Courting Civilians During Conflict: Evidence from Taliban Judges in Afghanistan

Abstract

Rebels regularly provide public services, especially legal services, but the consequences of such programs are unclear. We argue that rebel courts can boost civilian support for insurgency and augment attack capacity by increasing the legitimacy of the rebellion, creating a vested interest in rebel rule, or enabling rebel coercion of the civilian population. We study the impact of the Taliban's judiciary by leveraging cross-district and over-time variation in exposure to Taliban courts using a trajectory-balancing design. We find that rebel courts reduced civilian support for the government and increased it for the Taliban, and were associated with more attacks and more coalition casualties. Exploring mechanisms, we find that courts resolved major interpersonal disputes between civilians but also facilitated more insurgent intimidation of civilians, and that changes in public opinion are unlikely to have been driven solely by social desirability bias. Our findings help explain the logic of rebel courts and highlight the complex interactions between warfare and institutional development in weak states.

Rebels regularly provide public goods to civilians during conflict, with a goal of boosting productive capacity or winning supporters.¹ Although scholars have documented many predictors of and explanations for rebel service provision,² the consequences of such programs remain largely unclear. Despite rich theoretical exploration of rebel services,³ micro-level evidence of their impacts and adjudication between causal mechanisms have been difficult to obtain, though such evidence remains important for understanding conflict more generally.⁴

We argue that rebel courts can sway civilians to support insurgents through several mechanisms: legitimization of the rebellion, facilitating coercive control of the population, or creating a vested interest in continued rebel presence. By promoting civilian collaboration with the insurgency, rebels can use courts to advance their position in the struggle over the flow of information: understanding where, when, and how the state operates in a given conflict. Winning the battle over information can enable armed opponents of the state to anticipate military operations, coordinate attacks with more precision, and engage in more varied and intense violence.

We examine a common type of rebel service provision, judicial services, in the context of the war in Afghanistan. The Taliban implemented a system of mobile courts, whereby judges were rotated within districts to resolve major interpersonal, familial, and domestic legal issues that state courts and traditional dispute resolution were unable to resolve due to cost, inaccessibility, or lack of enforcement capacity. The Taliban's mobile court system was a “spot service,” which is common among rebels who wish to provide services but lack territorial dominance.⁵

We use new data on the time and location of Taliban courts in Afghanistan in the early 2010s to evaluate the impact of judicial service provision. We exploit the time-series cross-sectional variation in district exposure to Taliban courts to estimate their effect on civilian attitudes and insurgent attacks. Our econometric design reweights districts never exposed to courts to interpolate the counterfactual average using a trajectory-balancing design.⁶

We find that Taliban courts impact public opinion in the short run. We estimate a 15 percent decline in willingness to use state courts to resolve disputes after Taliban courts are introduced, and a 23 percent increase in support for the Taliban returning to power. Several studies have documented that exposure to rebel governance matters for civilian attitudes after conflict ends.⁷ We expand on these findings by showing how courting civilians impacts public opinion during conflict.

Further, we show that changes in rebel services have effects on the battlefield. The Taliban significantly increases direct fire and improvised explosive device (IED) attacks after courts are introduced, which leads to more coalition causalities. Consistent with the idea that shifts in public opinion facilitate more insurgent attacks by encouraging civilian collaboration with the rebels, we show a positive correlation between use of government courts and reporting of IEDs to the coalition. And although we find that civilians’ willingness to collaborate through information sharing depends on exposure to insurgent courts, we do not find any changes in recruitment, suggesting that intelligence sharing is perhaps the key pathway connecting collaboration to conflict after courts are introduced.

Our paper makes a number of contributions, the first of which is empirical. As Ginsburg notes, “The consequences of the use of law and courts are still rather obscure. Existing evidence does not suggest that rebels that use law are more likely to prevail against the state but does suggest that rebel legal institutions can compete rather favorably compared with state institutions.”⁸

We provide evidence that rebel law and courts have a significant impact on civilian attitudes which translates into material changes on the battlefield. The limited empirical literature on the consequences of rebel governance activities suggests that rebel governance has ambiguous effects on rebel combat strength. Rebel services can signal high organizational capacity, increasing the odds of a peace deal,⁹ but governance may have a null or negative relationship with rebel strength.¹⁰ Ginsburg finds no correlation between the existence of rebel law and courts and ultimate rebel success.¹¹ Our study provides micro-level evidence which leverages within-conflict variation, which can account for cross-country heterogeneity otherwise not accounted for in prior studies. In doing so, we provide some of the first concrete evidence that rebel justice shapes battlefield conditions.

Our second contribution is clarifying and testing causal mechanisms to explain our baseline findings. We elucidate three causal processes by which insurgent courts can secure civilian collaboration: legitimacy, coercion, and vested interests. Legitimacy emphasizes civilian “hearts and minds” being bought by rebels, whereas coercion is a case where civilians remain staunchly opposed to rebel rule but collaborate due to the threat of force.¹² We also highlight the under-discussed mechanism of vested interest. This differs from the legitimacy and coercion accounts in that civilians may not have their hearts and minds bought by rebels, and may also not be entirely forced into supporting them, but choose to side with them due to pragmatic concerns that arise from the type and quality of the service being provided.

We provide evidence consistent with all three mechanisms, suggesting they may work jointly, feeding into one another. First, we show that rebel courts led to an increase in insurgent intimidation of civilians, which is consistent with increased targeting of government collaborators enabled by a legitimate process to try and prosecute opponents. With a fair judicial process, insurgents can credibly promise not to punish civilians who did not work with the state, alleviating a potential trade-off between deterrence and backlash when rebels attempt to rule through fear alone. Thus courts facilitate more repression of political opponents by insurgents.

Next, we show that courts are not only enabling coercion of civilians: they also reduce major disputes between civilians. The evidence is consistent with vested interest. If the Taliban were attempting to resolve security issues in the community where courts operated, one would detect a decline in large-scale communal conflicts, which is what we observe. The dispute results suggest that courts are not just rubber-stamping coercion, which would have happened in the absence of judges: they are also meaningfully impacting civilian behavior.

Further, we connect our results on civilian attitudes to combat by exploring how civilian collaboration may facilitate insurgent attacks. We explore two possibilities: civilians change their informing behavior in response to courts, and civilians join the insurgency after courts. We find that use of government courts is positively associated with informing on insurgents at the micro level, which is consistent with the idea that insurgents can trade services for information.¹³ We find no evidence of changes in recruitment activity.

More broadly, our study joins many others that explore how war creates states and governing institutions.¹⁴ Scholars have long recognized that the pressure to win wars spurs governance innovations, specifically fiscal instruments to secure tax revenue to finance war efforts.¹⁵ We show that competitive state building during civil wars can also incentivize a race to provide judicial services to the civilian population. We build on the “war made the state” literature by focusing on how insurgent groups develop legal institutions to capture popular support in the context of ongoing conflict.

Although the Afghanistan case has unique characteristics and Taliban courts are not generalizable to all rebel governance institutions, the core strategic dynamics this study elucidates are highly relevant in a wide range of civil conflict contexts. In places as diverse as the Philippines, Colombia, and Northern Ireland, rebels have employed courts and dispute-resolution techniques in competition with government authorities. Worldwide data indicate that since World War II there have been more than 200 internal armed conflicts that have included informal justice processes.¹⁶ Our results suggest they may have shifted the political and battlefield conditions in favor of rebels courting civilians during conflict.

Theory

Courts are a key foundation of rebel governance, since they “allow the group to penetrate a community very effectively in relation to both important and mundane aspects of civilian life.”¹⁷ Figure 1 shows the distribution of countries that have had a civil war from 1945 to 2012 (coded missing if not) and highlights those where during the course of the conflict at least one rebel group offered legal institutions. Slightly over half of the civil wars during this period had rebel groups who provided law or court services.¹⁸ At the rebel level, Huang finds that 28 percent of rebel groups in her data employed some form of court.¹⁹

FIGURE 1. Rebel judicial service provision during civil wars, 1945–2012

Judicial services are institutions defined by three core functions: dispute resolution, social control, and lawmaking.²⁰ Arbitration and mediation services are implied by courts, but a judiciary does something stronger: it connects rulings with coercive power, which allows courts to modify behavior through social control and lawmaking rather than through creating self-enforcing agreements alone.²¹ Judicial institutions support the rule of law—the concept that all are accountable to the same set of restrictions.

Courts rely on coercion, but their purpose is not to provide social order solely through violence. If insurgents wished to compel civilians to behave in a particular way, they could secure compliance without appealing to a formal legal process to justify their method of control. Investing in a court system is costly during an insurgency, since rulings need to be enforced and judges need to be protected, paid, and trained. The shifting of resources to noncombat activities implies that rebels find value in securing civilian approval not just through coercion but also through legitimacy.²²

Courts and Territorial Control

Rebel courts may emerge when insurgents control an area, as in the emergence of the Tamil Tigers’ court system in Sri Lanka, but this does not require dominance over land area. Indeed, “while territorial control can certainly facilitate rebel governance, features of contemporary civil conflicts suggest there is much beyond territorial control that enables rebel groups to govern, foster social relations with civilians, and appeal for external support.”²³ Insurgent justice may be a spot service: a good provided by rebels based on local demand, without a fixed infrastructure. The Maoist insurgents in Nepal are a key example. They used a mobile court system that rotated across villages to adjudicate cases. Although Maoists did not have a monopoly on violence or dominance over the territory, they were able to exert influence and provide a service that helped them gain popularity. The Taliban relied on a similar mobile court system. Qualitatively, scholars have echoed this point in the context of the Taliban, declaring that “the Taliban do not have to take territory to control it.”²⁴ Thus, even without territorial control, rebels can leverage their influence and presence to shape civilian behavior, building their strength and enabling their future control over territory.

Here, we focus on mechanisms by which courts, rather than territorial control, can sway civilian attitudes and the tactical choices of insurgents. Empirically, we provide evidence that our mechanisms are distinct from pre-existing control by showing the robustness of our findings after conditioning on three measures of pre-treatment territorial control.

How Courts Secure Civilian Collaboration

Courts can influence civilian collaboration with the insurgency in several ways. We stress that these are not mutually exclusive. The existence of one pathway does not imply that another mechanism is crowded out in its entirety.

Legitimacy

First, courts may persuade civilians that rebels are legitimate forces. “By giving the community what it lacks, the group gives locals a reason to form positive beliefs about its involvement in local affairs.”²⁵ Further, courts can chip away at government legitimacy by demonstrating that plausible alternatives to the state exist.²⁶ To the extent that moral outrage can drive civilian collaboration with insurgents,²⁷ the presence of effective rebel courts can highlight the government's failings while providing an outlet for civilians to act on their negative assessments of the rule of law provided by the state. In much the same way as the state's expansion into lawless areas can secure civilian support by facilitating social bonds,²⁸ rebel judicial expansion may increase the legitimacy of the insurgency.

An example of courts enhancing rebel legitimacy is the civil war in Nepal. The Communist Party of Nepal-Maoist (CPN-M) gained significant support for the insurgency with their courts.²⁹ These “people's courts” provided a less corrupt judicial system, which was highly popular among the rural poor.³⁰ Loyle suggests that the “Maoists themselves” attribute part of their success to the competence that their judicial system demonstrated, which focused on property crimes, domestic abuse, and war crimes.³¹

Social control

Second, courts may enable insurgent coercion of the population by creating a process that legitimizes violence against opponents. If rebels rule through fear alone, civilians may rationally choose to inform to the government when they would not want to otherwise: if a civilian can expect to be victimized with the same probability should they collaborate with the state or not under coercive rebel rule, they are better off working with the state to try to change their situation. But with a criminal process and a court, a civilian may feel more secure in choosing not to collaborate with the state, since they can prove their innocence to the insurgents should they be accused. Thus, with courts, insurgents can punish informers without fear of a backlash, leading to more net coercion.

An example of judicial institutions facilitating social control and coercion is the Irish Republican Army (IRA) in Ireland. Dudai documents how the IRA used a complex system, of which courts were a key component, to manage and coerce government informers to keep the insurgency alive.³² The constant fear that one could be accused of being an informer led to defections, since members thought they had a better chance of staying safe by working for government victory. To create the perception that one could contest a false accusation of informing, the IRA relied on a court-martial procedure with due process to allow adjudication of informing claims, rather than executions on the spot, which helped maintain cohesion.³³ Similarly, the Islamic State has declared that violence is permitted only by the law, but has also made “spying for nonbelievers” punishable by death under their penal code.³⁴ The Islamic State established a court system but backed its governance with repression, which ruled civilians through fear.³⁵

Vested interest

We elucidate a third mechanism, discussed often in the Afghanistan context: vested interest. In this account, courts create a positive externality of social peace. Major disputes may remain unsettled by government courts for a variety of reasons—lack of access to government buildings, administrative delays, or corruption may all block civilian access to justice. Local forms of dispute resolution, such as arbitration and mediation by elders or elites, may not be effective at resolving distributional conflicts between civilians over productive assets because preserving a judgment requires some type of external enforcement. Whereas mediation and arbitration are efficacious when self-enforcing solutions can be found, judicial intervention with third-party enforcement is necessary to resolve a conflict without self-enforcing solutions.

Rebel courts can settle conflicts which were otherwise intractable. This is especially the case for mobile courts: if justice can be delivered on the spot, rebels can make a good that was difficult for civilians to obtain due to distance or transportation costs suddenly accessible.

After settling conflicts between civilians that were otherwise intractable, communities exposed to insurgent courts enjoy new levels of order. The order created by rebel courts is intrinsic to continued rebel presence. Should insurgents lose their grip on a community, the cases they settled could re-emerge because the legal and enforcement apparatus necessary for stability would dissipate.

The dynamic of vested interests has been used to explain judicial development historically. During a period of civil conflict in medieval England, the king extended judicial services to politically vulnerable areas to secure support. Legal protection locked in support: “there was no guarantee that rights extended by one ruler would be maintained by another. This gave subjects strong incentives to support a king under whom they enjoyed new protections.”³⁶ Since citizens knew the legal services provided by the crown might change under a new ruler, they increased support for the government relative to challengers.

The Taliban's court system illustrates the vested-interest mechanism. Vested interests in rebel rule created a cycle of dependence between civilians and continued rebel presence, which extended beyond the claimants of a particular case. Most obviously, winners of cases knew their preferred outcome was conditional on Taliban enforcement: “One could expect retaliation should the government reestablish its presence in any area that had been under Taliban influence for some time because the losers in disputes and criminal cases could turn to the government for support.”³⁷

Since verdicts had community-wide impacts which created social peace, rebel courts could foster dependence on continued insurgent presence among civilians who were not claimants in a particular case. Since only the Taliban could enforce difficult cases, “with each verdict the Taliban courts deliver, the insurgency's presence increasingly becomes the condition for sustaining the social peace it has produced.”³⁸ For instance, a village in Ghaziabad district was subject to constant insecurity caused by an interfamily quarrel over a forest. The Taliban's judges resolved the case, threatening arson should either family violate the judgment. Even villagers who had little sympathy for the Taliban approved of the ruling, and benefited from the end of the dispute.³⁹

Another example of vested interest can be found in the rural Andar district of mixed-ethnicity Ghazni Province, where there was a long-running dispute between two families over the ownership of a tract of land devoted to grape growing. All agreed that a neighboring tract had been sold in the 1950s by one family to a second, but the second family claimed that the sale also included the vineyard, which the seller denied. The case caused significant tension in the community, with the buyer reportedly viewed as stirring up trouble. After being approached by one of the claimants, the local Taliban judge issued a signed slip of paper requiring all parties to come and produce their paperwork pertaining to ownership of the vineyard. After two days, judgment was rendered for the seller: the vineyard had not been sold, and no further claims by the buyer family would be entertained.⁴⁰

How Courts Increase Insurgent Attacks

Civilian collaboration augments insurgent attack capacity. Rebels rely on obedience, spontaneous support, or enlistment by civilians for resources and labor for the insurgency.⁴¹ We highlight two pathways by which civilian collaboration could assist insurgents after courts: information sharing and recruitment.

Information sharing

Information is critical to civil-conflict dynamics: “counterinsurgents seek it, insurgents safeguard it, and civilians often trade it.”⁴² Civilian informing to the government will spoil insurgent attacks, enable government ambushes, and constrain rebel operations due to concerns about plans leaking to counterinsurgents. For instance, optimal bomb placement is along highly traveled paths; a group could help avoid civilian causalities if they inform civilians where they are planing on placing mines, but this risks the same information leaking to enemy forces.⁴³ If insurgents have little political capital with the community, they may not be able to pursue their best military strategy due to the possibility of civilian collaboration with government forces. On the other hand, with information from civilians and their support, insurgents can operate in secrecy to attack the government without fear of tips spoiling their attacks.

Recruitment

A second form of civilian cooperation with insurgents is enlistment.⁴⁴ Civilians can join rebel groups, providing labor for the insurgency and augmenting the capacity of the group to carry out attacks. But joining up costs much more than sharing information, and we expect this mechanism to be inactive.

Summary of Causal Pathways

Figure 2 outlines the causal pathways mapping courts to our outcomes of interest as articulated in the theory section. Courts can boost public support through legitimacy, by increasing social control through coercion, or by creating a vested interest by reducing disputes. The upshot of increased public support is collaboration from civilians, which leads to more attacks.

FIGURE 2. Mechanisms connecting courts to outcomes

Context: Afghanistan, 2008–2013

We study post-2001 Afghanistan to empirically test the effectiveness and mechanisms of rebel judicial services—in this case courts operated by the Afghan Taliban. In contrast with past work, we focus intensively on a single country and rebel group, leveraging within-country variation, rather than looking for generalities across many rebel groups. This has both inferential and theoretical advantages, allowing us to hold fixed many of the potentially confounding factors across countries to isolate the specific effect of courts on citizen attitudes and conflict outcomes.

The War in Afghanistan

After a devastating civil war (1992–1996) that followed the Soviet invasion of Afghanistan, the Taliban took control of the country in 1996, inaugurating a five-year regime that was notable for its uncompromising policies and close relationships with international Islamist organizations like al-Qaida. Following the terrorist attacks of 11 September 2001, the United States and NATO allies invaded Afghanistan and expelled the Taliban from Kabul, installing an internationally backed government.

The Taliban took several years to regroup, but then began mounting a large-scale insurgency against the new Kabul government.⁴⁵ In addition to violent operations against foreign troops and government security forces, the Taliban made “law and order” a core part of their appeal to the population. Dating back to their administration of the 1990s, the Taliban had branded themselves as a movement that would provide social stability by aligning Afghan society with Islamic law, with some traditional Pashtun social rules included for good measure.

Taliban Courts

In the mid-2000s, as the Taliban began to re-establish itself as a political contender, with support from Pakistan, its court system began to take shape. Providing courts was a mechanism for establishing and consolidating authority during a period of significant international military presence. Largely recruited from Deobandi madrassas in Afghanistan and Pakistan, the incoming Taliban judges were already trained in the prevailing Islamic legal theories that the movement wished to enforce nationwide.⁴⁶ Taliban courts existed in earlier days of the insurgency, but became well organized by 2011.⁴⁷ The institutionalization of the court system included non-local judges and rotations across Afghanistan, to reduce the risk of corruption and co-optation.⁴⁸

The case of the Taliban's judiciary is of particular interest because of the evolution of the court system as it pertains to rebel territorial control. The Taliban's judiciary was a core component of their governance framework, and it emerged prior to, rather than because of, territorial control. As Jackson writes:

The critical point, and one which is often missed in analyses of Taliban control, is that governance does not come after the capture of territory, but precedes it. Coercion, coupled with the more popular aspects of Taliban governance such as justice, softens the ground. Taliban governance does not supplant the Afghan government but co-opts and augments it, resulting in a hybrid service delivery arrangement.⁴⁹

Taliban courts are an example of “on the spot” services rebels deliver when they lack a monopoly on territorial control. The Taliban would leave contact information in villages and drive judges into town on demand to provide civilians with legal services should they have a dispute to settle. The Taliban's mobile courts are similar to the courts set up by the CPN-M in Nepal, which also dispensed justice from town to town.⁵⁰

During the period we study, the Taliban used “governance to keep the population at least marginally satisfied, and this, in combination with their coercive power, helps secure the population in areas under their influence or control.”⁵¹ A part of the appeal of the Taliban's court system is its reliance on an interpretation of Sharia that allows the Taliban to use popular religious appeals to justify their rulings while also incorporating local custom into decision making. Thus these courts can base their decisions on criteria that local townspeople find agreeable and more familiar in terms of procedure than the Western-based rules and norms used by the Afghan government.

The rise of Taliban justice after 2001 corresponded with a growing realization that the legal system of the Western-backed Kabul government was struggling to resolve a long backlog of civil and property disputes.⁵² Some dated back to the Soviet war era of the 1980s, with legal uncertainty and conflict interfering with final disposition.⁵³ By combining legal and religious training, as well as a careful understanding of the local cultural context, Taliban judges were able to render locally legitimate judgments.⁵⁴ Perhaps most importantly, they were able to do so quickly.

Taliban courts, like other judiciaries, mainly resolved civil cases, such as land disputes, property conflicts, or divorces.⁵⁵ This was strategic: solving these problems provided social order for the whole village, which helped create a vested interest. “Taliban judges might even suddenly turn up in a village asking about a particular dispute or a criminal case, presumably following a report by the Taliban's own intelligence system. This suggests that the Taliban did not merely conceive the judiciary as a service being provided, but as a strategy to penetrate rural communities.”⁵⁶

Judgments by Taliban courts were more effective than state courts or mediation for at least three reasons. First, they had religious legitimacy. One disputant reported, “Unfortunately, I lost the trial. I am not upset at the Taliban judges, they judge according to the Sharia, and I cannot oppose the Sharia.”⁵⁷ Although dispute cases produce nominal “losers,” the religious legitimacy undergirding the courts can shield the Taliban from backlash. One loser of a land dispute said, “The Taliban took my land from me, but to be honest I didn't understand how Shariah worked … Now, logically looking at it, when they told me I needed to build a proper building, it makes sense to me.”⁵⁸

Second, Taliban courts had greater enforcement capacity. Traditional mediation did not have a coercive backing that could compel a loser to abide by a decision; indeed, in the forest dispute in Ghaziabad district described earlier, both families had taken the dispute to arbitration only for the loser to violate the ruling without punishment.⁵⁹

Third, Taliban courts were more accessible than government courts. Sharia gave them religious legitimacy but also represented legal principles that were broadly familiar to the population. Relative to procedural rules largely transplanted by Western legal institutions for government courts, the Taliban system was far simpler to navigate. Government courts were also hard for many rural villagers to access, with court fees, corruption, and the cost of transportation creating high barriers to entry. In contrast, Taliban judges often arrived in response to a single phone call. In a US congressional hearing in 2020, John Sopko, special inspector general for Afghanistan reconstruction, said, “As much as you hate the Taliban, and I do, and I hate their brand of justice, to the average Afghan it is better than the justice provided by the National Unity Government.” He went on to detail how three separate Afghans he worked with had advised their families to use Taliban instead of government courts to resolve their disputes.⁶⁰

Data and Design

Courts

We use data on court locations and the years they were operational. The data were collected by a team of field researchers from two sources: Taliban officials and district elders and tribal leaders.⁶¹ Information from both sources was used for cross validation. The team collected baseline data in 2011 and updated information in 2012 and 2013. We use these field reports and harmonize the information available with district boundary information provided by the Empirical Studies of Conflict project. These boundaries are similarly used to merge the attitudinal outcomes and conflict event data we describe later. We map the spatial distribution of courts in Figure 3.

FIGURE 3. Taliban courts over time and space

Although most courts continued indefinitely after being established, a few (WD in the map legend) were withdrawn after some period. These withdrawals were largely unrelated to local conditions (such as the types of cases, public opinion, or conflict) but rather driven by disagreements over turf within and between the Quetta and Peshawar branches of the Taliban.⁶² Our primary analysis focuses on comparing the districts exposed to courts once to those never exposed. For robustness, we show our results remain unchanged when considering districts that received courts once as treated for the duration of the panel (details on Appendix page A50). This design choice is minor, in part because the share of withdrawn court districts is relatively small.

The presence of Taliban judicial services in 2011 does not necessarily mean that a court was established in 2011. Unfortunately, we lack precise information on start dates in districts that had courts in 2011; some of these courts may have been first deployed in 2009 or 2010. In practice, though, during these early years (the heart of the Obama-era US surge) the Taliban judiciary was small and was not highly active, only becoming mature by 2011.⁶³ From an empirical perspective, measuring start dates later rather than earlier should bias our design toward zero, since balancing in the pre-period will seek balance between the counterfactual outcome Y (0) plus the treatment effect, creating a weighted comparison group that subsumes the treatment effect already. We provide a mathematical illustration of this point in the Appendix on page A6.

Civilian attitudes

We measure civilian attitudes with survey data from NATO, which contracted ACSOR, an Afghan subsidiary of the international firm D3, to design and field a recurring household-level survey. The data we rely on are drawn from the Afghanistan Nationwide Quarterly Assessment Research (ANQAR) survey. ACSOR hired and trained local enumerators in household and respondent selection, data recording, culturally sensitive interview methods, and secure storage of contact information. ACSOR's use of local-to-area enumerators increases comfort with survey interviews and decreases anxiety that external actors are monitoring and tracking respondents. We use Waves 1 through 24 of these quarterly surveys, which were collected between November/December 2008 and May 2014.

We construct four outcomes with our survey data which we aggregate to the district-year level, the level of granularity available for our court data.

State court usage. ANQAR asks respondents whether they would take a case to a government court if they hypothetically had a dispute. We use this question to measure whether civilians disengage with state institutions after being offered an alternative service by insurgents.

Taliban approval. We use a question asked from Wave 6 to Wave 24 on whether a return of the Taliban as a governing body would be good for the country.

Government influence. ANQAR asks “Between the two, the anti-government elements [mukhalafeen-e dawlat] and the government, who has more influence in your area [mantaqa] now?” We code a 1 if a respondent says the government, and 0 otherwise.

Support for government index (gov't index). We follow Plumb and colleagues and build a index of support for the government using principal component analysis.⁶⁴ The questions are highly correlated and ask respondents to assess how well different layers of government (district, provincial, national) perform on a variety of dimensions (economy, security, corruption, development and reconstruction, and overall). Together, these questions represent civilian approval of the government's performance, which we expect to decline when the Taliban offers a compelling alternative to the government's services (such as courts).

Combat

We measure insurgent attacks, which we theorize will be a function of court presence due to changes in support from civilians. Our combat data are drawn from two sources.

First, we use event logs from the Afghan NGO Safety Office (ANSO) from 2008 to 2013.⁶⁵ During this period ANSO produced weekly spreadsheets of security-related incidents, recording for each event the timing, location, participants, and a description. The reports were submitted by a nationwide team of more than 100 enumerators working for ANSO, with nationwide coverage. Importantly, during this period the Taliban did not discourage or target NGOs or aid agencies; on the contrary, they encouraged aid agencies to operate, in an effort to skim resources and claim credit.

We use the text in event descriptions to code a range of event types, including disputes between civilians, property conflicts, IED events and armed opposition group (AOG) events (excluding IEDs to avoid double-counting), and crime (such as robbery, non-AOG homicide, burglary, or theft). AOG events usually refer to Taliban forces. Disputes occur when non-AOGs engage in a violent clash over a disagreement. An example of a dispute from the data is, “An altercation occurred between two local civilians over a tribal dispute, injuring one person.”

Second, we use declassified data collected by members of the International Security Assistance Force and their Afghan counterparts. During the conflict, these security forces documented the time and location of attacks, as well as attack type; there are more than 100,000 incidents between 2008 and 2013. We focus on two types of attacks: direct fire (attacks perpetrated at close range—direct line-of-sight encounters) and IEDs (usually roadside bombs).

Our data also track casualties among Afghan and foreign forces. We use these measures to capture the intensity of insurgent missions that successfully harm or kill security forces. We also use information about instances of nonlethal attempted coercion or intimidation of the civilian population.⁶⁶

Empirical Strategy

Our data contain N = N _tr + N _co districts i ∈ {1, 2, …, N}, where tr and co denote treatment and control, across time periods t ∈ {2008, 2009, …, T}. Districts received Taliban courts (which we simply denote as “courts”) in a staggered fashion, placing them in cohorts i ∈{2011, 2012, 2013, ∞}, where ∞ denotes that the district did not receive a court during the sample period. Further, let t ₀ be the time where t = cohort_i, meaning t ₀ is the first time period when district i is treated, and t ₋₁ is the time period before treatment. Define a treatment indicator court_it ∈ {0, 1}.

$${\rm cour}{\rm t}_{it}: = \left\{{\matrix{1& {{\rm if\;cohor}{\rm t}_i < \infty \ {\rm and} \ t \ge t_0} \cr 0& {\rm otherwise}\hfill}}\right.$$

We are interested in estimating the the average treatment effect on the treated (ATT) for the outcomes of interest Y :

$${\rm AT}{\rm T}_t = E[ Y_{it}( 1 ) -Y_{it}( 0 ) \vert {\rm cour}{\rm t}_{it} = 1] $$

where the potential outcomes are Y(court_it), Y _it(1) denotes the outcome under treatment and Y _it(0) what the outcome would have looked like under control, and court_it is the binary indicator for whether a court is present.

Our outcome of interest under control can be defined in four ways: the court group before courts, the court group after courts, the control group before courts, and the control group after courts:

$$Y( 0 ) = \left({\matrix{ {E[ Y( 0 ) \vert {\rm cohort} < \infty , \;t \ge t_0] , \;} & {E[ Y( 0 ) \vert {\rm cohort} < \infty , \;t < t_0] ) } \cr {} & {} \cr {E[ Y( 0 ) \vert {\rm cohort} = \infty , \;t \ge t_0] , \;} & {E[ Y( 0 ) \vert {\rm cohort} = \infty , \;t < t_0] } \cr } } \right), \;$$

However, we do not observe what the court group (those districts that received courts during the sample period) would have looked like had they never gotten courts. Replacing the foregoing matrix with values we observe, the counterfactual world is missing.

$$Y( 0 ) ^{{\rm observed}} = \left({\matrix{ {{\rm unobserved}, \;} & {E[ Y\vert {\rm cohort} < \infty , \;t < t_0] ) } \cr {} & {} \cr {E[ Y\vert {\rm cohort} = \infty , \;t \ge t_0] , \;} & {E[ Y\vert {\rm cohort} = \infty , \;t < t_0] } \cr } } \right), \;$$

Our identification approach is to model the relationship between E[Y _it(0)|cohort = ∞, t < t ₀] and E[Y _it(0)|cohort = ∞, t ≥ t ₀] (the second row in the Y(0)^observed matrix) to interpolate E[Y(0)|cohort < ∞, t ≥ t ₀].

Popular approaches for modeling the counterfactual of treated units had they not been treated include difference-in-differences (DiD) and the synthetic control method (SCM), neither of which is appropriate for our setting. SCM requires a small donor pool with a long pre-treatment history to arrive at an unbiased estimate of treatment effects. Our data have a short history prior to courts. DiD requires parallel trends, which will be violated due to selective service provision based on trends in combat activity and civilian support.

We outline and explain our choice of design in Figure 4. Since DiD and SCM are both inappropriate for our data, we use trajectory balancing to estimate the effect of courts on our outcomes of interest.⁶⁷ Trajectory balancing is a general reweighting approach for causal inference with panel data and binary treatment regimes where some units are exposed to an event in an absorbing fashion and other units are never exposed. This is an extension of Hazlett's method.⁶⁸ The intuition for identification is this. If units that are exposed have a similar pre-trend in the outcome of interest as control units, then the control units serve as a valid counterfactual for what the evolution of the outcome would have looked like in the absence of the event. Following this idea, one may select a set of weights to create a weighted control group such that the trend of the actual treated units is mean-equal to the weighted control group. The weights from that control group may then be used to project out what the counterfactual evolution of the outcome would have been among the treated units. We estimate the ATT as

(1)$$\matrix{ {\matrix{ {{\widehat{{ATT}}}_t = \displaystyle{1 \over {N_{{\rm cour}{\rm t}_i < \infty }}}\mathop \sum \limits_{{\rm cour}{\rm t}_i < \infty } y_{it}-\mathop \sum \limits_{{\rm cour}{\rm t}_i = \infty } w_iy_{it}\ {\rm with\;weights\;s}{\rm .t} .} \cr {\displaystyle{1 \over {N_{{\rm cour}{\rm t}_i < \infty }}}\mathop \sum \limits_{{\rm cour}{\rm t}_i < \infty } y_{it} = \mathop \sum \limits_{{\rm cour}{\rm t}_i = \infty } w_iy_{it}\ {\rm for}\ t\le {\rm year\;before\;treatment}} \cr {{\rm subject\;to}\mathop \sum \limits_i w_i = 1\ {\rm and}\ w_i > 0, \;\forall _i} \cr } } & {} \cr } $$

FIGURE 4. Decision tree illustrating choice of research design

Our approach is subject to three assumptions which we contextualize to our setting, test observable implications of, and relax when appropriate.

Assumption 1 Conditional ignorability: $Y_{it}( 0) \bot {\rm cohor}{\rm t}_i\vert {Y}_{i, {\rm pre}}\;\;\forall t > T_0$

That is, the potential outcome for treated units had they not been treated is independent of their treatment status, conditional on the trajectory of the outcome. The assumption is reasonable since the data generating process for outcome variables implies places that have similar values of an outcome are likely similar on fundamental baseline characteristics. For instance, if two places have the same level of support for the Taliban or same level of IED attacks, it would be very unusual if those districts were vastly different in terms of their history as a Taliban stronghold, assuming that past historical presence is a strong predictor of the outcome(s).

Assumption 2 Linearity in prior outcomes (LPO): $E[ Y_{it}( 0 ) \vert {Y}_{i, {\rm pre}}] = ( {1, \;Y_{i, {\rm pre}}} ) ^\top \theta _t + \eta _{it}$ for E[η _it|Y_pre] = 0

That is, the potential outcome for the treated under control (conditional on prior outcomes) can be expressed as a linear function of past outcomes with a common intercept shift. LPO is reasonable given the aggregation level of our data. A violation would occur if our outcomes followed strong seasonal patterns: for instance, if we measured our outcomes at the monthly level, the cycles of the fighting season could create nonlinearity in the potential outcome for the treated unit under control. However, since we aggregate our data to the yearly level, seasonal cycles would need to vary annually, and there is little evidence of this in the Afghan context.

We assess a violation of these assumptions as follows. Say Y _it(0) depends on a transitory shock η _it and that $E[ \eta _{it}\vert {\rm cohor}{\rm t}_i, \;{Y}_{i, {\rm pre}}] \ne 0$. Then our estimator will have the following bias.⁶⁹

$$\underbrace{{ATT}}_{{{\rm treatment\;effect}}} + \underbrace{{E\left[{\displaystyle{1 \over {N_{{\rm cour}{\rm t}_i < \infty }}}\mathop \sum \limits_{{\rm cour}{\rm t}_i < \infty } \eta_{it}-\mathop \sum \limits_{{\rm cour}{\rm t}_i = \infty } w_i\eta_{it}} \right]}}_{{{\rm bias}}}$$

Transitory shocks η _it may include things like economic fluctuations (commodity price shocks, weather fluctuations) or changes in conditions on the ground (surges of troops). As an example of this bias, say a subset of control districts received an adverse economic shock which temporarily increased violence and reduced civilian support for the government. These districts may receive a high weight from our algorithm because they had more violence leading up to the treatment period. However, since transitory shocks are temporary, these control districts may have simply reverted back to their mean levels of support and conflict, making a normal cycle look like an increase among the treated units.

We assess a violation by filtering our data to the pre-court period and checking for differences between court and (weighted) control districts. If future realizations of Y are independent of cohort_i conditional on Y _pre, then the difference between treated and control districts should be substantively small and statistically insignificant in the sample where no treatment effects are present. This should also be the case with time-varying covariates that courts ought to not impact as well. Intuitively, since the treatment effect is known to be zero in the pre-period, any differences in time-varying covariates are the result of bias rather than the impact of courts, which would suggest our estimates in the post-period are contaminated.

We check for differences between the court group and control group by looking for differences in important time-varying variables: nighttime lights, opium, wheat price, rain, and temperature shocks, the number of forward operating bases in a district, and population.

Figure 5 plots the estimated ATT and p-value for our main outcomes of interest (attitudes and combat) along with auxiliary covariates (night lights, opium shocks, wheat shocks, rain shocks, temperature shocks, population, the count of US and NATO bases) using 2010 and 2011 as placebo treatment time periods for (1) all cohorts and (2) the 2012/2013 cohort with Equation (1). We are unable to estimate a placebo year with 2009 since there is insufficient pre-treatment data to capture a trajectory (only one period).

FIGURE 5. ATT and p-value of placebo courts on outcome(s)

Results are substantively small (less than 0.2 standard deviations) and statistically indistinguishable from zero. The exceptions are opium shocks for the 2010 placebo year and rain shocks for 2011. Since we conducted thirty tests, finding two estimates that are statistically significant at the .05 level is approximately consistent with the null hypothesis. Because we do not find differences between the court and noncourt groups in cases where we should not see treatment effects, we have provided evidence against the possibility that lurking differences between court and noncourt districts explain the change in the outcomes, supporting Assumptions 1 and 2.

Assumption 3 Weight feasibility: there exists a set of weights w _i that are non-negative and sum to 1, such that

$$\displaystyle{1 \over {N_{{\rm cohor}{\rm t}_i\ne \infty }}}\mathop \sum \limits_{{\rm cohor}{\rm t}_i\ne \infty } y_{it} = \mathop \sum \limits_{{\rm cour}{\rm t}_i = \infty } w_iy_{it}\ {\rm for}\ t \le {\rm year\;before\;treatment}$$

We provide support for this assumption by plotting the data for the treated cohorts versus the weighted average of the control cohorts. The plots illustrate that a set of weights that satisfy the constraints exists.

To conduct inference, we use a jackknife procedure which sequentially drops districts and re-estimates Equation (1) to estimate variance and apply normal theory to obtain p-values.⁷⁰ We explore other methods to obtain standard errors for robustness (Appendix C.9).

In Table 1, we summarize the key assumptions, whether they have directly observable implications we can test, whether the assumptions can be relaxed, and how we relax them for robustness.

TABLE 1. Summary of assumptions and tests

Results

Baseline Results

We first show the impact of courts on civilian attitudes and combat, respectively. Then we explore causal mechanisms using other outcome data. We report our results graphically first, illustrating the trend of the court group versus the counterfactual as a function of years until courts are introduced. We then present the average difference between the groups, in tables. The results are summarized in Figures 6 and 7.

FIGURE 6. Public opinion trends: trajectory balanced

FIGURE 7. Armed conflict trends: trajectory balanced

Courts shift civilians toward rebels

Exposure to rebel courts shifts civilians’ attitudes toward rebels. Column (1) of Table 2 shows a 7 percent (nearly 0.4 σ) decline in the number of respondents reporting that they would take a dispute to a state court. This provides evidence that Taliban courts crowd out government service provision. And column (2) shows that this reduction in usage of state courts translates into an increase in support for the Taliban: after courts, 5 percent more respondents say that a return of the Taliban would be good for the country.

TABLE 2. Civilian attitude results

Notes: Outcomes are attitudes, measured with either ANQAR. Jackknife standard errors in parentheses. * p < .05; ** p < .01; *** p < .001.

Courts also shift the government's influence. Column (3) of Table 2 shows a 7 percent decrease in respondents reporting that the government has the most influence in their village. Finally, civilians score government performance lower—a 0.37σ decline. Together, the evidence suggests that after courts are introduced civilians disengage from state institutions, change their preferences regarding Taliban governance, rate government performance lower, and ultimately consider the government less influential in their district.

One may expect subgroup effects along salient demographic traits if courts cater to particular interest groups. But we find little systematic evidence of differences between Pashtun and non-Pashtun Afghans or between economic strata. We find some weak but mixed evidence that men reduce their use of state courts more than women do, but we find no difference between genders on support for the Taliban returning (Appendix page A39).

Courts facilitate insurgent attacks

Rebels are able to convert increased civilian support into attacks. Columns (1) and (2) of Table 3 use ANSO-based outcomes. First, we observe eleven additional events involving AOGs (the Taliban) and nearly eight additional events involving IEDs. These estimates are sizable relative to the mean of the control group (69 percent and 67 percent, respectively).

TABLE 3. Combat results

Notes: Outcomes are combat, measured with either ANSO or SIGACTS. Jackknife standard errors in parentheses. * p < .05; ** p < .01; *** p < .001.

Columns (3) to (5) show combat increases using SIGACTS outcomes. Insurgents execute thirty-three more direct fire attacks and nearly eight more IED explosions. The estimates of IED events from ANSO and SIGACTS are nearly the same; these data sets, which log events using different methodologies, are nevertheless consistent with one another. Finally, the rebel attacks are not immaterial or without collateral consequences: these events increase casualties among coalition and Afghan forces.

Robustness of Baseline Results

We conduct the following supplemental analyses to check the robustness of our results.

Relaxing conditional ignorability

Our first identification assumption is that Y _it(0) is independent of court_it conditional on ${Y}_{i, {\rm pre}}.$ Our placebo test, introduced in Figure 5, increases our confidence in this assumption, but it may fail in practice if some other covariate is prognostic of both exposure and the potential outcome of our covariates of interest.

We relax the assumption by conditioning on district features which may predict courts and the outcome: rain and opium shocks, as those covariates were unbalanced in our placebo test; and Taliban control of the district (Appendix page A25). Further, we use three measures of pre-court control (survey responses on government influence in a district, survey team access to districts, and the presence of forward operating bases) to capture the possibility that our results are driven by pre-existing Taliban influence in a district. Our results are consistent across measurements and specifications (Appendix page A69).

Relaxing LPO

Our second key assumption is that the potential outcome under no treatment is linear in the pre-treatment outcome history. Unlike either DiD or SCM, our approach can relax LPO. We do this by seeking balance on higher-order dimensions of the pre-outcome history, rather than just the mean, through kernel balancing. For intuition, a mean balance for a flat trend could be achieved by collapsing very volatile control units with high and low values, but the counterfactual that is projected will likely be unrealistic, since several units with different variances are unlikely to serve as a good counterfactual for a steady trend. Since kernel balancing accounts for volatility when creating weights, it avoids this pitfall. Further, by seeking balance on higher-order features, kernel balancing performs better when there is a short pre-treatment history.⁷¹ We replicate our results with kernel balancing and covariates (Appendix page A32).

Inference

We obtained standard errors from the jackknife in our main results, as described in the Methods section. Inference is complicated in our setting because the weights we obtain are also subject to uncertainty. Here we calculate the variance of our estimates in a different way, with a block bootstrap treating districts as blocks, and obtain similar t-statistics (Appendix page A64).

Alternative combat models

We measure combat as the count of events in our main models. We check that our results are consistent when we use the natural log of combat events per capita (Appendix page A16). ANSO and ANQAR begin coverage in only 2008, but SIGACTS goes further back. We extend the panel with SIGACTS outcomes from 2005 to 2014, documenting a similar pattern as found in the shorter panel (Appendix page A20).

Mechanisms

In this section, we provide evidence consistent with vested interest and social control. Further, we show that public support is connected to attacks, drawing on survey questions about use of court services and willingness to inform on insurgents.

Do Courts Resolve Major Disputes?

The vested-interest mechanism we propose suggests that major interpersonal disputes should decline in response to the introduction of courts. This decline likely occurs through two channels: it is likely that courts reduce disputes directly and also indirectly deter them because the existence of an effective legal architecture makes illegitimate claims to ownership less likely to succeed. If the Taliban's judiciary prevented major social conflicts in a locality, creating stability and thus locking civilians into support for continued insurgent presence, especially large and disruptive conflicts would decline in treated areas.

We test this mechanism using ANSO event data. We parse text for events involving communal conflict or violence between neighbors but not involving AOGs. We further restrict these data to disputes involving land. Since courts create vested interests when they solve disputes without self-enforcing solutions, the type of violent disputes measured in our data should be responsive to court introduction.

Further, as a falsification test, we include crime as an outcome: since the Taliban's courts largely focused on resolving disputes that disrupted day-to-day civilian life and not on more trivial forms of criminal activity, we expect to see no difference in observed criminal violence. Our results could suffer from bias if reporting of events decreased after courts, showing a decline in crime when the only thing that actually changed was the flow of information.

Disputes decline significantly after courts are introduced (Table 4). This can largely be attributed to changes in disputes over land specifically, which connects closely to the qualitative literature on the Taliban's judicial services. We observe no difference in crime. One might be concerned that we observe a decline in disputes only because Taliban courts cut off access to reporting. But then one would see a decline in reporting across the board, including crime. Yet we see no change in criminal activity, increasing our confidence that the observed change in disputes is driven by behavior on the ground rather than the data-reporting process.

TABLE 4. Trajectory-balanced dispute resolution and crime results

Notes: Trajectory balancing results for the effect of courts on dispute resolution and crime. Odd columns are logged counts (+1), and even columns are binary 1 (event > 0). “Taliban courts” estimates are derived from weighted difference in means, where weights are obtained from Equation (1). Standard errors nonparametrically computed through jackknife. ^†p < .10; *p < .05; **p < .01; ***p < .001.

Do courts increase coercion of civilians?

After courts are introduced, the social-control mechanism suggests that the Taliban will be able to target opposition forces more diligently and precisely. Indeed, the Taliban's court system gathered information about crimes from villagers who reported facts to insurgents for the trial.⁷² The network of spies for court cases had an additional value: the Taliban could rely on these individuals to report on people collaborating with the government, which the insurgency long considered a crime.⁷³ Thus the judiciary served to institutionalize the insurgents’ coercive apparatus.

We use SIGACTS data on insurgent intimidation events, which logs threatened or realized violence toward civilians—events where “an individual or group of individuals are murdered by insurgent action due to their association with a particular group or organisation,” of which “killing of informers” is a key example.

Taliban intimidation incidents increase in frequency once courts are established (Table 5). We note an increase of 15 percent in the probability of intimidation among treated units relative to the counterfactual. Our estimates suggest around 0.5 more intimidation events, which is 68 percent of the average number of intimidation events among control units.

TABLE 5. Insurgent intimidation results

Notes: The table reports the effect of Taliban courts on insurgent intimidation, or coercion, of civilians, measured as the count, log, or binary incidence of threats or use of lethal force against civilians by the Taliban. Estimates obtained via Equation (1). Standard errors are jackknifed. *p < .05; **p < .01; ***p < .001.

How can coercion covary with persuasion?

An increase in intimidation of civilians going hand in hand with an increase in support for the insurgency raises a question: why would civilians approve of greater repressive force? The core reason insurgent intimidation of civilians can increase while civilians also grow more approving of the insurgency is that a segment of the civilian population prefers order to disorder, and may view coercion as a necessary intervention to achieve stability. The combination of increased services with more coercion aligns with the “varying combinations of persuasion and coercion” insurgents use to consolidate control.⁷⁴ For example, brutality against alleged criminals received public approval in Colombia, where insurgents used violence on the pretext of establishing local authority and social order.⁷⁵

Further, the finding is consistent with the literature on state building, courts, and civil war. Judicial institutions inherently combine coercion and persuasion; the threat or use of force by the provider of a judicial service is an intrinsic and implied aspect of every court ruling.⁷⁶ Further, the shift from interpersonal coercion (in the form of violence over disputes) toward increased Taliban violence against civilians is strongly suggestive that force is being monopolized by the Taliban after the introduction of courts. Theoretical models of the transition from anarchy to state consolidation predict a decline in violence between civilians because of the increased capacity of the provider of security to use force.⁷⁷

We interpret the increase in coercion in tandem with public support as evidence of the courts’ dual role, and the activation of two parallel mechanisms. A rival account would be that coercion is the only active mechanism and that coercion leads to preference falsification among survey respondents who feel social pressure to express support for the Taliban. We address social desirability bias in two steps.

First, if respondents became reluctant to express their true feelings after the introduction of courts, we would probably observe respondents not only lying about their true feelings but also refusing to answer questions. We use this insight to construct a new variable: the rate of missingness of our key attitudinal questions by district-year. We estimate Equation (1) using missingness as the outcome(s) and find no evidence of divergence in response rates by group (Appendix page A43).

Next, we use an ANQAR question on respondent comfort during the survey to filter to respondents who are unlikely to feel social pressure. Respondents who express very high levels of comfort during the survey process are unlikely to be responding out of coercion or fear. We estimate Equation (1) using only comfortable respondents, and find the same results (Appendix page A43).

Do Attacks Increase Because of Civilian Collaboration?

We study two different ways that civilian collaboration may facilitate insurgent attacks after courts are introduced: information sharing and enlistment.

Information sharing

ANQAR data allow us to test whether court preferences correlate with tipping preferences at the individual level. Waves 20 to 24 of ANQAR include a question about providing tips on insurgent attacks to government/coalition forces: “If you knew that an IED had been planted, how likely would you be to report it to the local security forces?” (with possible responses very likely, somewhat likely, somewhat unlikely, and very unlikely).

The likelihood citizens provide information about IEDs in particular is salient to our context. Unlike other insurgent attacks, effective IED placement relies to some degree on civilian support of rebel goals. Unfortunately, this survey question was not asked prior to the establishment of courts, so we cannot test our argument using trajectory balancing or other pre/post forms of analysis.

Since we cannot capture the change in tipping preferences, we use the ANQAR question about where a respondent would take a dispute if they had one—to a government court, a local counci [shura or jirga], or elsewhere—to capture individual preferences regarding state institutions and their likelihood of reporting to the government. We regress willingness to report IEDs on the measure of use of state courts, adjusting for a host of geographic, temporal, and individual covariates:

(2)$$\eqalign{y_{i( d) ( w) } = \alpha + \delta _d + \omega _w + \gamma 1( {\rm {\scriptstyle USE}\;{\scriptstyle GOVERNMENT}\;{\scriptstyle COURT}}) _{i, ( d) ( w) } + \sum\limits_{k = 1}^K {X_i^k } + \eta _{i( d) ( w) }}$$

where i ∈ {1, 2 … 62, 199} indexes respondents, d ∈ {1, … 344} indexes districts, and w ∈ {20, … 24} indexes survey waves. The outcome of interest is civilians’ willingness to report an IED to local security forces. The outcome is measured on a scale from 1 to 4, with 4 representing the highest likelihood of informing. We also measure a binary measure 1(ied report ≥ 3) which is 1 when a respondent indicates a positive likelihood of reporting and 0 otherwise. Our regressor of interest is 1(use government court), which is a binary variable scored 1 if a respondent indicates that they would take their dispute to a government court.

We include k individual covariates—education, age, age², ethnicity, gender, perception of government control in the area, and whether the respondent would use a shura/jirga for a dispute—to capture observable traits that are correlated with support for the Taliban, which may codetermine willingness to report IEDs and the use of government judicial services. Further, we adjust for district (δ _d) and survey wave (ω _w) fixed effects. We cluster errors at the district, since this is the administrative level where government and Taliban courts were provided.

Consistent with our expectation, the use of the government's public services, especially for conflict resolution, is strongly correlated with willingness to collaborate with government forces in the neutralization of insurgent threats (Table 6). More broadly, this positive correlation likely reflects a dynamic relevant to court provision in settings of contested authority: armed groups can use conflict resolution and other public services to thwart government access to vital information by undermining ties between the civilian population and government institutions.

TABLE 6. IED reporting on court usage

Notes: Robust errors, clustered at the district, in parentheses. All models include age, age², education, ethnicity, gender, and perception of government control of the area. Regressor of interest is a dummy for whether individuals would take their dispute to a government court. Outcome is civilian likelihood of reporting an IED, measured either as binary (1 if positive) or in categories (high, medium, low, or very low). Question asked in Waves 20–24, which covered 344 districts. ***p < .001.

Although these descriptive patterns are robust to a range of model specifications addressing the most pressing concerns about nonrandom selection into use of government courts, we cannot fully rule out other sources of bias. We therefore caution against interpreting these results in the same manner as our trajectory-balanced estimates. However, these findings are consistent with a host of qualitative evidence from the Afghanistan context and beyond linking information sharing, civilian attitudes, and battlefield outcomes to provision of public services during conflict.

Recruitment

Another possibility is that the courts, by increasing sympathy, drive recruitment locally. We use SIGACTS data on terrorist recruitment. Unlike combat, which is easy to record given its violent and public nature, recruitment events are rarely observed and do not produce causalities. We measure levels and a binary specification. We do not find a consistent pattern in terrorist recruitment: although the ATT for the count outcome is positive and significant at the p < .10 level (estimate 0.104, SE 0.0585), we find no difference for the binary measurement, and the count increase is very small. We plot the counterfactual versus the observed data in Figure 8, which shows little evidence of changes in recruitment trends after courts. Thus it appears there is stronger evidence of civilian collaboration changing through information sharing rather than joining the insurgency.

FIGURE 8. Public opinion trends: trajectory balanced

Conclusion

Do rebel courts affect civilian attitudes and battlefield performance? We study the case of the Taliban courts in Afghanistan, one of the most intense counterinsurgency wars in the last twenty years. We leverage the timing and location of Taliban court expansion after the surge to explore the impact of rebel judicial institutions. Since the assumptions underlying common identification strategies will fail in this context, we adopt a novel design to reweight control units to construct the counterfactual.

We show that rebels can court public opinion by providing judicial services. After Taliban courts arrive in their district, civilians are less likely to turn to government courts and more approving of a Taliban return to power; they also consider government influence in their district to be weaker and government performance to be less adequate. Further, we document that the change in public opinion has material battlefield impacts: after courts are introduced, rebels increase their attacks and the coalition suffers more casualties, probably due to changes in civilian collaboration with the coalition rather than through recruitment.

Our study provides evidence that rebel judicial services can sway civilian attitudes during a conflict, with consequences for the war effort. The theoretical literature on rebel service provision suggests that the impact of governing institutions should be strong and in the direction we expect, but identifying the impact of courts has remained an empirical challenge. We supply evidence that judicial services are not mere window dressing: they meaningfully impact the course of the war. Our study is one of the few to show that rebel public goods have spillover effects on combat.

We also show the causal processes by which this occurs: courts change patterns of interpersonal disputes and enable increased rebel discipline and punishment of civilians. As a consequence, civilian collaboration changes primarily through willingness to share information. We corroborate this channel by showing that use of government courts correlates with willingness to inform on insurgents, and that patterns of recruitment do not budge after courts are introduced.

Our evidence on the theoretical mechanism should increase confidence that the findings we have for Afghanistan generalize to other conflict settings. Naturally, each civil war has unique features, setting it apart from other historical struggles. Yet many intrastate conflicts emerge under political pressures relevant to the scope conditions of our arguments about vested interests and social control. These include a weak legal system being provided by the government, a demand for legal certainty from the civilian population, and an insurgent group that has an ideology which can be used to legitimize their rules. Weakened state capacity, especially in the provision of conflict resolution, creates opportunities for armed opposition groups to mobilize civilian support. Insurgents can amplify grievances created by unstable judicial institutions, including fractured systems of land tenure, corrupted public administration (where the likelihood of legal consequences is low), and varieties of illegal yet unregulated economic discrimination. These dynamics are reflected in mobilization tactics used by the LTTE in Sri Lanka, the CPN-M in Nepal, and the IRA in Ireland and may well have had similar impacts on public support for these armed groups.

Rebel courts are a particular type of service, and our findings may not generalize to all types of public goods. For instance, a rebel-constructed road is unlikely to create a vested interest, because if the rebels leave, the government may be able to seamlessly transition maintenance to ensure that the good is provided uninterrupted. But this would not be the case for something like social order produced through judicial rulings, since the state cannot credibly commit to honoring all holdings by another court. Nor would roads serve the same information-gathering or coercive functions as a court.

Yet vested interests may not be a mechanism unique to judicial institutions. Another example is indigenous, linguistically inclusive education systems. In cases where governments have engaged in historical discrimination with respect to languages spoken or subjects taught, rebels may be able to provide a credible alternative that generates a vested interest among affected communities. By devolving control over curricula to local authorities or taking a more active role in promoting identity-based education systems, rebels may enable an approach to education that the government cannot credibly commit to maintaining. Like judicial systems, education is a platform for disciplining the local population, shaping knowledge formation and social behaviors through the subjects and languages taught.⁷⁸ Importantly, any deep understanding of how these systems shape civilians’ attitudes and behaviors will require future exploration.

The timing of our study also creates opportunities for future research in Afghanistan. This study focuses on how Taliban-led courts influenced civilian attitudes and combat outcomes in a relatively short period of several years in the middle of the twenty-year conflict in post–September 2001 Afghanistan. Future research may explore how these early episodes of rebel justice provision shaped the broader fight to re/establish the political legitimacy of the Taliban. For example, how did these courts influence Taliban attempts to consolidate authority after most international forces left in 2014? Did these courts give the Taliban a refined view of local power dynamics, enabling them to negotiate with political elders and elites in the run-up to the final military withdrawal in 2021? How did these early judicial proceedings shape public expectations and demands after the Taliban took over local, provincial, and national seats of power? Did these courts buffer local populations against the various economic and political shocks caused by the Taliban's takeover of Kabul and the country's various institutions?

More broadly, our work contributes to debates concerning conflict during the process of state formation. The need to levy taxes during conflict has long been credited as an explanation for the formation of governing institutions. Beyond this literature, civil war has been called a state-building process. We take a close look at how these institutions form and their impacts on civilian attitudes and actionable outcomes during an ongoing conflict. Future work might explore how public services provided by armed opponents of the state—on their own or as part of a group of governance institutions—shape short- and long-run interactions with the state, whether or not rebels seize the seats of power.