Does legal aid improve access to justice in ‘fragile’ settings? Evidence from Burundi

Access to justice is often described as key for building and consolidating peace and enhancing socio-economic development in fragile and post-conflict states. Since the 2000s, legal empowerment has been one of the most popular approaches to improve such access, and a growing literature has presented mixed evidence on the quality of its outcomes. We evaluate and discuss the impact of a locally provisioned legal aid program on justice-seekers’ use of dispute resolution fora, legal agency, and trust in judicial institutions. The program was implemented between 2011 and 2014 in 26 municipalities of rural Burundi. We consider its effects on 486 beneficiaries using various propensity score-matching methods and data on non-beneficiaries from two distinct control groups (n = 3,267). Forty-eight interviews with key informants help discuss judicial practices. We find that the program increased the use of courts but not trust in the judiciary. It had no significant impact on the use of alternative dispute resolution mechanisms. Qualitative and quantitative evidence suggests that justice-seekers’ perception of the treatment they received in courts, also known as procedural justice, shaped their perception of accessing justice. Qualitative evidence also points to a possible ‘watchdog effect’: in some cases, the presence of a legal adviser may have pushed judges to better comply with procedures. While legal aid programs can improve access to courts, it does not necessarily mean an erosion of judicial ‘forum shopping’ or that trust in state institutions is reinforced and rights fully realized.


Introduction
Effective access to justice through dispute resolution mechanisms is widely regarded as fundamental to enable social cohesion, sustainable socio-economic development, and peace (UNDP, 2000). Instability thrives on unresolved criminal, but also civil, disputes (Deininger & Castagnini, 2006). In the African Great Lakes region, for instance, land disputes are commonplace and amplify broader cleavages associated with violence and civil war (van Leeuwen & van der Haar, 2016). In the last two decades, states' and international aid's main interventions to improve access to justice in so-called 'fragile' settings have been the support or set-up of transitional and criminal justice processes and (para)legal aid mechanisms (Gisselquist, 2019). Research has looked at the national-level effects of legal aid (Carothers, 2003;Prettitore, 2015), but less is known of its micro-level dynamics (Prettitore, 2012) -especially in rural postconflict contexts that are also often marked by legal pluralism (Harper, 2011;Goodwin & Maru, 2017). This article seeks to advance the debates on legal aid and its capacity to improve access to justice through the evaluation of the Intercommunal Legal Aid Service (Service d'Aide Juridique Intercommunal; henceforth SAJI) implemented in rural Burundi between 2012 and 2015. The SAJI was typical of legal aid interventions: it consisted of the deployment of trained professionals to provide free legal services to populations without access to lawyers.
Burundi's situation in the first half of the 2000s was characteristic of post-conflict countries: state judiciary institutions had limited capacity, legitimacy, and, to some extent, authority (Stensrud, 2009). A preliminary objective of the article is, therefore, to better comprehend how judicial and non-judicial dispute resolution mechanisms were used and operated. It is only then that our focus can switch to the main objective of the article: an evidence-based discussion on the effectiveness of legal aid in helping citizens accessing justice. We proceed in two steps, as we consider two main dimensions of 'access to justice': the use of justice institutions -and especially courts in the case of legal aid -and the question of whether experience in courts leads to achieving justice, meaning an effective resolution of disputes and the realization of rights for justice-seekers. The second dimension is notoriously tricky to explore with only survey data -we use the justice-seekers' assessment of the fairness of the judicial process, their intention to use courts in the future, and their trust in institutions as proxies. Therefore, we also integrate substantial qualitative research, which allows exploring frontline officials' and judges' behaviors. The article provides new insights into the role of legal aid and revisits the meaning of 'access to justice' in a postconflict context.
Our study shows that legal aid beneficiaries are more likely to use courts of justice, but not other dispute resolution fora, especially when they have weaker social networks. Their trust in courts and legal institution is no different from non-beneficiaries -unless they have received substantial information regarding their rights and the legal proceedings while in court. Our qualitative evidence documents the interplay between local justice actors and suggests a possible watchdog effect in some cases: legal aid professionals are de facto monitoring judicial actors, thereby compelling them to fairer judgments and legal compliance.
Section 1 discusses the literature on legal aid in fragile contexts and access to justice in Burundi. Section 2 presents the dataset, which is then used to further characterize the context. Section 3 describes the SAJI. Section 4 presents the impact evaluation framework. Section 5 and Section 6, respectively, describe and discuss the main empirical results.

Literature and rationale
Improving access to justice and legal aid interventions Since the 1970s, abundant literature has underlined that access to justice is not only the effective use of state judicial institutions (e.g. UNDP, 2004;Bedner & Vel, 2010). Beyond the question of the ability of state institutions to deliver and enforce fair rulings (Rhodes, 2005), social scientists have stressed that justice is also sought and delivered through non-judicial institutions such as customary law, community arbitration, and other alternative dispute resolution (ADR) mechanisms that do not involve the judiciary (Glasser & Roberts, 1993;Le Roy, 2004;Ubink, 2011). In many contexts, and especially when state institutions are perceived to be 'fragile', non-judicial solutions are frequently depicted as more accessible, more predictable, fairer, and more concerned with social reparations (Chauveau, Le Pape & Olivier de Sardan, 2001;Harper, 2011). However, legal pluralism, the coexistence of several normative orders and dispute resolution fora, can also negatively affect the chances of justice-seekers solving their case: justice 'forum shopping ' (von Benda-Beckman, 1981) often leads to contradictory rulings between fora, and the multiplicity of fora is hard to navigate for the poorer and disenfranchised (Golub, 2003).
A key approach of the last decades for improving access to justice in such contexts has been legal empowerment. It builds on the critique of the 'rule of law [improvement] orthodoxy' (Golub, 2003: 1) as incomplete and insufficient to genuinely improve access to justice for all (Samuels, 2006;Krever, 2011) and advocates connecting the top-down/supply-side and the bottom-up/demand-side -people's use of the law and ADR (Gisselquist, 2019). In practice, the approach has focused on rights enhancement, which involves enacting laws that protect and improve the status of the most vulnerable in the judicial system and may push nonjudicial authorities to reform their judgments as well (Aldashev et al., 2012;Cecchi & Melesse, 2016). Importantly, the legal empowerment movement has also translated into rights awareness and enablement initiatives such as education campaigns, counselling, and (para)legal aid, which are all expected to improve legal agencythat is, people's ability and willingness to take action to solve their disputes and realize their rights (Goodwin & Maru, 2017). Beside legal agency and legal awareness (justice-seekers' understanding of their legal rights and options), rights awareness and enablement initiativesespecially when mobilizing a third party that acts as a facilitator or adviser (e.g. paralegals) -may also improve access to justice by improving people's trust in the judicial processes they go through and/or influencing the behavior of the judicial actors they face, such as judges. Similarly, it may improve procedural justice, which Lind & Tyler (1988) define as the perceived fairness of dispute resolution processes. Procedural justice has to do with the application of legal procedures, but also with the justice-seekers' confidence that their voices are heard by a neutral judge who seeks to understand the needs and concerns of all parties, and the fact that information is perceived as transparent and sufficient (Tsuchiya et al., 2005).
Empirical evidence on the effects of rights awareness and enablement is growing but still limited in scope: it tends to be based on small-scale studies and focused on democratic contexts (see Goodwin & Maru, 2017 for a recent review). In fragile contexts, the body of evidence is optimistic but also sparse. Sandefur & Siddiqi (2013) found that paralegals increase access to state legal systems in Liberia, and further work in Liberia by Blattman, Hartman & Blair (2014) found that an educational campaign advocating for ADR led to shorter and less violent land disputes and a decrease in violent youth-elders disputes, but also more extra-judicial punishments. Other studies present some evidence that community paralegals have a positive impact on settlement outcomes, litigant satisfaction and livelihoods, and intracommunity relations (Gramatikov et al., 2015;Maru & Gauri, 2018). This same literature also points to limitations with the approach, including in terms of the potential of legal empowerment to improve the conditions of those with limited social capital (Kapur, 2011) or to deliver immediate, short-run benefits (Mueller et al., 2018). Overall, the recognition is that legal empowerment approaches need to be 'twin-track': empowering the poor but also the frontline state officials (Waldorf, 2019: 441). Tanner & Bicchieri (2014) suggest that capacity-building work with justice officers -for instance, by explaining laws and how they are implemented in practice -is essential to foster changes in attitudes.

Accessing justice in Burundi (and in fragile settings)
Burundi constitutes an interesting case study. As in many fragile and post-conflict contexts, disputes are abundant and often related to land and forced displacement (Kohlhagen, 2009;Falisse & Niyonkuru, 2014;Nyenyezi Bisoka & Giraud, 2020;Tchatchoua-Djomo, van Leeuwen & van der Haar, 2020). Access to justice is limited for at least four reasons that echo the situation of other fragile and post-conflict countries.
First, the population has a limited understanding of official rules and laws (Ministère de la Justice du Burundi, 2005). In a recent legal analysis, Niyonkuru (2019) notes a significant gap between the 'proliferation of laws' and access to justice, and points to the population's many uncertainties about the functioning of judicial mechanisms. The current legal system is largely derived from Belgian law applied to non-Burundians during colonial times. Its appropriation by the administration has proven difficult, if only because it is mostly in French -a language fully understood by merely 12.5% of the population of Burundi according to the 2008 census. Recent reforms may not have improved this situation: Tchatchoua-Djomo (2018), who looked at land governance reforms, argues that reforms fueled confusion and institutional competition in a context of legal pluralism.
Second, there is significant mistrust vis-à-vis all official justice actors (Mukuri, 2002;Kohlhagen, 2011), which is best understood as 'legal cynicism' in the sense of 'skepticism about law and the actors who make and enforce it' (Gau, 2015: 404). This may contribute to explaining why, in a transitional justice survey conducted in 2007 focusing on the perception among the population of former combatants who perpetrated killings and rapes, Samii found a large preference for forgiveness and forgetting (Samii, 2013). Our study takes place seven years after Samii's and focuses mainly on cases non-directly related to war crimes, but we also find a low level of trust in courts (42.28% of respondents). We suggest that legal cynicism is not necessarily manifested by avoiding courts, but rather by engaging with many other dispute resolution mechanisms (see next section). Focusing on land disputes cases, Nyenyezi Bisoka & Ansoms (2011) explain how justiceseekers are, in practice, navigating 'real governance', a space where norms are not necessarily those theoretically applicable (and keep evolving as power relationships evolve).
Third, the literature on Burundi depicts an overwhelmed judiciary system. Kohlhagen (2009) explains that the backlog has roots in both a history of forced displacement (1972,1988,1993) that has created layers of land disputes and increased land insecurity, and Burundian law, which allows many means for proving land ownership (e.g. testimony, paper, location of a landmark), thereby generating longer debates and more frequent appeals.
Fourth, legal assistance is limited. In 2014, the Burundi bar association counted fewer than 400 members, of which over 250 had joined in the last two years. They were also overwhelmingly based in the capital city. The survey data we present below shows that less than 1.77% of justice-seekers who went to court were represented by a lawyer. Although the right to counsel is technically guaranteed by various international instruments and domestic law, legal assistance had never been implemented on any large-scale basis (ISSAT, 2014). Qualitative evaluations of past small-scale pilot legal aid programs suggested that legal advisers' activity may positively affect judges' behavior and the compliance of legal decisions with the law (Galand, 2007;Moriceau, 2016).

Data collection
Our study uses data from a household survey and semistructured interviews and focus groups with actors directly or indirectly involved in SAJI implementation and dispute resolution.
The survey was conducted among 3,245 heads of households in November 2014 as part of an evaluation of the European Union justice support program. The enumerators were professionals recruited in Bujumbura and underwent a four-day training on the survey instruments. The survey took place in: (1) municipalities randomly selected among the municipalities that benefited from the program in the provinces of Gitega, Karuzi, Bururi, and Ruyigi and (2) non-beneficiary municipalities selected because of their similarity with the beneficiary municipalities ( Figure 1). There is no significant statistical difference between the two sets of municipalities in terms of population (including ethnicities), violence, economic development, and vote in the 2010 elections (see Table O1, Online appendix).
Eight collines (or 'hills', a low-level administrative unit in Burundi) were randomly selected in each municipality, and 30 households were randomly selected in each hill (starting from the center of the hill and randomly selecting a direction and distance to fetch the next respondent, and replacing those who declined to participate with their neighbors). The statistical analysis takes into account these strata. The tablet-based survey lasted for around 40 minutes and was introduced as a survey on the usage of justice services. The study was green-lighted by the Burundi National Institute of Statistics.
In our impact evaluation below, we use the aforementioned survey and an additional survey of randomly selected SAJI beneficiaries using beneficiaries lists. The same questionnaire and team of enumerators were used for both. The total number of SAJI beneficiaries of the study is 486, out of which 58 were coincidently interviewed in the general survey. In addition, 48 semi-structured interviews were carried out with key stakeholders, including provincial judicial authorities, judges, clerks, SAJI program staff members, municipal administrators, civil servants, customary judges, and civil society representatives in the different municipalities (see Table O2, Online appendix). Each interview lasted for around one hour and a half and was conducted in French, except for a dozen Kirundi interviews with civil society organizations and local administrative leaders led by the survey coordinator. The qualitative material was used to refine the research questions, guide the quantitative analysis, and interpret the empirical results.

Disputes and justice pathways in Burundi
We begin by describing the reality of seeking justice in rural Burundi using the survey data.
Type of disputes. Seventy-two percent of the surveyed households had experienced an issue serious enough that they thought it could potentially be brought before a court. In 95.45% of cases, it had arisen less than six years before the survey. Figure 2 shows disputes overwhelmingly related to land and often lasting years. This is in line with the aforementioned literature and the Ministry of Justice reporting that 69.5% of the cases in courts of appeal (2013) and 80% of in the Supreme Court (2009) are land related. In a country where the livelihoods of 93% of the population depend on agriculture (2008 census), land disputes have a very direct repercussions on livelihoods.
Justice pathways according to the law. By law, there are three key justice actors at the local level in Burundi: (1) the hill council (French: conseil de colline), (2) the customary judges (Kirundi: bashingantahe), and (3) the municipal court (French: Tribunal de Résidence).
The hill council is made up of six elected members including the hill leader (French: chef de colline). It is responsible for 'ensuring arbitration, mediation, conciliation and neighborhood dispute resolution on the territory of the hill' but there are no specific rules or procedures pertaining to those mediations and conciliations (République du Burundi, 2010: art 36).
The council of custom-derived hill notables or bashingantahe has been a local actor in justice and administration since pre-colonial times (Ingelaere & Kohlhagen, 2012). A legal framework delineating the role of these 'wise-men' who mediate civil and neighborhood disputes was carved into Belgian law as adopted by Burundi during the decolonization process. Their importance has, however, been on the decline since independence and especially, after the 1993-2003 civil war (Matignon, 2014). Since 2010, Burundi law mentions the bashingantahe as local mediators only.
The municipal court is, by law, the lowest (or first) level of the judiciary and is responsible for dealing with land issues as well as all minor civil issues (civil and family status, family disputes, etc.) and minor offences (insult, minor damage). Each ruling in a municipal court is made by three judges, along with a president, in accordance with Burundian law. One municipal court sits in each municipality (commune), and decisions can be appealed to the provincial court (tribunal de grande instance) located in the provincial capital. It is worth noting that all fora are heavily male dominated (only 10% of judges were women according to Kohlhagen (2010), and bashingantahe are, by definition, men).
Justice pathways in practice. In practice, the situation is more confused. Anecdotal evidence suggests that municipal courts sometimes only agree to open a new case if the disputants have already attempted to mediate their case at the hill level, even though the law considers  Table O3, Online appendix. this mediation level optional (Kohlhagen, 2010). The lack of trust in justice institutions and the feeling that local political and administrative actors often interfere in dispute resolution processes have further been highlighted by many actors and reports, including by the Government of Burundi itself (Ministère à la présidence chargé de la bonne gouvernance, 2008). De facto, at least two other actors are involved in justice processes: local police forces and the municipal administrator who legally controls local police forces and supervises hill leaders. They are frequently described as arbitrators for different types of disputes, and they may or may not refer the case to other dispute resolution fora. Therefore, we include them as one of the dispute resolution mechanisms available to justice-seekers.
The municipal courts function at a crossroads between non-state and state justice. Only a minority of municipal judges (mostly the newly appointed ones) have a law degree, which frequently casts doubt upon the municipal court's decisions and proceedings' full legal compliance with the letter of the law. Since 2005, the Ministry of Justice and international donors have carried out considerable efforts to support municipal courts (Ministère de la Justice du Burundi, 2010). Courthouses have been rebuilt and a wide range of legal training, covering all major aspects of Burundian law, has been provided to judges. However, judges' remuneration levels remain low, and they do not receive training in mediation and ADR to the same extent that their counterparts do in other East African countries (Kohlhagen & Kanyonga, 2015).
Our survey found that people often use a combination of different dispute resolution fora, which we have categorized as local 'court' (tribunal de residence), 'administrative' (police, administrateur communal, chef de zone), 'customary' (bashingantahe), and 'community' based (hill councils, family, neighbors). The court is technically the state's only legal dispute resolution forum (the case of the bashingantahe and hills councils is slightly ambiguous). Figure 3 shows how the court, administrative, and customary fora (colored bars) are widely used across case types. Figure 4 shows the breakdown of case types (colors) according to each of the 16 possible pathways, that is, the combinations of fora (black dots below the bars). The variety of pathways, for each case type, is evident (see Table O4, Online appendix, for breakdown by region). The data do not provide information on sequencing, but anecdotal evidence shows that, depending on the case's nature and the recommendations of the traditional, community, court, or administrative judges, justice-seekers engage with the different fora either concurrently or sequentially. It is in this context, where courts are used as a sole dispute resolution mechanism in just 7.81% of cases, that the SAJI was implemented.
The SAJI: a 'local lawyer' The SAJI was part of the Gutwara Neza good governance program jointly implemented by the European Union and the Government of Burundi. Its aim was to provide legal aid to rural communities and especially to justice-seekers in vulnerable situations. SAJI's services were provided by a legal adviser hired jointly by the municipal administration and Gutwara Neza. The advisers had a contract with and were paid by the municipality (the European Union funded 90% of their salaries). They were required to live in their municipality of assignment and hold a university degree in law, but they were not lawyers -they could not officially represent people in court (but they often did accompany them to courts). The official role of the legal advisers was threefold: (1) organize awareness session about legal procedures, (2) provide legal advice services, and (3) support and advocate for the interests of the disenfranchised (the so-called indigents).
All justice-seekers were entitled to individual legal advice from the SAJI. The program's offices, located in the municipalities' offices (close to the municipal court in all cases but one) were open for free consultation one to two days a week. Minor issues would usually be solved by the direct intervention of the adviser and take one or two meetings. In some cases, they involved more than advice during the SAJI office hours: in family or civil issues, advisers would often organize a mediation or refer justice-seekers to mediation bodies, while in administrative cases, for instance, obtaining official documents such as birth certificates, it would not be uncommon for the adviser to visit the administrative authorities together Figure 4. Combination of dispute resolution fora used by justice-seekers, with case types with the justice-seeker. For more complex issues, the adviser would mainly give recommendations on the procedures and the rights of the different parties and refer them to the relevant judicial or administrative body. Only the justice-seekers recognized as 'indigents' -based on an official certificate and an assessment of resourceswere entitled to long-term legal assistance, including representation before the courts and support statements during the entire resolution process.
The SAJI was set up across 20 municipalities from late 2010 to late 2011 and was terminated at the end of 2014. The selection of the municipalities was based on (1) their inclusion in the geographical area of intervention of EU programs in Burundi, and (2) their readiness to join the program. Gutwara Neza initiated SAJI implementation by first approaching municipalities it had been working with and whose territorial administration was seen functioning well enough to support the program. If willing to participate in SAJI implementation, the municipality had to actively cooperate at every planning stage. Each adviser was responsible for managing and dividing his or her time between three municipalities, all of which contributed a portion toward his or her salary and out-of-pocket costs (e.g. travel and overhead costs). In many municipalities, the SAJI was a profound change: for the first time in Burundi history, a law graduate was available at a local level to give advice and provide services to the population, as well as interact with justice and administrative representatives.
The timeframe of the SAJI implementation is important to bear in mind: it took place seven years after Samii's transitional justice research, at a time of relative stability and freedom in Burundian society. Six years after the end of the SAJI, at the time of finalizing this article, the situation had changed again. The space for civil liberties had considerably shrunk, and it is likely that the issues that we explore in this article are only exacerbated. In fact, recent research has suggested that the unmet demands for justice and the dysfunction of dispute resolution mechanisms contributed to the frustration that led to the political crisis that erupted in 2015 (Razafindrakoto & Roubaud, 2015;Moriceau & Coster, 2019). Now that we have described the context of legal pluralism in Burundi and the SAJI legal aid program, we turn to explaining how we evaluated this program using the data described at the beginning of section 2.

Measuring the effects of legal aid
The quantitative evaluation focuses on the effect of the SAJI on legal pathways, which we study by looking at the use of different dispute resolution fora. This can be done through the evaluation of a linear probability model: where b is an estimator of the SAJI's effect on the outcome variable y i , which in our main estimations is a binary variable reflecting the use of a specific dispute resolution forum. X is a vector of observable characteristics of the household and the dispute case (professional activity, type of case, parties involved, distance to court), which includes controls for the observed differences between the control and intervention group (see Table I). We further control for the possible self-selection of beneficiaries -households that benefited from the program may have different characteristics than those who did not -by using a propensity-score matching approach to create an adequate control group (since conditioning on all relevant covariates is limited in case of a high dimensional vector X i , see Rosenbaum & Rubin's (1983) curse of dimensionality). Such an approach makes a strong hypothesis on the absence of selection bias arising from unobserved characteristics (such a hypothesis is, by definition, impossible to test) and we, therefore, use two counterfactuals that each presents a different potential bias. The first group is composed of justice-seekers from the intervention municipalities who did not participate in the program likely because, as qualitative evidence suggested, the program was not well advertised. Such a group is unbiased if knowing about the SAJI is random, which we cannot guarantee (or test). The second group is composed of justice-seekers from municipalities where the SAJI was not introduced. Here the bias has to do with potentially slightly different justice contexts.
Following the approach described in Becker & Ichino (2002), we selected matching variables -professional activity, type of case, parties involved, distance to court (see replication files for detail) -with the objective of obtaining an estimate of the propensity score that satisfies the balancing hypothesis. The matching variables are all very unlikely to be directly affected by SAJI participation. Two matching methods, nearest neighbor and kernel (non-parametric matching estimator), are used to mitigate the trade-off between bias and efficiency (Baser, 2006). In the robustness checks, we also used a covariate matching technique based on the Malahanobis distance, a measurement of similarity between two individuals in terms of covariate values.
In the third part of our empirical quantitative analysis, we explore the existence of heterogeneous effects. We use the propensity score obtained from the first stage of the matching models in a simple regression model of observed outcomes on intervention status. The variable that indicates the intervention status is then interacted with confounding factors (X i ): b 4 confirms whether a heterogeneous treatment effect exists. Standard errors are bootstrapped (Söderbom & Teal, 2015).

Effects of legal aid
Targeting the poor? All SAJI advisers interviewed described their central mission as helping marginalized justice-seekers to understand the proceedings and the role of judicial institutions: Marginalized people often fear institutions and are reluctant to resort to courts and administrations. Being Standard deviations in parentheses. Significance levels: ***p < 0.001, **p < 0.01, *p < 0.05; difference between the control group (I or II) and the intervention group using a t-test. a From none (0) to university (5).
in touch with the poorest and standing by them to gain their confidence was the first step of my work. I encouraged them to undertake procedures to solve their disputes. (SAJI facilitator) Beyond the intent, we found qualitative and quantitative evidence that the SAJI indeed benefited the poorest people. During the interviews and focus groups, administrative and justice actors acknowledged that the SAJI did reach the poorest. As a municipal employee explained: Before the implementation of SAJI, no one here in the municipality took care of marginalized people's issues. SAJI advisers showed that justice could also serve marginalized people's interests. Now, justice actors know that the poorest people can use justice institutions and be counseled. (civil servant, Bugendana municipality) Table I presents the average profile of legal aid beneficiaries compared with the rest of the population sampled in the intervention and control areas. Legal aid beneficiaries appear poorer in wealth and nutritional terms but not in terms of social capital. They are also more likely to belong to an officially recognized category of protected vulnerable persons: 'indigents' and displaced people (returnees and internal displaced persons, IDPs). Figure 5 presents the estimation of the SAJI effect on using the four main dispute resolution fora. Beneficiaries appear 14.54% to 19.71% more likely than nonbeneficiaries to use municipal courts (depending on the model specification), but the SAJI does not have robust effects on their use of other fora. There is no evidence of a substitution effect of legal aid, that is, that the SAJI pulls justice-seekers away from other dispute resolution mechanisms. These findings are consistent with both our earlier depiction of justice-seekers using several local dispute resolution mechanisms in parallel and the fact that the SAJI was a legal aid program oriented towards municipal courts. The skills and knowledge of the SAJI advisers are likely to be both more recognized and more useful in courts than in other dispute resolution fora. As non-native of the local area, SAJI advisers may not be very legitimate in, or even familiar with, customary processes. According to interviewees, SAJI advisers, local leaders, and customary leaders were in contact, but SAJI advisers had limited clout in nonjudicial fora.

Evaluation of the SAJI legal aid program
We then explored whether the SAJI makes a difference in terms of attaining justice by looking at three additional outcome variables: likelihood of going to court next time is a binary variable that takes value 1 when the respondent expressed their intent to use the local court next time they have an issue the court could potentially address; trust in court is a categorical variable that measures the level of trust of the respondent in the court on a Likert scale ranging from 1 (not at all) to 5 (totally); and the sense that courts are fair, based on their ranking of the court vis-à-vis six other actors (1 to 7, with 7 the fairest).
None of these variables is affected by the SAJI intervention, which points to the relative inability of the SAJI to meaningfully improve access to justice. These results, as well as those presented in Figure 5, are robust to several estimation techniques. The detailed results of the models used in these figures, as well as sensitivity analysis using linear probability, logit, and Mahalanobis distance models, are in Tables O5 and O6 in the Online appendix. The estimations of Figures 5 and 6 use dispute cases that started after the implementation of the SAJI program only; this mitigates a potential endogeneity issue arising from the fact that justice-seekers who already had a case in court at the time of the implementation of the SAJI program may be more likely to know about the program and consult the SAJI adviser. Using the full sample gives similar results.
The average effects we presented in this section hide complex local dynamics and interactions between local politics and access to justice (see, for instance, Nyenyezi Bisoka & Ansoms, 2011). It is not possible to explore all configurations in this article, but the main effects do appear to hold in each province: the one province that was a stronghold of the opposition (Bururi) sees people using courts and customary institutions more, but the effects of the SAJI are not substantially different (see Tables O4 and O7 in the Online appendix).

Heterogeneous effects
We now consider the possibility that justice-seekers with different profiles may be affected differently by the legal aid program. First, given the SAJI's ambition to help the most vulnerable, we explore whether one's level of vulnerability affects the effectiveness of the SAJI intervention. We consider heterogeneous effects on the use of courts depending on the SAJI beneficiaries' (1) social network, via a measure of the number of local notables known by the respondents (0 to 7) and whether they know the probably most influential of them, the municipal leader (administrateur communal); (2) vulnerability, and in particular whether they are forcibly displaced people; and (3) gender. As Table A1 in the appendix shows, social network matters: justice-seekers who have a network among notables or know a municipality leader are generally more likely to go to court. However, when exposed to SAJI, those well-connected justice-seekers are not more likely to go to court than SAJI beneficiaries with no personal connection with notables. A possible explanation may be that justice-seekers mobilize legal aid services and municipality leaders in similar circumstances. The other interaction effects are not significant: among SAJI beneficiaries, women and forcibly displaced persons are not more likely to go to court.
Second, we focus our attention on the treatment of the beneficiaries' case in local courts and the judicial outcome and whether it affects their intention to use courts in the future -a proxy of whether they consider it a way to access justice. We consider interactions with three variables: (1) whether the judgment was favorable to them; (2) whether the case is resolved (closed); and (3) whether the justice-seeker said they were well received and/or received proper information in court -our proxy for their treatment in courts. We only consider the case of justice-seekers who indeed went before a court. As shown in Table A2 in the appendix, whether the outcome is favorable and whether the case is closed do not Figure 6. Effects of the SAJI: intent to use, trust in, and ranked fairness of courts Bars are the 95% confidence interval (full results in the Online appendix). matter significantly. Justice-seeker's treatment in courts, however, appears important. Among both the beneficiaries and non-beneficiaries, justice-seekers who said they received good information or felt welcomed are more willing to use courts in the future than those who did not. There is a robust negative effect of the intervention on the beneficiaries who did not receive good information or did not feel welcomed: these are even less likely than the poorly informed/welcoming non-beneficiaries to be willing to return to court the next time. Considering the ranking of the courts in terms of fairness and trust in courts in place of the intended use of courts gives similar results ( Table O8 in the Online appendix).

Effectiveness and influence of legal aid
The results highlight three key points that help locate legal aid in the landscape of local-level dispute resolution and the broader context of Burundian society and statebuilding. The third point will be developed in more detail, with the support of qualitative data.
First, the absence of effects of the SAJI on the use of non-judicial dispute resolution mechanisms highlights the entrenchment of diverse and complex justice pathways in Burundi: only a minority of justice-seekers use only one forum to solve their dispute. Even after the introduction of a large-scale legal aid system, people tend to go to several dispute resolution fora with no resulting apparent erosion of the role of customary or non-judicial state actors. The inability of legal aid to curb the high number of actors may contribute to the subsistence of a situation where, as our data show, most disputes take a long time to be solved -and research has stressed the negative effects of long-lasting unresolved disputes on peace and stability (van Leeuwen & van der Haar, 2016).
Second, the increased access to state judicial institutions triggered by the SAJI does not necessarily indicate the full realization of the rights of justice-seekers. Our data shows that the beneficiaries of legal aid are not more likely to use courts in the eventuality of another issue. These results underscore the need to properly distinguish between access to justice and access to justice institutions. In our case, the SAJI increased access to justice institutions, but there is no substantial evidence that it improved access to justice. As a matter of fact, the beneficiaries of the legal aid program do not trust justice institutions more than non-beneficiaries. These results are concerning; they suggest that the SAJI legal aid may not properly empower citizens as it did not fundamentally change the relationship between them and state judicial institutions.
Third, and further characterizing our previous point, our results suggest that procedural justice, the treatment received by justice-seekers while in court, is key to their behavior vis-à-vis judicial institutions (especially their intended future use thereof). A dense literature has explored the idea of procedural justice and associates it with a better everyday compliance of people with the law and a higher acceptance of judicial decisions (Walters & Bolger, 2019). More data would be necessary for us to investigate these aspects fully, but our findings do suggest that the welcoming and level of information justiceseekers receive in courts -a reasonably good proxy of the global treatment received -has an incidence on their perception of and accessing justice. While the SAJI did not explicitly seek to change judges' and clerks' behavior (through trainings or incentives for instance), our qualitative data suggest, in some cases that we will now explore, a possible 'watchdog effect', which is, SAJI advisers acted as monitors of judicial actors, and thereby improved their behaviors.
After the SAJI was launched, in a context marred by bad practices and unlawful decisions or proceedings (RCN Justice & Démocratie, 2006), frictions rapidly emerged between SAJI advisers and judges. The SAJI advisers reported that judges initially appeared worried about the role the advisers would play, and municipal employees as well as judicial and administrative authorities also unanimously explained that they perceived the advisers as 'disturbing' the work of the judiciary. As a municipal employee put it, '[w]hen he [the SAJI adviser] arrived, judges wondered whether this graduate lawyer would watch them. Was he present to represent one party in lawsuits or to tell judges what to do?' In four of the eight municipalities of the study, we did find evidence that the SAJI advisers continued to be perceived as 'watching the judges' and had a positive influence. An adviser explained: When I was involved in a case, the various actors knew that they had to be careful: judges had to comply with the procedures. The municipal administration, the police or the wealthy, they knew that they couldn't interfere in the case.
Apart from the judges themselves, other actors involved in dispute resolution and provincial authorities -including presidents of provincial courts and appeal courts who supervised municipal judges as well as local civil servantsacknowledged this influence and explained that the presence of the advisers led to judges better complying with the law and procedures. In the words of a provincial court president: 'I have seen fewer unlawful decisions during the SAJI implementation'.
However, there are also clear limitations to this watchdog effect. The judges and clerks that we met typically stiffly denied that the SAJI advisers influenced them, which is an expected reaction from local notables who see themselves as embodying the law. While such an attitude does conceal the positive influence of the advisers in some cases, it also reflects an openly hostile attitude that does not benefit justice-seekers in other cases. As an adviser explained: 'I had many problems with one judge. I noticed that he systematically took decisions unfavorable to the party I represented. After that, I advised beneficiaries that I should not come when this judge chaired the trial.' The advisers, all university graduates, were seen as challenging the social status of judges who felt threatened -some reacted by behaving better, but others reacted with a less constructive form of hostility. The power dynamic between the SAJI adviser and judges is complex and often depends on local factors and individual characters, which future research must investigate in more detail. There are, however, good reasons to doubt the persistence of the advisers' influence in the long run. In fact, the SAJI stopped after a few years of existence. The 2015 political crisis, characterized by an increased authoritarianism of the administration, hindered the judiciary's independence, and would likely have severely altered the relationship between independent advisers and judges.

Concluding remarks
Our study has shown that a large-scale legal aid program did manage to reach the poorest fringes of society and increased the use of local courts. The program did not affect the situation of legal pluralism: justice-seekers continued to use other dispute resolution mechanisms even when accessing the courts more. Legal aid did not automatically increase trust in courts either. These findings suggest that, overall, legal aid does little to strengthen the relationship between citizens and state institutions (a relationship that is crucial to stability in post-conflict contexts). However, our research suggests that when legal aid is delivered in a context of better procedural justice, the perception of the courts improved slightly. Our qualitative results also suggest that legal aid services may, in some cases, act as a watchdog of courts, thereby improving legal actors' behavior (included, but not limited to, procedural justice).
There is a clear need for more research to probe the impacts of legal aid services further. Our research was limited in what it could realistically achieve: survey data like ours are often imprecise in capturing non-trivial aspects of the experience of justice-seekers such as the precise sequencing of the recourse to local institutions, corruption and other informal and illegal practices, and trust in institutions. We conclude on four points for future support of and research on access to justice, beyond the case of Burundi.
First, our study shows that despite the generally low level of trust in the judiciary, the provision of information to and support of disenfranchised justice-seekers do increase their ability to go to courts, and thereby their legal agency. This empirical result is consistent with qualitative studies conducted in DR Congo showing that even lowly functioning justice institutions maintain a certain legitimacy among the population (Rubbers & Gallez, 2012).
Second, the watchdog effect reveals the need for checks and balances at a local level to improve access to justice rather than merely access to justice institutions. However, it is essential to question whether the implementation of a local monitoring system is realistic in large rural and remote areas in fragile states characterized by very limited resources. More research is needed to fully comprehend the interplay between the different local dispute resolution actors. How do they relate to each other? How do local dynamics change in the presence of legal aid or other justice-seekers assistance in the medium and long run? The understanding of the complementarity or competition between actors involved in dispute resolution mechanisms seems, generally speaking, in its infancy.
Third, even if the SAJI experience led to outcomes that are probably positive, they are also limited and are dependent on significant external resources and funds. This pilot program was not sustainable. In Burundi, out of the 21 municipalities that had benefited from the SAJI experience, only one decided to keep the SAJI services (and a facilitator with a high school diploma replaced the law graduate, who was too expensive). Legal aid programs require the deployment of significant human and financial resources, which are typically not readily available in many low-income and fragile contexts. Given the limited effects of the SAJI program, one can question the cost-effectiveness of such intervention.
Fourth, our analysis suggests the importance of procedural justice, but more research is needed to measure the relative importance of the different dimensions that may matter in a court case, such as the quality of the decisionmaking process and the quality of the treatment of justice-seekers. Our findings about the role of procedural justice emphasizes the need for twin-track approaches that not only act to facilitate access to state justice institutions but also improve the quality of the experience in courts. For instance, a proper training of and dialog with judges may diminish the antagonism between judges and legal aid advisers and may, in turn, improve judicial proceedings. Then, a program like SAJI may enhance the demand for justice through state institutions not only in the short term, but also in the longer run. Such a dynamic is crucial for the stability of socalled fragile states as well as the legitimacy and credibility of legal institutions.
Ultimately, the challenge of legal aid in a context like Burundi was, and remains, to help establish fair and accessible judicial institutions as a credible reference point to improve the realization of rights in all fora. Better access to state judicial institutions is meaningless if access to justice does not ensue.