Motivation: Rational Peasants, Civilian Agency, and Wartime Attitudes

A near consensus now exists among practitioners around the notion that counterinsurgency wars are decided by the relative success each combatant enjoys in winning popular support from the civilian population (Kilcullen Reference Kilcullen2009; Mao Reference Mao1961; Thompson Reference Thompson1966; Trinquier Reference Trinquier2006; U.S. Army 2007). From the counterinsurgent's perspective, victory is obtained through a combination of service provision, material assistance, information campaigns, and, above all, restraint on the use of one's force with the goal of minimizing civilian casualties. From the Vietnam-era Hamlet Evaluation System (HES) to contemporary efforts in Iraq and Afghanistan, winning the “hearts and minds” of the civilian population, or at least an important subset of it, is a central element of counterinsurgency campaigns.

Surprisingly, however, there are few rigorous empirical studies of civilian attitudes toward combatants during wartime (see Beath, Christia, and Enikolopov Reference Beath, Christia and Enikolopov2011, for an exception). To be sure, this gap is due in part to the logistical and ethical issues that accompany survey research in a wartime setting. Even if surveys could be conducted safely, methodological issues abound: social desirability bias, high refusal rates, and preference falsification may frustrate efforts to obtain reliable answers, especially if sensitive issues such as support for combatants are investigated using direct questions (e.g., Berinsky Reference Berinsky2004; DeMaio Reference DeMaio1984). As a result, one leading scholar has argued we should “bracket [the] question of individual motives and attitudes” (Kalyvas Reference Kalyvas2006, 101) in favor of theories that emphasize the more easily observable structure of opportunities and constraints facing civilians in wartime (see also Leites and Wolf Reference Leites and Wolf1970, 45).

Indeed, the prevailing view of civilians in the civil war literature suggests they are “rational peasants” (Popkin Reference Popkin1979), that is, utility-maximizing agents driven by the short-term dictates of survival. Without strong ethnic or ideological attachments, civilians seek to minimize their exposure to harm from the combatants while maximizing the benefits offered by combatants, including material assistance and basic security. Based on this consensus, our theories concentrated on explaining which factors drive civilian behavior: the relative balance of control exercised by combatants (Kalyvas Reference Kalyvas2006; Kalyvas and Kocher Reference Kalyvas and Kocher2007; Leites and Wolf Reference Leites and Wolf1970), competitive service provision (Akerlof and Yellen Reference Akerlof, Yellen, Aaron, Mann and Timothy1994; Berman, Shapiro, and Felter Reference Berman, Shapiro and Felter2011; Crost and Johnston Reference Crost and Johnston2010), and relative levels of civilian victimization (Condra and Shapiro Reference Condra and Shapiro2012; Kocher, Pepinsky, and Kalyvas Reference Kocher, Pepinsky and Kalyvas2011; Stoll Reference Stoll1993), to name only three. Prior allegiances, if present at all, are thought to be easily recast in the face of a given mixture of (threatened) punishments and (promised) rewards levied by combatants who aim to exercise control over the civilian population.

Despite privileging different independent variables, these theories converge on the same empirical prediction: civilians should respond symmetrically to combatant actions. Put differently, civilians should be indifferent to which combatant provides material assistance (Berman, Shapiro, and Felter Reference Berman, Shapiro and Felter2011, 776; Collier and Hoeffler Reference Collier and Hoeffler2004, 569; Mampilly Reference Mampilly2011, 54–5) or victimizes them (Condra and Shapiro Reference Condra and Shapiro2012, 183–4; Kalyvas Reference Kalyvas2006, 111–19; Stoll Reference Stoll1993, 20). What matters is the type and amount of aid or violence, not which combatant is responsible for these actions, since civilians cannot afford to discriminate when their survival is at stake. We would therefore expect civilians to respond uniformly to similar assistance provided by foreign and domestic nongovernmental organizations, for example. Similarly, these theories predict that civilians will provide information to each side equally and routinely, with the choice of which combatant to inform dictated only by prevailing levels of control or violence.

Yet the empirical findings in these studies are often inconsistent with this baseline expectation of constant causal effects across combatants. The effects of aid programs on violence reduction in Afghanistan, for example, appear to hinge on the ethnic composition of a district, not the level of control or prior violence (Beath, Christia, and Enikolopov Reference Beath, Christia and Enikolopov2011). Similarly, small-scale aid programs in Iraq have heterogeneous effects across Sunni and non-Sunni dominated districts (Berman, Shapiro, and Felter Reference Berman, Shapiro and Felter2011). In addition, Condra and Shapiro (Reference Condra and Shapiro2012)’s findings suggest that Sunni districts exhibit markedly different pre- and postcivilian casualty patterns of violence than Shia or mixed districts. In each case, it appears that civilians are responding asymmetrically, rather than uniformly, to combatants. If responses are conditional on combatant identities, then we need to theorize how heterogeneity in the effects of aid and violence—variation in who is rewarding or punishing civilians—affects attitudes and subsequent behavior asymmetrically.

Second, these theories rely on indicators of civilian behavior that are observationally equivalent with multiple causal mechanisms. As a result, we are left unable to adjudicate between competing accounts or, more simply, to explain why a particular relationship exists. Kalyvas (Reference Kalyvas2006) acknowledges, for example, his central independent variable, the relative distribution of combatant control, generates its effects via at least six different mechanisms, three of them attitudinal in nature (124–32). Observational data cannot tease apart these different mechanisms; only by directly measuring attitudes can the relative influence of each mechanism be determined.Footnote ² Getting the mechanisms right also has important policy implications: it matters whether civilians are providing tips to insurgents because they fear punishment for noncooperation or if they genuinely support the rebel cause since the policy response is likely to be different even if rebel control is constant in these scenarios.

Why Identity Matters: Intergroup Bias and Support for Combatants

We argue that wartime attitudes are shaped by intergroup biases that create durable expectations about the responsibility and blame for combatant actions toward the civilian population. These biases lead civilians to condition their interpretation of events on the perpetrator's identity (Lyall Reference Lyall2010). As a result, we expect that support for combatants will depend on the combatant's identity, as will the effects of actions taken by the combatants.

We define intergroup bias as the systematic tendency by individuals to evaluate one's own membership group (the “in-group”) more favorably than a group one does not belong to (the “out-group”) (Hewstone, Rubin, and Willis Reference Hewstone, Rubin and Willis2002, 576; Tajfel Reference Tajfel1970; Tajfel and Turner Reference Tajfel, Turner, Austin and Worchel1979).Footnote ³ Positive actions by one's own in-group are therefore viewed as arising from the innate disposition of in-group members while similar actions by the out-group are interpreted as situational in nature: the out-group was compelled by the situation to undertake a positive act, whereas the in-group did so because of its inherent nature. Negative actions by an in-group, by contrast, are understood as situational in nature (“forced to be bad”), while negative actions by the out-group confirm biases about the out-group and its members as bad actors.

Laboratory experiments have shown that intergroup bias is especially prevalent in situations where the out-group possesses a significant power advantage and members of the in-group feel threatened by the out-group (Brewer Reference Brewer1999; Esses et al. Reference Esses, Dovidio, Jackson and Armstrong2001; Hewstone, Rubin and Willis Reference Hewstone, Rubin and Willis2002, 585–6). Such conditions characterize counterinsurgency wars, in which members of a group are challenging a much stronger out-group such as a national government or external occupier in a bid for independence, autonomy, or state capture.Footnote ⁴

One potential basis for in-group identification is ethnicity.Footnote ⁵ Ethnicity can perform two functions under wartime conditions. First, it acts as a shorthand or “social radar” (Hale Reference Hale2008, 34) that allows forward-looking individuals to anticipate the nature of in- and out-group actions based on prior interactions or information about ethnic types. In particular, ethnicity can convey information about often unobservable characteristics about other relevant actors like soldiers and rebels, such as the credibility of their threats and assurances for (non)compliance. Biases about (non)coethnics can thus be drawn on to facilitate wartime risk management by allowing individuals to quickly attach probabilities, if only crudely, to the expected (re)actions of combatants (Lyall Reference Lyall2010, 15).

Second, and central to our argument, group membership also moderates beliefs about responsibility and blame for past actions toward the individual and her group. Viewed through the lens of intergroup bias, individuals are more likely to punish out-groups for transgressions, simply confirming biases about the out-group's disposition. Harm inflicted by the in-group, however, carries a different meaning: victimized individuals, and the community at large, may be more forgiving, since such acts are justified by appeal to extenuating circumstances that forced the in-group's hand. This bias helps explain why external occupiers are often blamed by victimized individuals for harm that was actually inflicted by the rebels themselves.

This discussion generates several conditional hypotheses, which are characterized by two types of asymmetry. Table 1 summarizes these hypotheses. First, we anticipate that harm inflicted by the out-group will be associated with a sharply negative effect on civilian attitudes toward the out-group (Hypothesis 1) and a correspondingly large rally effect for insurgents drawn from the in-group (Hypothesis 2). Taking Hypotheses 1 and 2 together, we should observe a clear asymmetry of effect on attitudes that is conditional on the joint identities of the perpetrator and victim(s).

TABLE 1.

Home Team Discount: The Asymmetry of Combatant Support in Civil War

Second, the in-group's victimization of civilians will not result in a transfer of support to the out-group (Hypothesis 3) but will instead generate only a neutral effect on out-group support. We may observe a modest negative effect on in-group support (Hypothesis 4), though this should be substantially smaller than the negative effect on out-group support after civilian victimization. Here we observe a second, more subtle, asymmetry: Unlike out-group victimization, which leads to support for the in-group, in-group victimization does not lead to a corresponding increase for the out-group. We term this the “home team discount”: insurgents may be less constrained in their abuse of their fellow in-group members since such actions do not lead to a corresponding positive effect on support for the out-group.Footnote ⁶

We also hypothesize that as we move from an individual's direct experience with the combatants to indirect experiences shared by other in-group members, the effects of victimization on combatant support should attenuate. Put differently, the magnitude of the action's effect, but not its direction, should diminish as we move from direct experiences to indirect ones provided by secondhand accounts, combatant propaganda, or rumors (Hypothesis 5).

Simply put, we anticipate that the effects of combatant violence on civilian attitudes are asymmetrical, not uniform. Violence by the out-group toward civilians should have negative effects on support for that group (in many contexts the counterinsurgent or national government) while having a positive effect on support for the in-group. In-group violence, by contrast, should result in only a modest negative effect on in-group support (if any effect is recorded) and should not yield a transfer of support to the out-group.

The Endorsement Experiment

The mechanics of a survey endorsement experiment are straightforward. Randomly selected respondents are assigned to a treatment group and asked to express their opinion toward a policy endorsed by a specific actor whose support level we wish to measure (here, ISAF and the Taliban). These responses are then contrasted with those from a control group of respondents that answered an identical question without the endorsement. Higher levels of enthusiasm for a policy with an endorsement relative to those without it are viewed as evidence of support for the endorsing actor. Since each respondent is assigned only one condition for any endorsement experiment, it is impossible for enumerators or others to compare support levels across different conditions for any individual respondent.

Drawing on electronic and print media, as well as focus groups and two pretest surveys conducted in sampled districts, we identified four policies with the properties desired for an endorsement experiment. Successful endorsement experiments share four properties (Bullock, Imai, and Shapiro Reference Bullock, Imai and Shapiro2011). First, selected initiatives should be in the same policy space so that they can be combined for statistical analysis. We select policies that are all related to domestic policy reforms: prison reform, direct election of district councils, a reform of the Independent Election Committee, and the strengthening of anticorruption policies. Second, these initiatives should be well known by individuals to minimize “Don't Know” responses and to differentiate support for an endorser from learning about a policy from the endorsement itself. In the survey, only 4.8% of respondents replied “Don't Know,” while refusal rates were low in all provinces. Third, these initiatives should actually be endorsed by the particular actors in question so that the questions are realistic and respondents take them seriously (Barabas and Jerit Reference Barabas and Jennifer2010).Footnote ⁸ Finally, a wide range of views was held by the general public about these initiatives, enabling us to detect support for endorsers without suffering from ceiling and floor effects.

Here, we reproduce one endorsement question, calling for reform of Afghanistan's notoriously corrupt prison system, below. The text of the remaining questions is provided in the Appendix.

• • Control Condition: A recent proposal calls for the sweeping reform of the Afghan prison system, including the construction of new prisons in every district to help alleviate overcrowding in existing facilities. Though expensive, new programs for inmates would also be offered, and new judges and prosecutors would be trained. How do you feel about this proposal?

• • Treatment Condition: A recent proposal by the Taliban [or foreign forces] calls for the sweeping reform of the Afghan prison system, including the construction of new prisons in every district to help alleviate overcrowding in existing facilities. Though expensive, new programs for inmates would also be offered, and new judges and prosecutors would be trained. How do you feel about this proposal?

Plagued by allegations of widespread overcrowding and systematic torture, among other human rights abuses (Human Rights First 2011), the Afghan prison system has been publicly criticized by ISAF, the Taliban, and outside observers alike. When polled, only 48% of Afghans claim to place any faith in the Afghan prison system (The Asia Foundation 2010). As a result, both ISAF and the Taliban have at various occasions offered remarkably similar reform programs (Nixon Reference Nixon2011; International Crisis Group 2008).Footnote ⁹

We also asked whether individuals would support a proposal to allow Afghans to vote in direct elections when selecting leaders for district councils. While technically permitted under Afghanistan's 2004 Electoral Law, these elections have not been held to date. In early 2010, the Independent Directorate for Local Governance (IDLG)—a government agency led by a presidential appointee that works with ISAF to coordinate subnational governance policies—publicly floated the idea of direct elections, to be held in March 2011 (they were not). Surprisingly, the Taliban seized upon this idea in their own propaganda, suggesting that Karzai's claims of being democratically elected were hollow since the Electoral Law was not being followed (Nixon Reference Nixon2011).

Third, we gauged support for reform of Afghanistan's Independent Election Committee (IEC), a much-maligned institution that failed to prevent widespread fraud in the 2009 Presidential and 2010 Parliamentary elections. The Afghan public has also soured on the IEC; only 54% approved of its performance in 2010 (The Asia Foundation 2010). ISAF and various international organizations spent much of the post-September 2010 election period publicly discussing various reform proposals. Even the Taliban, which had sought to derail these elections through violence, raised the IEC as additional evidence of the Karzai administration's democracy deficit (Nixon Reference Nixon2011).

Our final question asked whether individuals would support strengthening the new Office of Oversight for Anti-Corruption. Alongside security concerns, there is perhaps no more salient issue in the minds of Afghans today than corruption; a full 76% rated it a “major problem” in 2010, ranking it among the top two domestic concerns of individuals (The Asia Foundation 2010). Both ISAF and the Taliban are aware of the issue's salience; each has issued repeated public statements concerning corruption in general and the need to strengthen existing institutions, including the Office of Oversight, in particular (Konrad-Adenauer-Stiftung, Royal United Services Institute and Transparency International UK 2011; Office of the Special Inspector General for Afghanistan Reconstruction 2009).

For all four endorsement questions, respondents were asked to assess their level of support for each proposal on a five-point scale: “I strongly agree with this proposal”; “I somewhat agree with this proposal”; “I am indifferent to this proposal”; “I disagree with this proposal”; and “I strongly disagree with this proposal.” Respondents were also permitted to answer “Don't Know” or “Refuse to Answer.”

Figure 1 displays the overall (top row) and provincial distributions of responses to the four endorsement questions. Three patterns are worth noting. First, there is substantial heterogeneity of interprovince support for these policies, even independent of particular endorsements. This is encouraging since it suggests that the questions, taken separately or together, possess strong discriminatory power. Compare higher support for prison reform in Logar, for example, with Khost, or anticorruption efforts in Logar with Helmand. Second, support differs substantially across Taliban and ISAF endorsements. In some provinces including the Taliban stronghold of Helmand or the warlord-controlled Uruzgan, support for Taliban endorsed policies is far higher than ISAF endorsed policies. In fewer cases, notably in the key battleground province of Kunar, ISAF endorsement translates into higher support for at least two proposals. To further validate whether the endorsement experiments are measuring the support for the Taliban and ISAF, the Appendix presents a detailed descriptive analysis, which argues that the patterns we observe across provinces and districts are consistent with conditions in each province at the time of the survey.

FIGURE 1.

Overall and Within-Province Distribution of Responses from the Endorsement Experiments

Notes: Plots depict the distribution of responses to four policy questions (columns) across three groups (Taliban/ISAF endorsement groups and control group) for the overall sample (top row) and each province (bottom five rows). Sample sizes are also shown.

Sampling Design and Target Population

Our survey experiment was conducted between 18 January and 3 February 2011 in five provinces of Afghanistan.Footnote ¹⁰ We chose to concentrate on Pashtun-majority areas rather than a nationwide sample for three reasons. First, estimation at the village and district level requires a sufficient number of individual respondents and selected villages, respectively, making a nationwide sample prohibitively expensive if we scaled up with the same level of coverage. Second, these areas are not only substantively important—they encompass the Taliban “heartland,” and are considered the battleground provinces of the Afghan war—but are also areas where intergroup bias between Pashtuns and ISAF should be present. Third, these areas have experience with both ISAF- and Taliban-initiated violence. Indeed, they represent the high end of the distribution of violence.

We devised a multistage sampling design to generate estimates of support for combatants at the individual, village, and district levels. First, five provinces—Helmand, Khost, Kunar, Logar, and Uruzgan—were randomly selected from the 13 Pashtun-majority provinces.Footnote ¹¹ Next, we randomly selected one-third of each province's districts for a total of 21 selected districts. Third, villages were randomly sampled from these districts with the restriction that at least 10% of each district's villages had to be selected, yielding 204 villages. Households were then chosen within these villages using the random walk method. Finally, male respondents aged 16 years and older were randomly selected from these households using a Kish grid. We surveyed nine respondents from villages with populations below the mean of sampled villages (about 680 individuals) and 18 in villages with above-mean populations. Table 2 summarizes our sampling design and the realized sample.

TABLE 2.

Overview of Multistage Sampling Design

Notes: The sampling design was conducted as follows: First, five provinces shown in this table were randomly sampled from a total of 13 provinces with a Pashtun majority. Second, districts were randomly chosen within these districts. Third, villages were then randomly sampled from within selected districts. Fourth, households were randomly selected within each of the selected villages. Finally, one male respondent 16 years or older was randomly sampled within each of the selected households. The table also displays the number of Taliban- and ISAF-initiated violent events during one year prior to the survey (from 18 January 18 2010 to 17 January 2011) inflicted at the provincial level.

We obtained an 89% participation rate (2,754 out of 3,097 attempts). Figure 2 presents the distributions of basic demographic variables for our sample of 2,754 respondents. Our average respondent was a 32-year-old Pashtun male who was likely married (77% answered yes), possibly employed (only 58% reported holding a job, with the most frequent occupation being “farmer”), and earning between US $1.40 and $6.97 a day. Respondents typically had little or no government-provided schooling, and on average had 18 months of madrassa religious schooling. In keeping with the high poverty rates of the sample area, respondents on average had less than 90 minutes of electricity a day. Less than half (48%) possessed cellphones, while nearly all (90%) owned a radio, the principal means of obtaining information. Ethnic minorities, mostly Tajiks, made up 6% of the respondents and had comparatively higher rates of education and employment.

FIGURE 2.

Distribution of Demographic Characteristics

Notes: Bar plots indicate the number of respondents in each category for demographic variables. Dashed lines represent median values.

Since the vast majority of Afghans live in small rural settlements, we avoided sampling from large urban centers such as district capitals. Our villages thus range from 29 to 6509 inhabitants (mean: 681 individuals). We also made the difficult decision to include only male respondents given the cultural and logistical challenges of interviewing women in these violent and deeply conservative areas. Our protocol included a special consent form for those aged 16–18 years. We elected to include these individuals since Afghanistan's median age is only 18, a fact typically overlooked by existing surveys of public opinion.

Of the original 204 villages, only four proved inaccessible due to a combination of Taliban hostility, the presence of criminal elements and, in two cases, the inability of the enumerators to locate the selected village.Footnote ¹² In all cases, village elders, who may have been members of the Taliban, were first approached by ORCA district supervisors with the relevant connections to describe the survey and to receive assurances of enumerator safety. Despite our extraordinary level of access to Taliban-controlled areas, a testament to ORCA's connections as well as its decision to hire locals enumerators, we nonetheless experienced myriad delays and logistical challenges. These included movement restrictions due to open war-fighting (especially in Helmand); Taliban and militia roadblocks (typically appearing after 2 p.m. in most of our districts); and attempted highway robbery. In one case, a clean-shaven enumerator was pulled from his car in Kunar where, along with other motorists, his face was blackened with motor oil by local Taliban to signify his “sinfulness.”Footnote ¹³ Our most serious incident involved a district manager in Uruzgan, who was wounded by a roadside improvised explosive device (IED) the day after placing the last bundle of completed surveys on a fruit truck (our method of returning surveys to Kabul).

Measuring Exposure to Violence

Table 2 outlines just how violent these provinces have been using two different datasets. The number of attacks initiated by the Taliban and ISAF from 18 January 2010 to 17 January 2011 was extracted from ISAF's Combined Information Data Network Exchange, which records the date, location, and type of event from 17 specific categories.Footnote ¹⁴ These data show that the sampled provinces have experienced variable levels of violence but, on average, are considerably more violent than other non-Pashtun provinces.

Figure 3 illustrates this pattern by overlaying the location of sampled villages (as black dots) on the distribution of Taliban and ISAF violence (as red dots) during 2010. Two spatial patterns emerge immediately. First, while some of our surveyed villages are within extremely violent areas, other sites experienced relatively few attacks nearby, giving rise to considerable heterogeneity in the level and type of harm experienced by respondents. Second, we overlay the spatial distribution of relative control by combatants, as measured by ISAF, at the district level. Nine of the 21 districts are defined as under Taliban control (red areas), while another nine were coded as “contested” (green areas). Only three districts in our sample are thus considered under central government or pro-ISAF local control (blue areas). These data underscore both the difficulties of research in these areas as well as the fact that, for nearly all respondents, experience with harm at the hands of ISAF or the Taliban (or both) is not an abstract notion but a realistic possibility—and, for many, a reality. In the Appendix, we demonstrate that though these provinces are more violent than non-Pashtun provinces, the sampled villages exhibit violence levels similar to nonsampled villages in these districts. We also show that sampled villages are similar to nonsampled villages on other substantively important dimensions. As expected with multistage sampling, in-sample balance at the district and province level is slightly worse, given the small samples of districts (21) and provinces (5).

FIGURE 3.

Sampled Villages

Notes: Black dots represent randomly sampled villages within five Pashtun-dominated provinces. Red dots represent Taliban- or ISAF-initiated violence in 2010. Three colors are used to distinguish Taliban-controlled (red), government-controlled (blue), or contested (green) districts.

We measure our central explanatory variable, exposure to harm at the hands of ISAF or the Taliban, with several questions. We distinguish between two models of harm: direct exposure, in which an individual or his family are directly affected in the past year; and indirect exposure, in which individuals are asked about their awareness of ISAF- or Taliban-inflicted harm within their manteqa (area) over the same time period. These self-reports, elicited prior to the endorsement experiments, were preceded by a script that defined “harm” as both physical injury and property damage. These questions and detailed descriptive analyses appear in the Appendix.

Our respondents reported a staggering degree of exposure to ISAF and Taliban violence. Some 37% of respondents indicated that they had personally experienced victimization by ISAF, while a similar 33% acknowledged Taliban victimization (left corner plot in Figure 10(a) in the Appendix). About 19% of respondents reported that they had been victimized by both combatants. Nearly 70% of respondents had heard of ISAF harming civilians in their manteqa, while about 40% had similarly heard of Taliban-inflicted harm in their area over the past year (left corner plot in Figure 10(b)).

These self-reported measures were followed by questions about whether the individual had directly experienced or was indirectly aware of the efforts by the responsible combatant to make amends for harm inflicted. These questions are designed to test whether intergroup biases are irretrievably fixed or, instead, there remains the possibility of softening, if only partially, these biases in a wartime setting. We are therefore able to classify individual experiences by exposure to harm (yes/no), by combatant, and whether the individuals were directly approached by the combatants following victimization or had heard of such efforts within their manteqa.

To be approached by ISAF signifies that an individual or family received a one-time solatia payment, typically around $2,500, that absolved ISAF of criminal liability for civilian casualties or property damage.Footnote ¹⁵ By contrast, to be approached by the Taliban signifies that the aggrieved party received a funeral oration by the Taliban extolling the virtues of the fallen individual(s) as well as modest monthly payments or basic staples such as kerosene or foodstuffs.

Yet can we rely on these self-reports, especially those obtained via direct questions? At least two objections might be raised. First, we may be concerned that questions of wartime harm are simply too sensitive to elicit truthful answers. This, too, was our initial concern, but one subsequently mitigated when enumerators reported that respondents, far from being reticent about sharing their experiences, were only too eager to describe them. In fact, the combined rate of “Don't Know” and “Refuse to Answer” for these questions was only 2% for Taliban and 1% for ISAF-inflicted harm questions.Footnote ¹⁶

Second, there may be endogeneity between the certain covariates—for example, the level of combatant control, or an individual's socioeconomic and cultural attributes—and survey responses. Individuals did report a much greater (second-hand) awareness of ISAF victimization than Taliban misdeeds, for example. This may suggest a greater willingness to speak openly about ISAF, especially in the Taliban-controlled districts that represent 43% of our sample. Coethnic bias itself may also skew these self-reports: Pashtuns could conceivably be more likely to report ISAF victimization than harm by the almost-exclusively Pashtun Taliban.

Without dismissing the dangers of endogeneity, there are reasons to believe such concerns may be muted. First, minimal nonresponse rates were recorded; respondents clearly were not hiding their answers behind appeals to “Don't Know” and “Refuse to Answer.” Second, almost identical levels of victimization were recorded at the hands of ISAF and the Taliban. Third, substantial variation exists within Taliban-controlled and contested districts (and villages), suggesting that a heavy Taliban presence does not necessarily dictate pro-Taliban answers. Fourth, the large discrepancy between indirect exposure to ISAF and Taliban victimization is likely due to the Taliban's own impressive and far-reaching information campaign. Unlike ISAF, the Taliban is adept at disseminating news of ISAF-inflicted civilian casualties (real or imagined) via cell phone videos and SMS as well as routine visits to villages. The efforts of ISAF lack the reach of those of the Taliban and remain largely confined to urban centers and traditional media.

Finally, to complement respondent self-reports, we calculated the number of attacks initiated by the Taliban and ISAF occurring in or near each village. To be consistent with our survey questions, we count only the attacks that occurred within a five kilometer radius of each village during one year prior to the survey. This event data offers an extraordinarily fine-grained measure of violence that can act as an important, if limited, alternative measure to our self-reported measures of exposure to harm. Intriguingly, we find little correlation between these individual self-reports and ISAF's own dataset of violent events. We constructed measures of violence using the number of Taliban and ISAF-initiated events during the year prior to the survey at different radii around these villages and found little correlation.Footnote ¹⁷ Moreover, there is almost no correlation between these self-reports and ISAF's own civilian casualty data. ISAF's data suggest that the Taliban are responsible for inflicting 89% of incidents involving the death or wounding of civilians in 2010 (or, by UNAMA's estimate, 83% of civilian deaths). By contrast, our self-report data reveal a much more even distribution of harm inflicted, with a slightly larger proportion of respondents reporting victimization at the hands of ISAF compared to the Taliban.

Several factors may account for this discrepancy. First, the two measures operationalize harm differently. ISAF's data focuses solely on deaths and wounded, while our self-reports measure both physical harm and property damage. The latter category, which ISAF does not collect systematically, is critical because it represents a far more frequent occurrence than the rare mass-casualty events that command media attention. Second, ISAF's coding rules for assigning responsibility for a particular event rely on objective indicators (e.g., if a Taliban-placed IED kills civilians, then the Taliban is assigned responsibility for the event) while the population may not be making the same calculation (e.g., ISAF is blamed for IED-conflicted casualties since ISAF's presence made the IED likely). Finally, these self-reports measure individual-level exposure, while ISAF's event data in present form can only be pegged to the village level. For these reasons, we believe that our self-reports offer a more comprehensive measure of harm inflicted than ISAF's own data.

Main Results

Our multilevel models generate estimates of a respondent's level of support for the particular combatant; estimates are measured in terms of the (posterior) standard deviation of respondents’ preferences within a single dimensional policy space (i.e., ideal points). If a respondent is assigned to the control group, no estimate is generated. Imagine, for example that a respondent is neutral toward the proposed reform initiative. If this individual's support level for the Taliban was a “1,” then it implies that a Taliban endorsement can shift this individual's ideal point by one standard deviation in the proreform direction.

Attitudes Toward Combatants

Who is more likely to support the Taliban or ISAF? As an initial test of the validity of our survey instrument, we plot individual-level support estimates from the direct exposure model against key demographic covariates in Figure 4. We note that estimated support for both combatants decreases with age, though the Taliban enjoy more support among the youth than ISAF. This may be somewhat surprising given the Taliban's public image as a conservative gerontocracy, but the war has created opportunities for advancement among the youth, and the Taliban appears to be given credit for this mobility. Predictably, support for ISAF—or, more bluntly, decreased dislike for ISAF—is associated with increased education and with fewer years spent in madrassa education. The reverse is true for the Taliban: its estimated support levels rise with years of madrassa education and fall with years of government education. Finally, per capita income is positively associated with estimated ISAF support (or, again, less dislike of ISAF) while there appears to be no clear pattern between income and Taliban support.

FIGURE 4.

Bivariate Relationships between Demographic Variables and Individual-level Support for the Taliban and ISAF

Notes: Support estimates are given in terms of the standard deviation of ideal points. For age (the left column), dots represent respondents and the lowess curve is fitted as a solid line. For remaining plots, box plots are used: the width of each box is proportional to (the square root of) the number of respondents for each value of the demographic variables.

Attitudes by Levels of Victimization

What are the average support levels for ISAF and the Taliban among those harmed by these combatants? As an initial investigation of the effects of victimization, we estimate the level of support for each combatant according to whether respondents reported harm by ISAF only, the Taliban only, by both combatants, or by neither (Figure 5).Footnote ²⁵

FIGURE 5.

Estimated Support Levels for Each Combatant by Victimization Type

Notes: We derive and then plot the posterior predicted mean support level from our multilevel models with 95% confidence intervals.

We find that most respondents are not supporters of either combatant: among those not harmed by either combatant, respondents exhibit a strongly negative effect toward ISAF (a −0.62 effect in standard deviations of the ideal points) and are only slightly less negative about the Taliban (at −0.29 standard deviations of the ideal points). Interestingly, those respondents who were harmed by ISAF only or by both ISAF and the Taliban exhibit a positive effect toward the Taliban (at 0.27 standard deviations and 0.15 standard deviations, respectively). Respondents who had been harmed only by ISAF or the Taliban were the least supportive of these combatants (at −0.89 standard deviations for ISAF and −0.40 standard deviations for the Taliban), though the magnitude of the effect is clearly asymmetrical across combatants.

Asymmetrical Effects of Victimization

We now provide the key evidence to support our central claim that the effects of violence on civilian attitudes are conditional on combatant identity. Figure 6 illustrates the estimated average effects of ISAF (left panel) and Taliban victimization (middle panel) on ISAF and Taliban support at two levels: the individual/family (solid circles) and the manteqa (open circles). In the right panel, the net differences between the effects of victimization on support for the Taliban and ISAF are also provided.

FIGURE 6.

Estimated Effects of ISAF and Taliban Victimization on Support Levels for Each Combatant

Notes: The right panel presents the differences between the results in the middle and left panels. Posterior means of coefficients derived from multilevel models are plotted with 95% confidence intervals.

A number of important observations stand out. First, as expected by Hypothesis 1, we note that ISAF victimization is associated with a significant negative effect on ISAF support. ISAF victimization is also associated with a substantively large positive effect on Taliban support (Hypothesis 2). These data confirm that harmed individuals are both punishing the out-group for inflicting harm and are transferring support to the out-group as expected if intergroup biases are in play.

Second, Taliban victimization has no clear effect on support for ISAF. As expected by Hypothesis 3, we do not observe a blossoming of support for ISAF after Taliban-inflicted harm, suggesting that the out-group is not rewarded for its (relative) restraint. Moreover, Taliban victimization itself only slightly decreases Taliban support and certainly does not approach the penalty on ISAF support witnessed after ISAF victimization. While it would be incorrect to conclude that the Taliban can victimize civilians with impunity, they appear to enjoy a much larger cushion of public support than does ISAF, for whom civilian casualties are associated with a steep negative effect on support.

Third, we observe the hypothesized asymmetry of effects between ISAF victimization on Taliban support (middle panel, left column) and Taliban victimization on ISAF support (left panel, right column). This difference is statistically significant and is estimated to be 0.37 standard deviation of the ideal points (with the 95% confidence interval of [0.02, 0.73]) at the individual/family level. The right panel in Figure 6 underscores the asymmetry between the effects of ISAF victimization on relative support for the combatants (Taliban support minus ISAF support) and those associated with Taliban violence. Once again, it is apparent that while ISAF victimization creates a substantial positive effect on net Taliban support, Taliban victimization generates at best only a very modest positive effect on relative support for the out-group.

Are these results driven by personal exposure to harm, as suggested by Hypothesis 5, or are indirect influences equally as important? Returning to Figure 6, we observe that manteqa-level effects (as represented by open circles) of ISAF victimization are attenuated when compared with individual self-reports. We find that the net difference in the effect of ISAF violence on ISAF support between the individual and manteqa level is nearly statistically significant, with a 0.203 (95% confidence interval at [0, 0.65]). We do not observe a similar pattern of attenuation for Taliban violence, in part because the individual-level effects are already small. While the general pattern of effects is therefore similar across individual and manteqa levels, the magnitude of estimated effects on support levels for each combatant is typically diminished at the manteqa level, and significantly so for ISAF.

One interesting wrinkle also emerges from these findings: both Taliban and ISAF endorsements are, in some cases and regions, associated with reductions in support for the proposed reform relative to the control endorsement.Footnote ²⁶ While caution is warranted, one interpretation of this result is simply that the control most closely resembles an official (government) endorsement. What we may be measuring, then, is the relative shift in support for Taliban and ISAF combatants associated with victimization against a more general backdrop of war weariness in which neither combatant is especially favored. That said, however, since support in war is relative, not absolute, it is also clear that the Taliban do enjoy a greater measure of support, and more tolerance of inflicting harm on civilians, than ISAF.

Effects of Post-Harm Mitigation Efforts

Can combatants offset the negative effects of their violence by rendering postharm assistance? We offer a preliminary exploration of this issue in Figure 7, which presents the estimated combined effect of being harmed but not subsequently approached (circle), the effect of being approached (square), and the combined effect of victimization and receiving post-harm assistance on attitudes toward the combatants.

FIGURE 7.

Estimated Effects of Victimization and Subsequent “Approach” by Combatants on Support

Notes: The right panel presents the differences between the results in the middle and left panels. Posterior means of the corresponding coefficients are plotted along with 95% confidence intervals.

Two broad conclusions stand out. First, postharm mitigation efforts by both combatants are associated with a marked negative effect on support for the other combatant. Taliban victimization alone has a small positive effect on attitudes toward ISAF (as reported above) but, once aid has been proffered, the effect is a substantially negative effect on attitudes towards ISAF. Similarly, experience with ISAF victimization alone is associated with a positive effect on support for the Taliban. Yet if aid has been provided by ISAF, the result is a large negative effect on Taliban support (second column). Indeed, the net difference in effect between individuals who were harmed but not approached by ISAF and those who were harmed and then approached is a highly significant −1.00 standard deviation movement away from a pro-Taliban position (95% confidence interval of [−1.43, −0.53])

An optimistic take on these results would suggest that hearts and minds can in fact be swayed, if not bought outright, via postharm mitigation efforts. Yet a second finding is also important: these efforts appear to have little effect on support for the actual combatant that rendered the assistance. While ISAF does appear more successful at generating modest positive effects after its efforts, there is much less positive movement in attitudes toward the combatant responsible for the victimization than there is a shift away from the other combatant. In this situation, postharm aid appears less about persuading civilians to move toward one's side as it is about shifting away from the rival combatant.

We must also be careful not to overstate the finding about the effects of ISAF assistance. There is a severe selection problem at work here, because ISAF managed to “approach” only 16% of those who claimed that they had been harmed by ISAF. By contrast, the Taliban “approached” over 60% of those who self-identified as suffering Taliban victimization. ISAF's decision-making on the extension of condolence and battle damage payments is often haphazard, but is at least partly conditioned on anticipated reception in a given village (Campaign for Innocent Victims in Conflict 2010b). The more likely a unit is to receive armed resistance from aggrieved villagers, the less likely ISAF will return to disburse payments. Given closer media scrutiny, mass casualty events are more likely to be followed by postharm mitigation efforts. Aggrieved individuals in small-claims cases, by contrast, are often forced to go to military bases to receive their funds, discouraging all but the most determined and risk acceptant claimants (Campaign for Innocent Victims in Conflict 2010a). As a result, the finding regarding ISAF's ability to use condolence payments to overcome intergroup bias relies on a small subset of individuals who were specifically selected to receive this assistance. Indeed, these individuals may be the ones most likely to be swayed by ISAF's efforts.Footnote ²⁸

Alternative Explanations

Our findings strongly suggest that intergroup biases are responsible for the observed asymmetries of effects of combatant actions on civilian attitudes. Yet what of alternative explanations? We consider four alternative identity-based theories and three additional explanations that emphasize battlefield factors such as the distribution of aid, relative control by the combatants, and Taliban service provision.

Village-Level Analysis

These findings also pose a substantial challenge to studies that presume individual- or group-level motives and mechanisms but test these arguments with data aggregated to larger territorial units such as villages or districts. As noted above, our violent event datasets are only weakly correlated with respondent perceptions of harm. This lack of a relationship carries over into our multilevel model, in which ISAF and Taliban violent events from January to December 2010 are not associated with respondent attitudes toward the combatants regardless of whether we rely on a 1-, 5-, or 10-km radius around the village. Simply put, once we control for variance at the individual level and varying intercepts at the village, district, and province level, other factors at the village and district levels are not strongly associated with combatant support.Footnote ³⁴

To be sure, not every research question requires individual-level data, and we must be cautious when assessing causality in the absence of a clear identification strategy. Yet these nonfindings suggest that the now-standard practice in civil war studies of relying on aggregated event counts may be problematic. At a minimum, these data are unlikely to be useful in testing the attitudinal mechanisms thought to link violence to behavior. Even studies that draw on fine-grained data on civilian casualties (Condra et al. Reference Condra, Felter, Iyengar and Shapiro2011) are likely to be misleading if these events are only loosely connected with civilian perceptions or are aggregated into larger territorial units far removed from the victimized individuals. In other words, civilian attitudes may represent a substantial omitted variable in most statistical accounts of civil war dynamics. These mistaken inferences are likely compounded if the event data are not disaggregated by perpetrator and victim identities, thus missing the conditional nature of violence's effects on civilian attitudes.

Endogeneity

Two potential objections to our intergroup bias theory should also be addressed here. First, it is possible that perceptions of harm inflicted are themselves driven by prior attitudes toward the combatants. Willingness to assign blame for harm may therefore be endogenous to preexisting beliefs, particularly in circumstances where the actual identity of the perpetrator is unknown. Addressing this issue is difficult due to the absence of completely reliable measures of violence inflicted on individuals. In addition, research in conflict settings is typically conducted after repeated interactions between combatants and civilians have already occurred. Nevertheless, our theory generates a clear set of predictions about asymmetrical effects, whereas claims about endogeneity between attitudes and perceptions of harm do not. Specifically, the presence of endogeneity implies that self-reported harm by any group is negatively correlated with support for that group and does not predict that these effects are conditional on the perpetrator's identity.

It is worthwhile to note that information may be unreliable, and uncertainty rife, in wartime settings. Individuals themselves may not know which combatant to blame for the harm inflicted, especially if they are literally caught in the crossfire. Individuals may also be biased against the outsider, leading to the over-reporting of harm by the out-group. Under this scenario, blame attribution may be at work.Footnote ³⁵ Nevertheless, in contrast to our intergroup bias argument, this alternative theory does not predict the asymmetry in the marginal effects of victimization. Instead, a theory of blame attribution would suggest that the effects of victimization are uniform across combatants even though self-reports may incorrectly attribute some of the harm inflicted to the out-group.

A second concern is that since groups vary along multiple dimensions, not simply ethnicity, these differences in political objectives and military strategies may matter more than intergroup bias. In general, addressing such a concern would require the researcher to manipulate certain aspects of the identity and/or actions of combatants while keeping the ethnicity difference constant. Clearly, this is a difficult task in a wartime context. In our case, we can rule out specific explanations. For example, one may hypothesize that since ISAF is an external intervener, it possesses a shorter time horizon than the Taliban, the local in-group, and thus individuals will respond differently to their actions. However, this hypothesis is unlikely to hold in our case. While external interveners can of course withdraw, this prospect does not necessarily predict asymmetrical effects of harm. Moreover, an individual's “shadow of the future” would also have to loom exceedingly large for these distant concerns to trump short-term survival calculations.Footnote ³⁶