 on the normalized scale. This pattern points to the importance of considering the possibility that the correlates of support for wartime violence may differ along certain dimensions depending on the identity of the prospective victim. This localized negative relationship between the two indices also helps account for why the proportion of cases falling at the lower ends is more pronounced for each index individually in Figures 1b and 1c than in the overall abuse index in Figure 1a.

Figure 2 offers a better sense of patterns in support for different forms of wartime violence by plotting the cumulative distribution function for each index. While all three indices follow a similar pattern, opposition to wartime violence actually seems slightly more noticeable for abuses directed toward prisoners than toward civilians. This may seem surprising given the general emphasis in much of the discourse on the rights of civilians, along with the lesser threat or blame generally attributed to noncombatants that could be used to justify violence. However, the different distributions may partly be a function of the constituent items used to construct each index—several of the prisoner questions were much more direct, such as torturing or killing combatants, while a number of the civilian items involved less direct actions (denying food, planting land mines) or emphasized “attacking” rather than the more charged language of “killing.” It is thus difficult to compare the absolute levels across the various abuse indices directly. Of course, it might be preferable if the questions asked were exactly the same except for the group being victimized. However, closer congruence would come at the cost of ignoring the most relevant types of violence directed against civilians versus prisoners, which often differ importantly in numerous respects (Morrow and Jo Reference Morrow and Jo2006). Ultimately, for the purposes of this study, overall levels of support for various forms of wartime violence are generally less central than are differences in individual attitudes based on key attributes of respondents, in particular their understanding of IHL.

Figure 2.

Cumulative Distribution Functions of Support for Wartime Abuse Indices

C. Main Explanatory Variable—Knowledge of IHL

The main independent variable concerns individuals’ knowledge of IHL and the consequent limits these rules seek to place on the resort to abuse in times of armed conflict. While some experimental studies have used some form of prompt stating the illegality of a given violation to proxy for exposure to international law, the POW survey offers a more direct measure of individuals’ actual knowledge of IHL, which is intended to be separate from their policy preferences.

The IHL regime complex has expanded greatly over the last century to regulate a wide range of behavior.Footnote ¹⁷ The 1949 Geneva Conventions nonetheless remain the foundation for much of contemporary IHL (Best Reference Best1994, 136; Evangelista and Tannenwald Reference Evangelista and Tannenwald2017). While mainly addressing international armed conflicts, Common Article 3 of the conventions also guarantees a fundamental set of protections for victims in civil wars, or “conflicts not of an international character” (Solis Reference Solis2010, 96–104). Protections afforded under the Geneva Conventions to civilians and prisoners, among others, codified many prior rules and customs, while continuing to serve as a basis for later agreements, including the conventions’ two Additional Protocols of 1977 and the 1998 Rome Statute establishing the ICC (Schabas Reference Schabas2011, 131–33). As such, understanding the conventions introduces individuals to basic principles of restraint and humanity at odds with the violations at the core of the various abuse outcome indices.

To capture exposure to IHL, I rely on a pair of questions posed to all respondents in the survey. Respondents were first asked: “Have you ever heard of the Geneva Conventions?” If they answered affirmatively, this was followed by an open-ended question: “Could you tell me what the Geneva Conventions are about?” Enumerators for the survey then recorded whether these respondents accurately identified the main purpose of the conventions based on mentioning any reference to the placement of limits on wartime conduct. I then used these items to construct the dichotomous variable, Know About Geneva Conventions, which equals 1 if a respondent had heard of the conventions and correctly identified their purpose, and 0 if respondents had either not heard of the conventions or incorrectly described what the laws were about. This variable exhibits clear variation, as around 25 percent of respondents demonstrated a general understanding of the Geneva Conventions, suggesting exposure to basic IHL principles is evident but far from universal.Footnote ¹⁸

As a measure of IHL knowledge, the variable admittedly is unable to make fine-grained distinctions in a person’s relative level of knowledge or awareness of particular rules of warfare. Nevertheless, the measure offers a number of distinct advantages. First, the second item is open-ended, requiring respondents to provide an answer in their own words, making guessing and false positives more difficult and less of a worry than a closed-ended format. Second, the generality of the construct is beneficial as it is not directly linked to specific items gauging preferences for or against abuse, mitigating concerns over consistency bias between responses. Third, the two Geneva Conventions questions were asked well after the key items used to construct the abuse indices outcome variables, helping to minimize worries over question ordering and contamination across responses.Footnote ¹⁹

D. Additional Explanatory Variables

A number of other variables are also included that past research suggests should have some bearing on individual support for wartime abuse. The first category of variables concerns the experiences of respondents during the conflict given that such events can fundamentally change their political attitudes (Blattman Reference Blattman2009). Lived in warzone is a dichotomous variable that equals 1 if fighting took place in the area where the respondent lived, and 0 otherwise.Footnote ²⁰ In addition to their general proximity to the fighting, the next two variables capture respondents’ personal experiences during the conflict. Although suffering any form of abuse or deprivation is deplorable, these variables seek to distinguish between severe and more moderate forms of violence that respondents may have encountered, as each type may have different implications for their attitudes (Balcells Reference Balcells2012). Respondents were presented with a list of twelve wartime experiences that might have happened to them. More severe experiences are those that involved the infliction or grave threat of direct physical violence, while experiences considered more moderate included indirect deprivations or where violence was experienced secondhand against a close relation.

Severe wartime experience is a dichotomous variable that equals 1 if the respondent reported experiencing any of the following acts—being imprisoned, kidnapped, or taken hostage; tortured; or wounded by the fighting; and 0 otherwise. Moderate wartime experience is a corresponding dichotomous variable that equals 1 if the respondent instead reported experiencing any of the following—felt humiliated, lost contact with a close relative, property was seriously damaged, food was taken away, house was looted, forced from home, an immediate family member was killed, or a person they knew well was raped by combatants, and 0 otherwise. This variable was further coded 0 for those respondents who reported that one or more of the severe acts had also happened to them to ensure the two wartime experience variables do not overlap.

While the previous three variables capture acts or situations encountered by respondents, the next two variables examine the nature of an individual’s contributions and actions taken during the war. Took sides is a dichotomous variable that equals 1 if the respondent actively supported one of the warring parties during the war, and 0 otherwise. Unfortunately, the question does not provide an opportunity to distinguish between supporters of the government versus opposition forces. Using religion or ethnicity as a rough proxy for support is inappropriate, since it is not uncommon for co-ethnics to choose or fight on opposing sides (Kalyvas Reference Kalyvas2008; Lyall Reference Lyall2010), not to mention that the ethnic identity measures in the ICRC survey suffer from a number of problems (Spini, Elcheroth, and Fasel Reference Spini, Elcheroth and Fasel2008). Nevertheless, the variable provides a reasonable measure of whether individuals who are more politically engaged in the conflict may be more or less likely to support the use of violence (Balcells Reference Balcells2010). To assess whether those with past military experience possess different attitudes (Feaver and Kohn Reference Feaver and Kohn2001), Veteran is a dichotomous variable that equals 1 if the respondent took up arms, and 0 if he or she remained a noncombatant throughout the conflict.

The second set of indicators covers several socio-demographic characteristics that tend to be associated with a variety of political attitudes (Holsti Reference Holsti2004). Age is a categorical variable that separates respondents into a series of age ranges.Footnote ²¹ Gender is a dichotomous variable that equals 1 if a respondent is male, and 0 if female. Has children is a dichotomous variable that equals 1 if a respondent has children, and 0 otherwise.

Education is a particularly important variable for evaluating the theoretical argument put forward. Higher levels of education have been associated with support for more liberal policies, such as free trade and immigration (Hainmueller and Hiscox Reference Hainmueller and Hiscox2006, Reference Hainmueller and Hiscox2007). On the other hand, the role of education in armed conflict and political violence is much less settled (Thyne Reference Thyne2006; Brockhoff, Krieger, and Meierrieks Reference Brockhoff, Krieger and Meierrieks2015). If respondents who know more about the Geneva Conventions also tend to be the most educated, then any observed relationship between attitudes toward violence and awareness of IHL may simply be a function of educational attainment. The survey recorded the number of years of schooling attained by each respondent. Countries often have different national systems for primary school, secondary school, and postsecondary education, which makes a single common educational classification system problematic. To rectify these concerns and still allow a comparable cross-national measure, Education is a four-level categorical variable that divides the full sample of respondents into quartiles based on the number of years of schooling completed.Footnote ²²

Along with these socio-demographic items, the survey also included an experimental element in which respondents were presented with slightly different wordings for a number of items, including some of the abuse questions, such as a generic scenario versus a partisan situation involving combatants from the respondent’s own side.Footnote ²³ Partisan wording is a dichotomous variable that equals 1 if the respondent received the more partisan scenario for relevant questions, and 0 if he or she instead received the more nonpartisan wording. While there are differences in overall responses for some items based on the particular version asked, they are not always in the same direction—in some cases, the partisan version is formatted to elicit responses more supportive of abuses while in others it leads to greater opposition. Additionally, the relative distribution of values across the various indices do not differ markedly when subset by either of the two experimental groupings, suggesting that the different wordings should not overly affect the results of interest for this study.Footnote ²⁴ Nevertheless, the variable is included to take into account that those receiving the partisan wording may differ in their responses. Table 2 provides summary statistics for the main variables used in the analysis that follows.Footnote ²⁵

Table 2.

Summary Statistics for Main Variables Used in the Analysis

A. Overall Findings

Because the indices forming the main dependent variables dealing with attitudes toward wartime violence are continuous, regression models were estimated using ordinary least squares (OLS). To control for any country-specific effects, separate dichotomous variables were included for each country in the survey.Footnote ²⁶ Standard errors were clustered by country to take into account that observations are unlikely to be independent within each country.

Figure 3 reports results from a series of OLS models estimating individual attitudes toward the use of violence across each of the three indices—overall, civilian, and prisoner abuse. The figure plots coefficients for the main variables of interest across each abuse index, with horizontal lines indicating 95 percent confidence intervals.Footnote ²⁷ Since each index was normalized to values ranging between 0 and 1, interpreting the coefficients is relatively straightforward. Positive (negative) coefficients indicate a given percentage point increase (decrease) in the level of support for abuse. Supplementary Appendix C presents more complete regression results from this figure as well as for the subsequent analyses discussed in further detail below.

Figure 3.

Substantive Effects of Support for Wartime Abuse

Note: Coefficient plot (OLS) indicating percentage point change in support for each abuse index. Lines indicate 95 percent confidence intervals. Results for partisan wording and country variables not shown.

To give a better initial sense of raw levels of support, I calculated the predicted absolute values for each abuse index, while holding all explanatory variables constant at their median values.Footnote ²⁸ The resulting value for overall abuse was 0.23, which is generally in line with the patterns reported in Figure 1, where respondents on average demonstrate a relatively low propensity to countenance abuse, but they are nonetheless willing to condone violations in a variety of circumstances. Based on their corresponding models, the values were 0.18 for civilian abuse and 0.27 for prisoner abuse. While this may seem to reflect a lower willingness in general to support wartime violence against civilians than against prisoners once other factors are taken into account, as noted earlier, any direct comparisons can be suggestive only given the difference in question wordings and the normalized scales across the two victim-specific abuse indices.

Figure 3 instead focuses on the first differences associated with each independent variable, since our main quantity of interest is the relationship between knowledge of international law and support for wartime abuse. The results show that prior awareness of IHL is associated with greater restraint toward violence, though this relationship is not consistent across all forms of abuse. Looking at the combined overall abuse index, individuals who have prior knowledge of the Geneva Conventions are 3 percentage points less supportive of the resort to violence, a modest though statistically significant finding. Importantly, this result holds even when controlling for educational attainment, suggesting that IHL knowledge is not simply a function of those respondents who happen to have more schooling and, presumably, a greater level of general knowledge.Footnote ²⁹ By contrast, in the specific case of violence against civilians, the association with IHL knowledge disappears; the coefficient is still negative but greatly diminishes in size and statistical significance. The initial results for the overall abuse index appear instead largely driven by attitudes toward the treatment of enemy combatants. Individuals with greater IHL knowledge are more than 5 percentage points less supportive of abuse toward prisoners. This same pattern holds across various alternative specifications of explanatory variables, as well as combinations of the component items used to construct the abuse indices.Footnote ³⁰

The consistent non-finding for the association between IHL knowledge and civilian abuse does not necessarily imply that respondents do not care about the plight of noncombatants. These weaker results may instead actually point to the possibility of a more widespread acceptance of underlying norms of civilian protections, as suggested in the absolute values noted above, and along the lines enunciated in what later became known as the RtoP norm. Of course, individuals may still condone the targeting of civilians for strategic reasons, even while acknowledging that limits likely constrain such actions in principle. The discrimination between combatants and noncombatants and, by extension, the protection of the innocent remains one of the foundational tenets of just war theory (Walzer Reference Walzer2000, 151), and it may thus be more commonly accepted as such in mass opinion. Having more specific IHL knowledge may bring little new to the minds of individuals in conflict situations and, in turn, have little discernable ability to change their attitudes toward civilian targeting.

By contrast, the appropriate treatment of prisoners may be less settled given the uneasy status of combatants who had been fighting, or potentially even perpetrating abuses, not long before capture (Kinsella Reference Kinsella2005; Wallace Reference Wallace2015, 42–45). The War on Terror highlights how states can treat civilians relatively humanely, while simultaneously brutalizing enemy combatants (Kahl Reference Kahl2007, 8 n3). To the extent that the Geneva Conventions and similar laws can disseminate new beliefs about appropriate treatment of detainees, or bolster existing suppositions with greater legitimacy, it follows that IHL may exert greater restraint on attitudes toward prisoner abuse. Of course, it is important to note that this interpretation of the different results for civilian versus prisoner abuse is only suggestive. The available data unfortunately do not provide the ability to conclude definitively that individuals’ baseline understandings of varying forms of abuse are driving the results.

Although differences in support for both overall abuse and prisoner abuse associated with greater knowledge of IHL are modest, they are of comparable size to those found in related experimental studies (Wallace Reference Wallace2013; Chilton Reference Chilton2015). Caution must also be exercised before drawing any conclusions of a causal effect attributable to international law based on the observational data used in this study; a point discussed in greater detail below. It is nonetheless promising that the size of the observed differences in attitudes toward wartime violence based on a real-world measure of exposure to the laws of war closely tracks the causal estimates gleaned from prior experimental studies, which exhibit high levels of internal validity.

Looked at from a relative standpoint, the coefficient for knowledge about the Geneva Conventions is also at least of equal or greater size to the other explanatory variables measuring many other common explanations for attitudes toward violence. In fact, in the prisoner abuse model, IHL knowledge represents the single largest coefficient, with a greater associated change in attitudes toward wartime violence than direct personal experiences, gender, or being a veteran. The substantive importance of prior awareness of international law is thus in line with existing research and is impressive when compared to alternative factors.Footnote ³¹

Turning to the other explanatory variables, many are generally consistent with past research. For wartime experiences, only the gravest types of past events have any systematic relationship with support for abuse or similarly harsher viewpoints (Balcells Reference Balcells2012). Living in the midst of fighting has no discernable association on its own with an individual’s willingness to condone or condemn wartime abuses. More moderate (though nonetheless serious) experiences, such as having one’s house looted or losing contact with close relatives, similarly appears to have little bearing on attitudes toward appropriate conduct. On the other hand, severe wartime experiences like being wounded or tortured are positively and significantly associated with greater levels of approval for all three of the abuse outcomes. The finding thus confirms that exposure to violence can breed an escalating cycle of abuse, as those who are victimized are more likely to support similarly brutal actions in the future (Elcheroth Reference Elcheroth2006).

Partisans are also more likely to approve of greater levels of violence, suggesting that those who are more supportive of armed groups are also more willing to condone the resort to abuse. Similarly, those who actually took up arms were generally more supportive of abuse, which is consistent with prior research showing the greater proclivities of those with military experience in approving higher levels of violence (Feaver and Gelpi Reference Feaver and Gelpi2004, 60). Interestingly, veterans are no more likely to support the abuse of captured combatants, perhaps suggesting that the prospect of retaliation may make soldiers more hesitant to support violations against their enemy counterparts in ways that are more salient than when the victims would instead be civilians (Axelrod Reference Axelrod1984, 73–76).

Looking at the remaining socio-demographic control variables, those who are older are generally less likely to support violence, though the coefficient does not attain statistical significance in the case of civilian abuse. Given past research pointing out that countries with demographic shifts experiencing so-called youth bulges are more conflict prone (Urdal Reference Urdal2006), the results suggest that younger populations may also be more supportive of higher levels of violence in the event of war. Unlike other research pointing to a fairly robust relationship between gender and the use of force (Eichenberg Reference Eichenberg2003; Brooks and Valentino Reference Brooks and Valentino2011), there are less definitive differences between men and women when it comes to explaining attitudes toward wartime conduct, with the exception of civilian abuse. Education similarly shows few consistent findings—the coefficient is negative and significant for prisoner abuse as expected, but it fails to achieve statistical significance in the other models and it is actually slightly positive for civilian abuse. Although education may be a strong predictor of more liberal attitudes for a number of other foreign and domestic policies, this does not appear to be the case for wartime conduct. On the other hand, having children does appear to be associated with a small though persistent decline in approval of wartime violence, which corresponds to evidence examining general attitudes toward the use of force (Smith Reference Smith2005; Horowitz and Levendusky Reference Horowitz and Levendusky2011). Perhaps because of concerns over possible future negative consequences for their children and later generations, respondents who are parents appear less likely to condone the resort to extreme forms of violence during wartime.Footnote ³²

B. Who Might the Law Influence Most? Assessing Conditional Relationships

One potential issue with the findings discussed so far is that the relationship between IHL knowledge and support for wartime violence may be masking important conditional relationships, or lack thereof, among specific groups of respondents. This may be of particular concern if the role of IHL is greatest simply for those who already espouse fairly humanitarian ideals, while having little influence on those most skeptical of such constraints. International law might be succeeding in moving attitudes, but the effect may be limited to likeminded individuals, while playing no role for harder cases. Indeed, one of the most frequent charges against international law is that such rules and principles only work in easier situations, or when actors are already positively predisposed, but they fail under more trying circumstances (Mearsheimer Reference Mearsheimer1994–1995).

To evaluate this possibility, I conducted a series of analyses distinguishing individual respondents by key traits commonly associated with different baseline attitudes toward wartime violence. Each model was estimated with the same general specification of explanatory variables as in the original analysis, but it was then limited to respondents from the relevant subsample—for instance, civilians versus veterans.Footnote ³³ Figure 4 summarizes the results of these subsample models across four covariates—civilian-veteran status, type of wartime experience, whether the respondent took sides during the war, and gender. Although each of these covariates finds some (though varied) support for overall differences in attitudes toward wartime abuse, past research also suggests differences in terms of sensitivity to international legal appeals (Brooks and Valentino Reference Brooks and Valentino2011; Meernik and King Reference Meernik and King2014; Wallace Reference Wallace2014).

Figure 4.

Knowledge of Geneva Conventions and Support for Wartime Abuse

Note: Plot for Knowledge of Geneva Conventions coefficient (OLS) indicating percentage point change in support for each abuse index by relevant subgroup. Lines indicate 95 percent confidence intervals.

Figure 4 reports the coefficient (measuring the percentage point change in support for abuse) for knowledge of the Geneva Conventions for the relevant subsample, while holding all other explanatory variables constant. The subfigures suggest that the constraining relationship between international law and support for wartime abuse is not only limited to those already predisposed to international legal principles, but also evident even for those who at first glance may not be very welcoming of humanitarian appeals. If anything, the results are slightly stronger for seemingly harder cases like veterans and those with more severe wartime experiences, though differences between subgroups sometimes do not attain standard levels of statistical significance.Footnote ³⁴ This is especially apparent for the civilian abuse measure, which is negative for veterans, severe wartime experiences, and those who took sides, while actually slightly positive for their counterparts (civilians, individuals with moderate or no serious wartime experiences, and those who did not choose sides). Consistency in the role of international law across certain groups is evident in Figure 4d, which shows similar results across both men and women. Although the differences between the two subgroups are not statistically significant, they demonstrate that the role of IHL knowledge is comparable, and potentially even more pronounced, across subpopulations that may approach both violence and legal principles in very different ways.

C. Discussion and Implications

Although the results point to an understanding of IHL being associated with a decline in support for wartime violence, a number of concerns should be kept in mind. Perhaps most importantly, in the preceding analysis I have been careful not to assert that a definitive causal relationship exists between knowledge of IHL and attitudes toward wartime violence. Assessing causality on the basis of observational data is inherently problematic and suffers from a number of threats to validity (Winship and Morgan Reference Winship and Morgan1999)—several of which I discuss in turn.

First is the possibility of reverse causality, where those who already possess more humanitarian inclinations are more likely to seek out sources of rules and principles that are consistent with their beliefs. Rather than international law playing any independent role in shaping individual preferences, knowing about these rules may simply be a function of preexisting attitudes toward appropriate wartime conduct. Although by no means definitive, the conditional relationships shown in Figure 4 mitigate some of the worst worries that reverse causality is driving the results. If prior preferences were decisive, the role of international law should have been especially prominent for subgroups tending to be approving of more benevolent practices in general. Despite veterans, partisans, and those experiencing traumatic wartime events exhibiting higher overall support for violent conduct, the association with IHL knowledge in these subgroups was of similar magnitude as that of their more humanitarian counterparts.

This points to a second and more general inferential concern over the nonrandom nature of the key explanatory variable in the study—knowledge of IHL. Unlike in an experimental setting, respondents were not randomly assigned to whether or not they were exposed to the key principles governing the laws of war. While the ICRC survey may offer a novel and effective measure for individuals’ actual IHL knowledge, those respondents who gained knowledge about the laws of war may differ in important ways from those with little or no awareness of international legal principles. Especially problematic is whether those individuals with a greater propensity to be aware of IHL also have systematically different preexisting attitudes toward wartime violence. This suggests an individual-level corollary to the larger problem plaguing many observational studies of international law. It can be difficult in such studies to discern whether international legal principles are actually exerting any independent effect, or whether some other confounding factor is explaining both commitment to international law and the outcome behavior the agreement is intended to regulate (Simmons and Hopkins Reference Simmons and Hopkins2005; von Stein Reference Von Stein2005). In the case of the survey data analyzed here, perhaps there is some unmeasured individual factor, such as pacifism, accounting for both lower support for wartime abuses and a person’s greater desire to seek out and understand more about the very laws intended to regulate such violence.

Although the evidence reported here may only indicate a correlation between international law and policy attitudes, as noted earlier, the results track closely with the findings of a number of past survey experiments that have demonstrated with greater confidence a causal relationship, whereby exposure to an international legal treatment prompt affects public opinion toward controversial policies (Tomz Reference Tomz2008; Wallace Reference Wallace2013; Chilton Reference Chilton2015). Moreover, further experimental work shows that the effects of international law may actually be weakest for those with strong policy preferences (Chaudoin Reference Chaudoin2014), suggesting that the current results may, if anything, underestimate the importance of international law.

The conditional relationships from Figure 4 continue to offer further insight into whether some hidden confounder, such as pacifism, is explaining the observed results. To the extent that pacifism is interrelated with other types of traits, such as gender, civilian-veteran status, or being a partisan during the war, it would be expected that the role of international law in support for abuses should be significantly different for these subgroups. However, the results from Figure 4 indicated relatively similar coefficients for knowledge of international law between these different subgroups.

Additional statistical analyses also suggest unobserved confounding is unlikely to be driving the main results. Matching has been increasingly offered as an approach to lessen the risk that any observed differences in outcomes between a treatment and control group is a function of nonrandom assignment to the treatment (in this case, knowledge of the Geneva Conventions). Matching pairs a “treated” observation (here, someone who knows about the Geneva Conventions) with an observationally similar “control” unit (someone who does not know about the Geneva Conventions) based on a defined set of covariates. The procedure generates more evenly balanced data that minimize observed differences on the covariates between the treatment and control groups (Ho et al. Reference Ho, Imai, King and Stuart2007). This helps increase confidence in an observed relationship between a treatment and an outcome—in this case, between knowledge of international law and support for wartime violence.

I employed a genetic matching routine using all the main explanatory survey variables and country indicators as the baseline covariates. Genetic matching is a multivariate matching method designed to achieve maximum balance in the covariates between treated and control groups (Diamond and Sekhon Reference Diamond and Sekhon2013). Although not reported here, the results using matching are not substantially different from those reported earlier based on more conventional OLS regression models.Footnote ³⁵ Of course, the reliability of any matching findings depends on the assumption that assignment to the treatment group (knowledge of the Geneva Conventions) is captured by the observed covariates (Rosenbaum Reference Rosenbaum2010, 76–79). However, further sensitivity tests indicate that the results for attitudes toward abuse are robust to the potential exclusion of a fairly strong unobserved confounder that could account for people’s propensity to know about international law.Footnote ³⁶

Third, and building in a more specific direction off the prior point, the positive finding for awareness of IHL may simply reflect respondents’ general levels of knowledge, which presumably should be higher for those who have completed more years of education. However, the relationship between education and attitudes toward violence is far from clear in the existing literature. While some work suggests that education serves as a constraint on the resort to violence (Collier and Hoeffler Reference Collier and Hoeffler2004; Thyne Reference Thyne2006), other research shows that education can serve as a basis for heightened conflict (Posen Reference Posen1993; King Reference King2013, 7–11).

Looking at the survey evidence, Figure 3 shows that the role of IHL knowledge continued to hold even after controlling for educational attainment. Yet it may instead be the case that IHL is only able to shape individual preferences among those with a sufficient baseline level of general knowledge and the ability to grasp relevant humanitarian principles. To test for the possibility of a conditional relationship with education, I separated the analysis based on those with greater (upper two quartiles) and lesser (lower two quartiles) levels of educational attainment (Figure 5). The results indicate that respondents across education levels are similarly influenced by IHL, and for civilian abuse the coefficient is actually slightly stronger (though not statistically distinguishable) for those with lower levels of educational attainment.

Figure 5.

Knowledge of Geneva Conventions and Support for Wartime Abuse by Level of Education

Note: Plot for Knowledge of Geneva Conventions coefficient (OLS) indicating percentage point change in support for each abuse index by relevant subgroup. Lines indicate 95 percent confidence intervals.

Fourth, concerns of social desirability bias loom large in studies of controversial topics (Podsakoff et al. Reference Podsakoff, MacKenzie, Lee and Podsakoff2003), especially with the present survey’s focus on the victimization of vulnerable populations. Faced with sensitive questions, respondents may be motivated to give answers they expect to be socially appropriate and what interviewers want to hear, rather than revealing their sincere preferences. Such biases may be even more severe for respondents with knowledge of the Geneva Conventions, since they are already more aware of existing constraints.

Several reasons suggest social desirability concerns are not unduly influencing the results.Footnote ³⁷ Despite the direct nature of some questions, many respondents remained willing to condone fairly brutal behavior explicitly. For instance, looking at the raw results for some of the component abuse items, more than 25 percent of respondents approved using land mines that would endanger civilians, and over one-third reported that they would refuse to save an injured or surrendering soldier. A similar lack of concern for prisoners was evident even among those with prior knowledge of IHL—over 30 percent still supported using torture. Moreover, the prior conditional plots again illustrate that individuals who should be more sensitive to social desirability dynamics, such as civilians, exhibit similar influences concerning IHL as former combatants and others normally espousing more hard-line perspectives. In sum, although there are a number of reasons to be cautious about inferring a direct relationship between international law and attitudes toward wartime violence, the preceding discussion has sought to address a number of the more serious likely threats to the validity of the findings.

D. Assessing Potential Mechanisms

The main analyses presented above offer only a general sense of the pathways and reasoning through which international law may be related to individual attitudes. Yet one of the strongest calls has been for empirical scholarship to understand better how international law and institutions shape preferences and outcomes (Martin and Simmons Reference Martin and Simmons1998). Building on the prior analysis establishing an association between international law and public attitudes, I turn below to assessing some of the specific mechanisms that may account for this relationship.

For instance, while the content of the conventions appears to matter, it is unclear whether individuals are thinking specifically about the international legal nature of those rules. Constraints on the resort to wartime violence reflect contemporary IHL, but they may also be rooted in other sources, such as domestic law, or religious and other values (Howard, Andreopoulos, and Shulman Reference Howard, Andreopoulos and Shulman1994). Persons exposed to IHL may become aware of the accompanying principles, but they may in fact be motivated by unrelated belief systems. The ICRC survey largely focuses on IHL by design, and it does not contain a comparable set of extensive questions tapping into alternative value systems. However, an item is included where respondents were asked what they believed was the basis of rules the violation of which perpetrators should be held accountable. Given the nominal nature of the outcome variable, a multinomial logit model was estimated. Figure 6 summarizes how IHL knowledge is associated with differences in the probability a respondent gave each answer choice.Footnote ³⁸ The results indicate a close and statistically significant correspondence between knowledge of IHL and beliefs that wartime limits are based on international legal rules. By contrast, respondents with prior exposure to IHL are less likely to believe that such rules are rooted in domestic legal systems; the result is similarly negative and statistically significant for religious values.

Figure 6.

Knowledge of the Geneva Conventions and Beliefs on the Basis of Rules for Which Violators Should Be Held Accountable

Note: Multinomial logit estimation. Average marginal effect (percentage point change in the predicted probability of the outcome) of Knowledge of Geneva Conventions, holding all other variables at their median values. Lines indicate 95 percent confidence intervals. Model is limited to respondents who believed there were rules for which people should be held accountable if they violated them.

The lack of findings for religious or other values does not necessarily imply that individuals exposed to international law view such principles as divorced from normative belief systems. Indeed, debate continues over whether international law primarily influences beliefs and behavior through normative appeals, or more instrumental motives; although the line between the two is often blurred (Simmons Reference Simmons2010, 277). A few other items from the ICRC survey offer an opportunity to examine the specific ways in which IHL may inform individual attitudes.

Respondents were presented with a question asking: “Is there anything that combatants should not be allowed to do in fighting their enemy?” Estimating a logit model, the first point estimate in Figure 7 shows the change in the predicted probability a respondent answered affirmatively that limits do exist on soldiers’ conduct when fighting.Footnote ³⁹ Not surprisingly, respondents with prior knowledge of IHL are over 15 percentage points more likely to say such limits exist. Those respondents who answered affirmatively were then asked a follow-up question about precisely why limits on wartime conduct should exist. The two available answer choices tap into related but distinct reasons underlying these beliefs—one more normatively centered on a sense that such actions are morally wrong, and one more instrumentally focused on the problems that may result from violations. The second point estimate in Figure 7 from a corresponding logit model shows that those who know about the Geneva Conventions are 7 percentage points more likely to believe limits exist because violations are morally wrong rather than for more instrumental reasons.

Figure 7.

Knowledge of Geneva Conventions and Beliefs of Limits on Wartime Conduct

Note: Logit estimation separately for each outcome. Average marginal effect (percentage point change in the predicted probability of the outcome) of Knowledge of Geneva Conventions, while holding all other variables at their median values. Lines indicate 95 percent confidence intervals. Model for ‘Why Limits?’ limited to respondents who answered affirmatively to prior question on ‘Any Limits on Fighting?’

Delving further, those who offered a more normative rationale were asked an additional follow-up question inquiring into why they believed such violations were morally wrong. Similar to the earlier item on reasons for holding violators accountable, some options centered on more universal ideals, such as the importance of the law and human rights, while others highlighted religion, culture, or a personal code.Footnote ⁴⁰ For this question, however, answers were not mutually exclusive, as respondents were allowed to give multiple reasons. Since a conventional multinomial logic model could not be used, I instead estimated a series of separate logit models for each outcome. The results in Figure 8 show that normative beliefs in the limits of wartime conduct for those with prior exposure to IHL tend to be rooted in more universal appeals to law and human rights. Although the confidence intervals for the law/human rights first differences overlap with some of the other answer choices, neither cultural principles nor personal codes were statistically distinguishable from 0, or no apparent effect. Moreover, those with prior IHL knowledge were significantly less likely to attribute any moral beliefs on the resort to abuse to religious ideals.

Figure 8.

Knowledge of Geneva Conventions and Beliefs for Why Breaking Limits on Wartime Conduct Is Wrong

Note: Logit estimation separately for each outcome. Average marginal effect (percentage point change in the predicted probability of the outcome) of Knowledge of Geneva Conventions, while holding all other variables at their median values. Lines indicate 95 percent confidence intervals. Models limited to respondents who answered ‘It’s wrong’ on previous item of ‘Why are there limits on fighting?’

Taken together, the previous discussion responds to a number of different concerns arising over various claims regarding the nature of the relationship between knowledge of IHL and attitudes toward wartime violence. The analysis indicates that exposure to the Geneva Conventions does have a discernable relationship with support for abuse, the specific humanitarian content espoused in the agreements plays a key role, and these principles appear to be associated with heightened moral calculations among respondents that are rooted in universal legal norms of restraint on the use of force and protections of individual rights.