Context

Liberia is a West African nation of roughly 3.5 million people. Between 1989 and 2003, civil wars killed hundreds of thousands and displaced a majority of the population. A 2003 agreement ushered in peace. Two democratic elections followed. Police and court systems are slowly rebuilding, but they have little reach outside a few towns and are largely expensive, inefficient, and corrupt (Isser, Lubkemann, and N'Tow Reference Isser, Lubkemann and N'Tow2009).

As a result, Liberians mostly rely on local, informal institutions to manage disputes. The volume of those disputes is great. Nationally, in 2011, 16% of Liberians reported a land dispute since the war's end, and 10% reported another major dispute, such as over money or inheritances; 20% of land disputes turned violent (Vinck, Pham, and Kreutzer Reference Vinck, Pham and Kreutzer2011, 49). Roughly 40% of land disputes and 16% of nonland disputes remained unresolved since the end of the war (61).

Local disputes are also difficult to resolve. There is often no single acknowledged authority to mediate or enforce bargains, and there are rival forums for resolution, ranging from customary leaders, administrative leaders, elder councils, local peace committees to courts and the police. Parties to a dispute can thus “shop” for forums in search of favorable treatment.

Moreover, property rights are often unclear. Agreements are seldom recorded, few records survived the war, boundaries are poorly marked, and there are often competing claims to the same house, market spot, or farmland. For example, a newcomer may occupy a market stall vacated during the war. Even when she acknowledges the original inhabitant's claim, there may be a disagreement over compensation for the newcomer's structural improvements. In other cases, rights are more poorly defined, such as when a farmer uses fallow land that was historically tilled (but not formally held) by his neighbor.

Intervention Design

In 2009–10 the government and the UN directed the UN High Commission for Refugees (UNHCR) and a nongovernmental organization (NGO), the Justice and Peace Commission, to run an ADR campaign in rural Liberian communities (mapped in Figure 1).

FIGURE 1.

Map of Liberia and study communities

In treated communities, the NGO mobilized roughly 15% of adults to participate in the educational workshops. Groups of 35 residents participated in each workshop, which involved eight days of training and was led by two facilitators. These eight training days were spread over two months, allowing trainees to practice in between. The composition and content of the workshops were fairly homogeneous across communities, with some idiosyncratic variation in quality and delivery.

The training was designed to strengthen existing and longstanding informal methods, such as adjudication by customary leaders, as well as train and encourage ordinary residents to negotiate their own disputes or mediate those of their neighbors. The training drew on a wide range of examples, including community and group conflicts, but emphasized interpersonal disputes, especially land, money, domestic, and neighbor disputes. The workshops focused on the following tools, skills and practices: (1) direct engagement in one's own or others’ disputes; (2) strategies for problem solving and negotiation; (3) face-saving and “positive-sum” resolutions; and (4) avoidance of forum shopping and reliance on informal rather than formal justice mechanisms for civil disputes. Workshops combined lectures with group discussion, role-playing, and opportunities for people to share experiences. No information campaign was provided to the broader community.

Facilitators typically lived in the communities for two to four months. After-hours interactions with participants provided an opportunity to demonstrate, facilitate, and reinforce the ideas and norms taught in workshops. As we discuss later, however, facilitators seldom played a direct role in disputes as mediators. The principal treatment was the workshop.

Target Population and Participants

The intervention targeted 3 of Liberia's 15 counties: Lofa, Nimba, and Grand Gedeh. These counties were selected because they were more densely populated and more affected by the war than other areas and thus were expected to have more disputes and weaker social bonds. County officials nominated 246 communities they felt could benefit from the intervention, ranging in size from 100 to 5,000 persons. In 2008, 10% of these 246 communities reported a violent strike or ethnic dispute and 7% a peaceful protest; 9% of their residents reported a dispute over money in the past year, and 24% reported a land dispute since the war (Appendix Table 1). Therefore, although officials tried to select conflict-prone communities, compared to the national data the sample was not extreme, with rates of land conflict only moderately higher than the national average (Vinck, Pham, and Kreutzer Reference Vinck, Pham and Kreutzer2011).

Community members also had prior exposure to the ideas underlying the intervention. At baseline, 28% of residents reported prior “peace trainings,” and 40% said they were members of a “peace group.” However, our interviews suggested that these figures overstated prior exposure: Prior trainings were usually brief and covered a varied set of topics, such as reflections on the war or ethnic relations, and the peace groups included any group with a mission of postwar recovery or ethnic harmony (e.g., a multiethnic youth club). Nonetheless, any treatment effects must be considered in light of some prior exposure, and so we searched for differential treatment effects.

The intervention was not very selective within the communities. Community leaders typically invited leaders and opinion-makers to participate, but were forced to mobilize ordinary residents to meet the ambitious 15% target. In the end, those who participated looked much like those who did not. A comparison of pre-intervention traits shows that minority status had little association with attendance nor did several of the strongest correlates of land conflict—having one's land or house taken during the war, having been a refugee or displaced, or having been a victim of war violence. However, trainees were slightly more likely to be older, male, and lifetime residents with land (Table 2).

Noncooperative Bargaining

Noncooperative bargaining is usually modeled as a series of alternating offers between two parties with an interval in between the offers (Kennan and Wilson Reference Kennan and Wilson1993). If both parties have complete information and bargains are enforceable, a self-enforcing agreement is quickly reached. Costly delays in bargaining, then, come from incomplete information (usually the private cost of delay), commitment problems, and the delays between alternating rounds.

Fearon (Reference Fearon1998) developed a simple formal model where two parties bargain over two possible deals. Their bargaining resembles a common 2×2 coordination problem (Battle of the Sexes), whereas enforcement resembles a repeated Prisoners’ Dilemma. Both parties would prefer coordinating on one deal to no deal, but they prefer different deals. Lengthy bargaining is costly to both sides. Rational behavior leads to the classic “war of attrition”: Each party rejects the other's offer until the party with the highest cost of delay concedes. With complete information, the party with the lower cost of delay receives his or her preferred deal immediately. It is private information, such as uncertainty over the opponent's cost of delay, that causes long, costly wars of attrition.

Once the parties agree on a bargain, they begin an enforcement phase in which both have a short-run incentive to renege. As in the bargaining phase, there is an interval between observing each other's actions (without it there would be no gains from defection and no enforcement problem).

This simple two-phase game yields a fairly standard set of comparative statics. The length and cost of the bargaining process decrease as (1) the cost for not reaching a deal rises, (2) uncertainty and information asymmetries decrease, and (3) the value of future payoffs declines. Meanwhile, enforcement becomes easier as (1) the short-run benefits of defection decrease, (2) the costs of not finding a deal rise, and (3) the value of future payoffs increases. Finally, in every phase of the game, there is an interval between rounds that is essential to the costliness of bargaining (otherwise alternating offers would proceed rapidly to a conclusion). As we discuss later, ADR and informal institutions are explicitly designed to affect nearly every element of this process.

Three implications of this model of noncooperative bargaining are noteworthy. First, if an intervention improves both bargaining and enforcement, then there is an ambiguous effect on the length of disputes. Better enforcement raises the stakes of the bargaining phase and thus gives disputants an incentive to bargain harder and longer. Second, a drawback of noncooperative bargaining theory is that it does not necessarily generate predictions about when, or if, violence will erupt. For simplicity, we interpret violence as a risk that increases with the length of delay that leads to a pause or cessation in bargaining.

Third, this model treats the parties as rational, but in practice we know that disputants often behave irrationally. A large body of behavioral decision-making research emphasizes that people rely on simplifying strategies and cognitive heuristics that are prone to a number of errors. Bazerman et al. (Reference Bazerman, Curhan, Moore and Valley2000) and Kahneman and Tversky (Reference Kahneman, Tversky, Arrow, Mnookin, Ross, Tversky and Wilson1995) summarize the evidence as it applies to two-party negotiations: Parties are often loss averse; they assume their preferences are incompatible; they undervalue concessions, falsely assume a fixed pie, and miss mutually advantageous moves; they allow conflict to escalate even when the optimal decision would be to change their strategy; and they hold self-serving recall biases that are exacerbated by ambiguous information.

Less is known about the sources and effects of emotion and anger. In practice we know that emotion is important and may exacerbate the irrational tendencies just described. In the context of noncooperative bargaining, we might think of anger much as we do violence: Emotion can induce parties to leave the negotiating table and thus prolong the delay between bargaining rounds. Together, irrationalities and emotion may increase information asymmetries, slow the bargaining phase, and thus increase the length of disputes and the risk of violence.

Alternative Dispute Resolution

Behavior Change through Mass Education

Recall that this intervention did not introduce and impose ADR. Rather, it merely educated and persuaded communities to adopt its practices and principles. As a result, the previous theoretical discussion—especially the predictions of bargaining theory—is predicated on the effectiveness of the education campaign in changing behaviors.

There are several reasons for skepticism about the effectiveness of education campaigns in changing behavior. First, foreigners and elites, who may not understand the context, promulgate most education campaigns. Second, campaigns are often short (in this case, eight days of workshops). Third, a rationalist might argue that norm and behavior change is the product of changed economic fundamentals and constraints, on which an education campaign has no direct effect. Fourth, we may worry about a “Hawthorne effect”—a temporary increase in behaviors due to training and observation. Finally, we have cause to be skeptical of social engineering in general: Efforts to change practices and rules often fail to achieve their intended purposes or have unintended consequences when thrust on strong, preexisting social relations and obligations (Moore Reference Moore1973).

The advocate-centered theory of change underlies a range of social engineering interventions, from public health to voter education. As we noted in the introduction, there is a growing base of evidence that information campaigns can change political behavior in the short term, such as around elections. Furthermore, a large case literature on international norm diffusion emphasizes the power of persuasion in explaining change across and within countries, from the adoption of human rights (Finnemore and Sikkink Reference Finnemore and Sikkink1998) to reduction in female genital cutting (Cloward Reference Cloward2010) and Chinese footbinding (Mackie Reference Mackie1996). These accounts argue that third parties can use their status, resources, and skills of persuasion to convince a core of influential actors to change their actions and value systems. Once this core grows large enough, the rest follow in a “cascade effect.” The same idea underlies the design of this intervention. UNHCR deliberately chose to target 15% of adults in the hopes of reaching some “critical mass” of adults and inducing a cascade of new skills, practices, and norms.

Persistent change, then, could come about for two reasons. First, much like a new agricultural technology, ADR may be widely adopted for its own value. Second, education could change individuals’ shared expectations of how others will act in a dispute, as well as their beliefs about the sanction or praise that should accompany those actions.

Experimental Design

We worked with UNHCR to randomize the intervention at the community level. Of the 246 nominated communities, 116 were initially randomly assigned to treatment, stratified by the three counties. We were unable to randomly assign participants to the workshops within treatment communities.

Twenty-four facilitators, working in pairs, visited communities sequentially, implementing the intervention over 21 months (March 2009 to November 2010). We randomly assigned communities to one of five phases, thus introducing randomness into the order of treatment.Footnote ³ We did this not only to measure the impact of time since treatment but also to guard against the intervention being interrupted. It was fortunate that we did so—resource constraints meant that UNHCR stopped the intervention after Phase 4. Our control group thus has 160 communities: the 30 randomly assigned to Phase 5 plus the original 130 controls.

Sixteen of the 86 treatment communities in Phases 1 to 4 (26 of the original 116) were also assigned to an “intense” treatment where 25% of community members were targeted for training instead of the standard 15%. The purpose was to determine the marginal effect of increased training beyond the NGO's target of reaching 15% of the adult population.

Because of unexpected delays, 68 of the 86 treatment communities completed their training before the endline survey, all from Phases 1 to 3. Seventeen communities received the training at the same time as the endline, two from Phase 3 and 15 from Phase 4. Only one community, from Phase 4, did not receive the training at all. Running the endline survey concurrent with treatment in Phase 4 was an unfortunate necessity given implementation delays as well as financial and weather constraints. Assignment to Phase 4 was random, however, and so we are able to estimate the effect of concurrent treatment separately.

Communities were located far from each other, with little risk of spillovers between them. A comparison of baseline individual and community characteristics—including demographics, prior levels of conflict and cohesion, and prior exposure to NGOs and education campaigns—shows that treatment and control communities are statistically similar to one another (i.e. randomization was balanced) (Table 1).

TABLE 1.

Key outcomes

Survey Data

We collected baseline data from March to April 2009 and endline data from November 2010 to January 2011. However, we have endline data on only 243 of the 246 communities, because surveyors could not reach two extremely remote villages and one tiny village disbanded before the endline. All three of these villages were in the control group.

The survey was brief and focused on individual and community disputes and outcomes—the incidence, nature, and resolution of disputes—rather than mechanisms. To measure community-level outcomes (e.g., ethnic violence) and traits (e.g., population) we surveyed four leaders from each village at the baseline and endline—typically a town chief and a female, youth, and minority leader.

The intervention and our theory emphasized interpersonal rather than intergroup disputes. Hence the survey focused on individual outcomes and traits. We surveyed random cross-sections of roughly 20 residents per community at both the baseline and endline.Footnote ⁴ Nonresponse was typically less than 5% to 10% per community, and the only attrition came from the three unsurveyed villages.

At baseline, before assigning treatment status, we also asked leaders to propose three “targeted residents”—one elder, one “influential person,” and one “troublemaker”—who would be invited to attend the training if the community were treated. We followed these residents as a panel in both the treatment and control communities, mainly to ensure a minimum sample with a high likelihood of training. Attrition of these targeted residents was 13%. We pooled targeted and randomized residents in our analysis, and so the sample slightly overrepresented persons targeted by the intervention. Because we were measuring impacts at the community level (rather than individual impacts) this did not pose a problem for inference.

Qualitative Methods

We also collected longitudinal qualitative data using several methods to deepen our understanding of disputes and resolution processes, to assess implementation quality, and to generate hypotheses and explore causal mechanisms. First, researchers acted as participant-observers in 15 community trainings. Second, we interviewed 15 facilitators to solicit their opinions on intervention successes and shortcomings. Third, in conjunction with two Liberian research assistants trained by the authors, we interviewed leaders and residents in 20 purposefully selected communities: 15 in treatment and 5 in control.Footnote ⁵

This design allowed us to compare treated communities before and after the intervention, and treatment and control communities to one another. We conducted 104 interviews with 52 respondents between April 2009 and December 2010. We interviewed town leaders plus a convenience sample of community residents and trainees. Sampling was purposefully unsystematic, providing a wide-ranging sample.

The interviews followed a semi-scripted, open-ended questionnaire covering a range of topics, including dispute behavior, community relations, and reactions to the intervention and subsequent use of ADR in the community. In treated communities, we typically interviewed respondents twice: before the training and several months after it. In control communities, we attempted to interview the same individuals twice as well. Interviewers took detailed notes and recorded interviews, which were then transcribed, reread, edited, and annotated using a set of thematic coding rules that reflected key program outcomes and other factors hypothesized to influence outcomes.Footnote ⁶ After the initial coding and analysis, we selected and coded additional categories we believed to be important.Footnote ⁷

Empirical Strategy

Given that our survey took place 1 to 21 months after the intervention, we focused on short-term effects. To examine Predictions 1 to 8, we calculated average treatment effects (ATEs) for self-reported dispute outcomes. By far the most important and common disputes were those over land, followed by money, and so our survey concentrated on these two types of disputes. In part because of the need for such a large-scale survey to be brief, we obtained minimal survey data on ADR norms and skills. As a result, we were mostly unable to test Prediction 6.

Nonetheless, we did collect data on the incidence of disputes in the previous year (Prediction 7), whether each dispute resulted in violence (2), whether at the time of the survey the dispute had been resolved (1 and 4), whether it was resolved through an informal institution (3), and whether the parties were satisfied with the resolution (5). We did not have data on the specific length of the dispute, but one implication of Predictions 1 and 4 is that at any point in time we should observe fewer unresolved disputes (especially relative to the total number of disputes). We did not have direct data on bias toward low-powered groups (7), but we did look for evidence of lower impacts among youth, women, and minorities.

We also looked for early indications of the long-term goals of the intervention. We tested whether the treatment effects diminished over time (using random assignment to phase as an instrument for months since the intervention). We asked respondents to report their perceived land security and major investments (9 and 10).

Finally, and importantly, we looked for changes in behavior of the untrained residents (11). Absent random assignment of residents to training, we could not identify the direct causal effects on trained versus nontrained residents. We developed a technique, however, that bounded the effect on untrained residents.

ATE Estimation

Our preferred ATE estimator was the average treatment effect on the treated (ATT) measure, which uses random assignment as an instrument for being treated.Footnote ⁹ We estimated the ATT using two-stage least squares regression, controlling for concurrent treatment (instrumented with assignment to Phase 4), a vector of baseline covariates,Footnote ¹⁰ and district fixed effects. We clustered standard errors at the community level. The ATT thus included the direct effect of the intervention on trained residents plus spillovers onto untrained residents, and it averaged earlier and later treatment. We did not weight by population sampling. We tested for robustness to intent-to-treat (ITT) and alternate ATT estimates, probit estimation, population weights, and exclusion of controls (Appendix Table 4).

Measurement Error

All outcomes were self-reported. Therefore if disputes were underreported, then the ATE would be underestimated. We were more concerned with measurement error that is correlated with treatment. If training led to social desirability bias (so that residents underreported disputes or repeated back norms) then the ATE would be overestimated. Although this is certainly a risk, we saw that the pattern of effects we observed was inconsistent with social desirability bias: Even though some ATEs were consistent with the normed messages of the intervention (e.g., less property destruction), other normed messages showed no change (e.g., egalitarian attitudes), and residents even reported an increase in illegal behaviors (e.g., trial by ordeal).

Communication and Self-Enforcing Agreements

First, we observed a change in discourse about how communication can help resolve disputes and about who should make that effort. Respondents were asked to describe recent disputes and the process of resolution. In control communities and in pretraining interviews, several raised the importance of communication, but most often in reference to a community leader (referring, for instance, to his skill as an orator). In contrast, several treated respondents explained not only how facilitated communication was an important element of the dispute resolution process but also how it is an individual's responsibility and role to engage in that communication. A major change was the idea that ordinary residents have as much legitimacy and ability to resolve disputes as traditional authorities. One interviewee explained, “I bring people together, I tell them to communicate, to bring their position forward until they can reach an agreement” (MM, Zwedru, 5/1/2009). In treatment communities we also saw an increase in people reporting direct engagement in their own disputes.

Second, we found evidence of a shared vocabulary of dispute resolution in treated communities. For instance, when describing the process of reaching agreements, we observed respondents in trained communities using words and phrases such as “bringing people together,” “talking,” “finding the common way,” “working as one,” “internal conflict,” “external conflict,” “win-win,” and “calming them down.” Respondents may have been merely repeating lessons learned in the workshop, and in some cases this was our impression. In other cases, however, the vocabulary came up naturally in a narrative description of a respondent's experience with a particular case, emphasizing not only communication but also mutually agreeable bargains (such as “win-win solutions”).

Promoting Rational Dialogue and Behavior

Third, we saw signs that the program helped residents manage their emotions, identify and experience empathy, and increase their recognition of wrongdoing by both sides. In control and pretreatment interviews, for instance, not one respondent mentioned apologizing or admitting wrongdoing as part of the dispute resolution process. In contrast, several posttreatment respondents demonstrated such self-reflection. For instance, one respondent explained, “If I have offended someone, I must be able to realize that I did something wrong. Both parties must admit it and say ‘I am sorry'” (GP, Barteh Jam, 2/15/2010). We also saw some evidence that, in treatment communities, interviewees exhibited an increased capacity for self-reflection and empathy. In no control communities did the respondents mention the importance of seeing the problem from the other person's perspective.

Similarly, respondents in treatment communities described how self-control helps mitigate conflict. One interviewee described another resident after the workshop: “Now if he gets angry, for example when his children disobey him, he remembers the workshop, he thinks about the things the workshop leader told him and he tries to control himself” (EB, Lawalazu, 3/20/2010). Consistent with a new inclination to empathy and anger management, several interviewees spoke of reconnecting to brothers, wives, and other family members with whom they had contentious relationships in the past.

Our impression is that recognizing wrongdoing and managing emotions helped respondents express more open views about opposing parties. For instance, when describing another party to a dispute, they would highlight that person's humanity, making statements such as “we are all human.” This response echoed a lesson from the workshop, which encouraged people to focus on similarities not differences when they are divided by conflict. We must weigh this impression, however, against the absence of survey evidence for a large change in egalitarian and progressive attitudes, especially a reduction in prejudice (Table 8 and Appendix Table 6).

Decentralized Negotiation and Mediation

Fourth, after the intervention we noticed an increase in ordinary residents’ involvement in others’ disputes and of disputants engaging directly with one another rather than through third parties. Before the intervention and in control communities, it was customary for disputants to say they would take their cases to “powerful people.” One problem, however, is that the disputants seldom agreed on the appropriate authority. In a separate study of land dispute dynamics in the same counties, one of this article's authors shows that a serious obstacle to dispute resolution is disputants’ inability to commit to one authority's process (Hartman Reference Hartman2012). In 35% of cases, individual disputants brought their problem to different authorities when first trying to resolve their dispute, and in 20% of cases individual disputants visited three or more authorities to try to resolve their dispute. When asked why they went to different authorities, respondents commonly described authorities’ corrupt behavior and lack of transparency. Frequently, they also accused different authorities of “favoritism” toward one group or another. In spite of these problems, respondents in the control and pretreatment interviews in our study seldom raised instances where they attempted to tackle their own disputes directly outside these traditional forums.

Lack of transparency and incidents of favoritism by traditional leaders were by no means absent in communities that hosted the workshop. During interviews after treatment, however, we observed a shift toward more decentralized mediation and negotiation. Respondents gave specific examples of how they personally helped resolve problems and how disputants accepted their intervention. These were among the most common instances of change that were described. For example, in one community, a respondent explained how she had intervened during a dispute over livestock between two neighbors and that she helped them find a mutually beneficial solution—a new experience and new role for her (MM, Zwedru, 5/1/2009). Another explained that bringing people together was his “favorite lesson from the workshop” and something he regularly did since the training (AZ, Toe Town, 10/3/2010). We also witnessed the development of new informal structures designed to promote dialogue in the community, such as “peace groups.” In a village in Lofa, one participant explained, “After the workshop, we sat down together. . . we decided that in order to work together we need to organize a club. . .we never had a club in this town here, but after the workshop were able to establish one” (TS, Shandadu, 9/3/2010).

Self-enforcing rather than Enforcing Agreements

Our fifth observation is that the nature and objectives of informal third-party interventions shifted away from adjudication and toward mediation. Before the workshop, disputes were generally taken to customary authorities, who mostly used a combination of mediation and adjudication, often without being able to describe how they resolved a dispute other than saying that they had “cut’ (decided) a case. After training, respondents in treated communities appeared to hold different expectations of authorities’ roles: The appropriate objective was to bring parties together to agree on a solution as opposed to working with just one party for a judgment. One resident explained, “I am not adjudicating a case to decide who is right. Instead I bring people together, I tell them to communicate, to bring their position forward until they can reach an agreement” (MM, Zwedru, 5/1/2009). Similarly, another noted of participants and leaders who were trained that “now they are available and it is okay for a person to go to them, and these people use the same skills they learned in the workshop and talk to both people involved in the dispute to solve it” (AG, Toe Town, 10/3/2010).

Sixth and last, respondents in treatment communities also spoke about pressuring disputants to commit to solutions that did not require external punitive enforcement. The program training manuals repeatedly emphasized that disputes resolved through ADR are self-enforcing, because both parties agree to a solution that serves their interest. In interviews following treatment, respondents spoke about engaging with disputants and then pressuring them to commit to working through their problem until both parties were satisfied with the resolution. In contrast, no control or pretreatment interviewees focused on this aspect of dispute resolution. Instead, they focused on punitive methods and norms. Indeed, threatening fines and the external adjudication of disputes are the most common approaches to dispute resolution in rural Liberia. As one interviewee in a control community put it, “If you're wrong then you will pay that fine and when the town people call you to pay that fine if you refuse then they'll carry you to the town chief” (J.J., Barglor, 6/1/2009).

Unintended Consequences

Finally, although we did not collect explicit qualitative data on the unintended consequences that we found in the quantitative results, we did observe that the workshop inflamed disputes between youth and elders. Discussions of equal rights in the community gave a space for traditionally low-powered groups, such as youth, to speak up and make complaints about the status quo with support from the workshop facilitators. These opportunities led to passionate and sometimes unresolved debates about whether new ideas about “human rights” and sharing power were suited to the community. These observations could explain the increased incidence of disputes, especially if the intervention increased the willingness of certain groups to stand up for their rights rather than submit to existing power structures.