Coercion,governance,and political behavior in civil war

Introduction

In 1980, Peru democratized after a 12-year spell of military dictatorship. The May 1980 municipal elections were dominated by the centrist Popular Action party, which also swept the presidency and both houses of Congress. But an unusual pattern materialized in the election returns. In 16 of Peru’s 150 provinces – mostly provinces scattered across the rural Andean highlands – over 40% of votes were not cast in favor of any party. Rather, these ballots were either left blank or spoiled, deliberately or accidentally marked so as to render the ballot invalid.

A number of factors could lead voters to cast a blank or spoiled ballot. Peru has had compulsory voting since 1933, so voters forced to turn out may have left ballots blank out of apathy; others may have unintentionally spoiled their ballots. But the quantity of blank-and-spoiled ballots was surprising. In Peru’s most recent municipal elections, in 1966, only one province had recorded more than 40% blank-and-spoiled votes. By the 1983 elections, 28 provinces had crossed that threshold.

What else could explain this pattern? Latin American democracies have a long tradition of protest balloting: voters casting intentionally blank or null votes as a rejection of the political options on offer or the electoral system itself (Driscoll & Nelson, 2014). In the case of Peru, this explosion in blank-and-null ballots coincided with the rise of the maoist insurgent group Shining Path. Shining Path, which sought the overthrow of the democratic state, pressured voters to either boycott elections or to cast a blank or spoiled ballot in protest. Large amounts of blank-and-null voting would represent a symbolically powerful repudiation of democracy; sufficiently high levels would cause elections to be annulled. Many of the provinces where blank-and-spoiled ballots were most prevalent were also areas in which Shining Path was present.

Existing accounts of how armed groups intervene in elections suggest a straightforward connection between Shining Path’s presence and high levels of protest voting. ¹ Scholarship on armed actors and elections centers on electoral violence: how armed groups use threats and acts of violence to coerce voters to turn out, stay home, or vote for the group’s preferred candidates. The framework seems a natural fit to the case of Shining Path, which is widely conceived of as a radically violent organization with little concern for civilian welfare (Weinstein, 2006).

Indeed, the insurgents were brutally violent against civilians throughout the war. Shining Path combatants committed widespread abuses, including the killing or disappearance of tens of thousands of largely poor and indigenous rural residents. Yet the historical record belies the hypothesis that Shining Path coerced voters into casting protest ballots through violence: the eight local and national elections held during the war saw no incline in insurgent victimization of civilians.

How, then, do we understand the sharp increase in protest voting in Shining Path-influenced areas? How do armed actors affect electoral outcomes? I argue that a narrow focus on electoral violence misses a key channel through which armed actors like Shining Path influence elections. Beyond directly threatening voters to abstain, protest vote, or vote for their preferred candidates, armed groups like Shining Path can wield coercive power in indirect ways to sway political behavior. The ability to use violence converts armed actors into potential political authorities. Armed groups build a wide range of formal and informal governance institutions, from courts and policing to government councils, schools, and health and humanitarian services (Huang, 2016). These institutions forge ties of political authority between armed actors and civilians, incorporating local populations into armed groups’ political projects and increasing the credibility of their messaging.

I argue that armed actors can draw on these relationships of political authority to influence the political behavior of voters. For armed groups like Shining Path, which developed complex institutions of social organization and wealth redistribution, the popular support, political mobilization, and social control enabled by governance offer a means to sway voters’ political behavior without resorting to election violence. Both groups that seek to discredit democracy, like Shining Path, and those that aim to channel votes to particular candidates, like some militias and paramilitary groups, can employ this strategy to appeal to voters. This approach holds important strategic advantages over the direct use of coercion, which risks both alienating the civilians that militants rely on for political and material support and triggering state repression.

To assess this argument, I draw on a combination of qualitative and quantitative evidence about Shining Path governance, micro-level data on voting and violence from 1980 to 1993, and a synthetic control case study of governance and voter behavior in a province in the Peruvian highlands. I find that Shining Path created a diverse set of institutions for governance, influence, and social control in communities it influenced, and that these efforts to leverage relations of political authority with civilians – not acts of violence against voters – explained the high levels of protest balloting observed during the war.

This article is organized in five parts. In the following section, I provide a brief overview of the literature on electoral violence and armed groups in democracy, before outlining my theory of governance, authority building, and political behavior. The fourth section evaluates the argument through an analysis of Shining Path’s governance activities and its attempts to discredit democracy and sway voters. The final section considers the implications of this theory for scholarship on armed group governance, electoral violence, and political participation.

Armed groups and elections

A host of armed actors have a deep interest in elections and their outcomes. The reasons for this interest are many. Democratic regimes derive legitimacy from the ability to conduct free and fair elections. For insurgents aiming to overthrow the state, the success or failure of elections affects public and international support for the government’s political project. The outcomes of elections also matter: for secessionist groups, elections and referenda may lead directly to increased autonomy or independence. Paramilitary organizations and profit-oriented criminal groups have their own electoral stakes. Election outcomes shape their ability to wage state-sanctioned violence, avoid crackdowns on illicit economies, gain immunity from prosecution, and further their broader political agendas. A wide range of armed actors thus face strong incentives to influence democratic outcomes.

Given these incentives, how do armed actors influence elections? A growing research agenda in political science centers on electoral violence – the use of violence against voters, political opponents, and state officials in the context of an election in order to sway electoral outcomes (Höglund, 2009). ² Though this literature largely focuses on the use of violence by political parties (e.g. Birch, Daxecker & Höglund, 2020), recent work has begun to explore the use of violence against voters by non-state armed actors as well.

Matanock & Staniland (2017) situate electoral violence as an element of a broader set of strategies by which armed groups participate in elections, including running their own candidates and covertly or overtly supporting allied parties. They argue that relatively weak armed groups use violence to sway election outcomes, whereas strong groups prefer direct democratic participation. Other research emphasizes the connection between groups’ goals and their use of electoral violence. Armed groups with ties to mainstream political actors often coerce voters into supporting those actors at the ballot box. Acemoglu, Robinson & Santos (2013) and García-Sánchez (2016) show that Colombian paramilitary groups connected to elite interests channel votes to conservative politicians, while Turnbull (2021) demonstrates that armed militias in Nigeria are deputized by political parties to turn out voters.

Insurgent groups may leverage electoral violence to different ends. Insurgents aim to discredit and replace the state, and often lack ties to mainstream political actors. As a result, they may use coercion to reduce turnout and impede elections. Condra et al. (2018) show that the Afghan Taliban used violence against voters and election officials to depress turnout, arguing that elections present a unique opportunity for insurgents to deal both instrumental and symbolic damage to state institutions. Gallego (2018) demonstrates that Colombian rebels surged acts of violence in election years, reducing turnout in national elections.

That electoral violence is a key tool with which armed groups influence democratic outcomes is unsurprising. Armed actors are, after all, defined by their ability to wield violence and coercion; an ample empirical record demonstrates that they sometimes use this power to punish and coerce voters and politicians in the context of elections. However, a narrow focus on electoral violence offers only a partial picture of how armed actors influence democratic outcomes. Like states, armed groups can leverage their coercive, economic, and social endowments in sophisticated ways to achieve their political aims. By using their coercive endowments to position themselves as credible political authorities, I argue, these groups can sway voter behavior without resorting to direct electoral intimidation.

Coercion, governance, and political behavior

The ability to wield violence opens avenues of influence beyond pure coercion. In addition to enabling electoral violence, coercive capacity transforms those who wield it into potential political authorities. Coercion renders these actors capable of generating social order, providing public goods, making and enforcing laws, and pre-empting challenges by other would-be rulers. For groups that choose to govern, I argue that these relations of political authority provide a potent channel through which to sway political behavior and election outcomes. Rather than directly coercing voters, armed groups can draw on these institutions of service provision, political mobilization, and social control to influence voter behavior.

Relations of political authority underpin state–society relations in the modern world. To be sure, states rely on direct coercion of citizens and political rivals – police, prisons, and security forces – to enforce social order. More commonly, however, they induce citizen compliance through less direct means. States often supply public goods strategically to enforce popular loyalty. In what Albertus, Fenner, and Slater (2018) term ‘coercive distribution,’ authoritarian regimes use public goods provision to induce loyalty by ‘enmeshing [citizens] in relationships of material dependence […] and establishing the central state as the indispensable purveyor of life’s necessities’ (2018: 9). The threat to withdraw critical public services can be as potent a tool as the threat of force.

States also turn to soft power and symbolic politics to cement their political authority. States manufacture legitimacy and enforce obedience through symbolic politics, mobilizing citizens into public rituals that reinforce the authority and benevolence of the state (Wedeen, 2015). They build communications infrastructure to inundate citizens with propaganda about the legitimacy and power of the state and to promote loyalty by appealing to national or ethnic identity (Warren, 2014; Stapleton, Uribe & Wright, 2022). Though ultimately made possible by the ability to use force, these strategies, like the coercive distribution of public goods, generate popular compliance without resorting to the point of a gun.

Non-state actors often follow the same playbook. A broad set of armed actors attempt to construct relations of political authority with local communities. Rebel governance during civil war – defined by Huang (2016: 51) as a strategy ‘in which rebels use political organization to forge and manage relations with civilians’ – is common. Like state–society relations, political interactions between rebels and civilians can take many forms. Insurgents provide a variety of public goods to local populations, ranging from courts and dispute resolution services to welfare, education, and infrastructure development. Many rebel groups also impose formal or informal rules on public life, setting and enforcing regulations that mandate or forbid certain social, political, or economic behaviors. Finally, some insurgents create institutions to incorporate local civilians into their political projects. These groups invite civilians to participate in town councils and local government bodies or to otherwise provide feedback to rebel leaders (Wickham-Crowley, 2015; Arjona, Kasfir & Mampilly, 2015; Arjona, 2016; Huang, 2016). Paramilitary groups and militias govern as well. Biberman and Turnbull (2018) document extensive social service provision by militias in Pakistan and Nigeria, encompassing education, healthcare, dispute resolution, and disaster relief.

Why engage in governance? A large body of literature on insurgency (e.g. Mao, 1954 [1938]; Galula, 1964; Kalyvas, 2006; Berman, Shapiro & Felter, 2011) has shown that rebels gain important military benefits from creating systems of regularized, positive interactions with civilians. Rebels govern not out of altruism or irrationality, but because governing furthers their political and military goals. Effective, responsive governance creates legitimacy and mobilizes civilians in favor of a group’s aims. Governance institutions also act as valuable monitoring mechanisms, leveraging rebel administrators and civilian participants as intelligence collectors to identify potential collaborators and defectors. Governance incentivizes civilians to provide material support, information about enemy movements, and a stream of recruits to sustain the military effort.

I argue that the benefits of constructing political authority through governance extend to the realm of elections and political behavior. Governance serves as a powerful amplifier of armed group messaging. Armed actors who build governance institutions have the opportunity to demonstrate their commitment to the political goals they claim to espouse, increasing the credibility of those claims. If governance is effective and fulfills local demand, offering public goods can generate support for the agendas of armed groups and a willingness to act according to their wishes at the ballot box. Similarly, institutions that directly or indirectly incorporate civilians into armed group political structures can increase popular buy-in to the group’s political agenda. Like states, armed groups can use popular, effective governance institutions to foster genuine support among voters, inducing them to vote or abstain when the armed group asks them to.

Governance may also generate a less genuine form of support. As Wedeen (2015) argues, empirically disentangling genuine support for political authority from coerced compliance is challenging. Armed actors often force civilians to consume their political propaganda by embedding it in school curricula or mandatory indoctrination sessions. Like states, armed groups often have the power to control local information dissemination. By manipulating the ‘marketplace of ideas,’ armed groups can manufacture popular buy-in to their political agendas by convincing civilians that an armed group is a viable, perhaps inevitable ruler, one worth cooperating with rather than resisting. Armed groups often reinforce this impression by adopting the trappings of statehood, appropriating the symbols and performing the behaviors of the state to legitimize their claims to authority and demands for compliance (Mampilly, 2015).

Governance can therefore foster both genuine and coerced loyalty to the political programs of armed groups. When an armed group appeals to voters to support its allies at the ballot box, abstain, or cast a protest ballot, voters who have been exposed to or participated in the group’s governance institutions are more likely to comply. To be sure, I do not expect this relationship to hold always and everywhere: unpopular or abusive forms of governance, for instance, may repel voters and cause them to reject the armed group’s demands. But in many cases, I contend, governance represents an important channel through which armed actors can sway political behavior.

To varying degrees, governance activities by non-state actors are enabled by their ability to wield coercion. Building propaganda infrastructure to project soft power necessitates little coercion, but enforcing social order, redistributing wealth, and resolving disputes requires the ability to punish transgressors; evicting existing political authorities to carve out space for governance is intensely coercive. Yet the threat of coercion that underpins these governance activities has little in common with electoral violence. Unlike electoral violence, it makes no specific demands of voters and entails no direct threat to punish them their actions at the ballot box. Coercion lurks in the background of governance and only indirectly affects political behavior.

A strategy of influencing political behavior through governance holds two key advantages over the use of direct electoral coercion. First, it decreases the risk of alienating voters. While direct coercion may be effective, victimization of civilians in war sometimes causes them to defect and support other actors (Condra & Shapiro, 2012; Lyall, Blair & Imai, 2013) or to otherwise resist militant rule (Kaplan, 2017). This is especially problematic for armed actors if the civilians they victimize belong to a social or economic constituency they see as key to their political project; either intentional or collateral harm inflicted on civilians can reduce material and political support (Turnbull, 2021). Condra et al. (2018) show that this concern shapes how insurgents use electoral violence against civilians, forcing rebel leaders to balance the political value of attacking elections against the political and military cost of alienating civilians whose support they need.

Second, electoral coercion may backfire in the short term. A large literature on responses to victimization shows that people often respond to violence with increased participation in politics. Victims are more likely to turn out in elections, protest, and engage with their communities (Bateson, 2012; Bauer et al., 2016). Balcells & Torrats-Espinosa (2018) suggest that this effect is immediate: in the direct aftermath of terrorist attacks, people are more likely to vote. In addition to alienating civilian constituencies that militant groups rely on, by triggering defiant mobilization electoral coercion may fail in its immediate goal of shaping who people vote for or whether they vote at all.

In the following section, I study how Shining Path used governance to influence voting behavior. I focus on two forms of governance emphasized by the insurgents: wealth redistribution and political indoctrination. I draw on archival and secondary evidence to examine how Shining Path commanders created and leveraged governance institutions, violent event data to trace their use of electoral coercion, and election results to show effects on voter behavior. To further isolate the effects of these interventions, I conduct a synthetic control study of protest voting in a province in central Peru, Daniel A. Carrión, in which Shining Path constructed ties of political authority with voters but did not engage in electoral coercion. If my argument is correct, I expect to find that Shining Path’s governance activities – replacing state institutions with its own mechanisms for wealth redistribution and propaganda – increased the proportion of voters who heeded the insurgents’ demands to cast blank-and-null ballots.

Evidence from Peru

The Peruvian civil war ran from 1980 to 2000 and claimed roughly 70,000 lives. The conflict represented a pivotal event in contemporary Peruvian history: in addition to the immense human toll of the violence, public backlash to the war led to the unexpected election of Alberto Fujimori in 1990 and the subsequent collapse of Peruvian democracy, which had only emerged from dictatorship a decade earlier. The legacy of the conflict remains a deeply salient issue in electoral politics and popular culture (Burt, 2021).

The war was primarily fought between Shining Path and the Peruvian state, though smaller insurgent and paramilitary organizations also participated. Although initially treated as an annoyance by President Fernando Belaúnde (1980–85), Shining Path’s alarming military success led to increasing state crackdowns starting in 1983. By 1985, the government had declared an emergency zone in the southern highlands, where Shining Path was most active, and deployed over 2,500 troops to the region (McClintock, 2001).

The conflict escalated through the mid-1980s, with Shining Path holding territory in rural areas up and down the Andes and carrying out major attacks in Lima and other major cities. Both insurgents and counterinsurgents committed large-scale human rights abuses as fighting intensified. After Shining Path’s leader, a philosophy professor and uncompromising ideologue named Abimael Guzmán, was captured by security forces in a surprise raid in Lima in 1992, the group began to disintegrate and violence decreased sharply. Fragments of the group have survived to the present but have little popular support or military capacity.

National analysis, 1980–93

Did governance by Shining Path affect voting behavior? Estimating this effect poses two related empirical challenges. First, systematically measuring governance by armed groups is difficult. Civil war datasets generally focus on violent events – clashes between combatants and violence against civilians – and are rarely able to capture the contextual details that speak to the motives or goals surrounding each act of violence. This presents an obstacle for measuring governance, which may not be violent; when it does entail violence, governance may be difficult to distinguish from other logics of victimization. Second, the degree to which Shining Path governed was positively correlated with the group’s coercive control, since consolidating territorial control created opportunities for complex forms of governance. This presents an inferential problem: given that the same conditions that enable militants to influence voters through governance also enable electoral coercion, how do we disentangle the two?

I adopt two empirical approaches to address these challenges. First, I use testimonies from the Truth and Reconciliation Commission to identify acts of governance by Shining Path. The CVR conducted a large-scale national fact-finding mission, collecting nearly 17,000 testimonies from people affected by the conflict (Heilman, 2018). Though data missingness concerns remain, these testimonies offer a wealth of data on the insurgents’ violent and nonviolent activities across the country and throughout the war. They contain text fields with descriptions of each violent event, the circumstances surrounding it, and the victims’ understanding of the motivation for the act of violence. I digitize these testimonies from the CVR report and use keyword searches to identify acts of redistributive governance or mandatory propaganda sessions. ⁴

I also code acts of electoral violence by keyword search ⁵ and add two control variables from CVR data: counts of civilian victimization by state security forces and civilian harm resulting from state-insurgent clashes in order to capture variation in conflict dynamics across time and space. ⁶ I take the natural log of these variables to account for leftward skew in the counts of these events. Finally, I generate an indicator for Shining Path presence that takes the value of 1 if the insurgents committed one or more acts of violence in a given district-election cycle. The summary statistics for these variables are reported in Online appendix 2.

To evaluate the effect of governance on voter behavior, I scrape district-level election returns for the 1983, 1986, 1989, and 1993 local elections from the National Jury of Elections’ Observatory for Governability web portal. To help separate governance from electoral coercion, I follow McClintock (1984) in estimating the effect of Shining Path governance on blank-and-spoiled ballots rather than on turnout. Turnout is highly sensitive to coercion. Armed actors usually target electoral violence at polling stations, making it risky to vote at all. They are also able to observe who turns out and who does not. In contrast, protest voting is unobservable to armed actors. Assuming ballots are kept secret, a reasonable assumption in the case of 1980s Peru, armed groups have no way of knowing whether someone who entered the voting booth voted blank/null or cast a valid vote. This inferential tool is not perfect – armed groups may, for instance, punish communities collectively if they are unhappy with aggregate election results – but helps ameliorate concerns of omitted variable bias. Treating blank-and-spoiled ballots as votes in support of Shining Path is common in the secondary literature on the Peruvian internal conflict (McClintock, 1984; Palmer, 1986; McClintock, 2001); the Peruvian National Office for Electoral Processes notes that blank-and-spoiled ballots were interpreted as ‘votes in support of PCP-SL [Shining Path]’ (ONPE, 2005). ⁷

I build a national panel of districts spanning the four local elections listed earlier. I first regress protest voteshare on the indicator for Shining Path presence, adding district and election cycle fixed effects to account for

OPEN IN VIEWER

Table I.

Determinants of protest voteshare in local elections, 1983–93

	(1)	(2)	(3)
Shining Path present	0.0419 *
	(0.0176)
ln(Election violence+1)		0.0314
		(0.0296)
ln(Governance events+1)			0.0418 *
			(0.0184)
ln(Clashes+1)		-0.0372	-0.0439
		(0.0376)	(0.0384)
ln(State victimization+1)		0.0199 *	0.0180 †
		(0.0092)	(0.0103)
District FE	Yes	Yes	Yes
Election cycle FE	Yes	Yes	Yes
Observations	5,273	5,273	5,273
R 2	0.56097	0.56191	0.56245
Within R 2	0.00208	0.00423	0.00544

Standard errors clustered by district. ** p < 0.01 , * p < 0.05 , † p < 0.1.

district-level unobservables and global time trends and clustering standard errors within districts. Model 1 (Table I) suggests that the insurgents had a clear effect on the number of blank-and-spoiled ballots cast: districts where Shining Path was present saw a 4.2 percentage point increase in protest voteshare ( p < .05 ), representing roughly a fifth of a standard deviation.

What explains this effect? Model 2 evaluates the conventional wisdom, that electoral violence is responsible for the increase in protest voting. I find no support for this thesis: the effect of electoral violence is indistinguishable from zero. As theorized, electoral violence seems to have limited or counterproductive effects on voter compliance. I then turn to assessing the effect of governance. Model 3 shows that rebel governance events do predict increased levels of blank-and-spoiled ballots ( p < .05 ), with a 100% increase in governance events predicting a 4.2 percentage point incline in protest votes.

These results are robust to a range of alternative model specifications (presented in Online appendix 2), including adjustments for spatial autocorrelation, dichotomization of the count variables, and exclusion of covariates. Across specifications, exposure to Shining Path governance consistently predicts an increase in protest voteshare – suggesting that the insurgents effectively leveraged governance institutions to sway voter behavior.

Synthetic control analysis: Carrión Province, 1983

To be sure, the independent causal effect of governance on voter behavior may still be confounded by other elements of Shining Path’s activities. To further disentangle the effect of governance from the straightforward application of coercion, I study a relatively controlled episode of Shining Path governance. Drawing on qualitative and quantitative data from the CVR, I identify a province, Daniel A Carrión, in which Shining Path engaged in governance during the first three years of the war but did not consolidate territorial control or commit acts of electoral violence. This province, the only one for which the CVR reports governance activity by Shining Path but not other forms of violence, should offer a strong ‘governance’ signal while minimizing the confounding effect of direct electoral coercion. ⁸ It should thus allow me to assess the effect of governance on voter behavior relative to a baseline of no Shining Path intervention.

Shining Path in Carrión Province

Shining Path insurgents arrived in Carrión Province in 1983. Shining Path’s original base of operations was in the southern highlands region of Ayacucho, but the rebels quickly started to make their way north along the Andes corridor as the war progressed. In 1983, they began to make inroads in Carrión, one of three provinces in Pasco Region. Shining Path began with forced indoctrination. In May 1983, a group of insurgents entered the town of San Juan de Yacán and forced the residents to convene on the campus of the local high school. The CVR quotes one resident:

Once we were all in the classroom, the attackers identified themselves as members of Shining Path and talked to us about their ideology, [they kept us] until 5 pm teaching us songs.

The militants then named some residents as representatives to the group, demanded that the local authorities resign, and left.

At roughly the same time, approximately 200 Shining Path militants entered the district of Páucar. A month later, they returned and publicly assassinated several local government officials before proceeding to San Juan de Yacán, where they executed four other officials and two community members. These attacks on political structures were followed by redistributive governance. In June, 30 militants killed Alcira Benavides Albeo de Madrid in Yanahuanca district after a ‘popular trial’ in which they accused him of being a wealthy landowner; they looted his estate and redistributed his belongings and livestock among community members. According to the CVR report, news of ‘the assassination of Alcira Benavides […] rapidly spread through all the towns and annexes of Yanahuanca.’ Shining Path also assassinated two people in Páucar for ‘having extensive land and livestock’ and redistributed their belongings. A few days later, troops from a nearby military base responded by moving into San Juan de Yacán and detaining some civilians in Páucar, suggesting that Shining Path did not maintain territorial control of the province (Truth and Reconciliation Commission, 2003, Vol. 4: 197–199). Shining Path did not commit any acts of violence in Carrión province between these events and the elections held three months later.

Shining Path thus used a combination of governance strategies to position itself as a political authority in Carrión Province. It combined mandatory propaganda sessions with an effort to incorporate local residents into the group’s project, attempts to dismantle state-led political structures, and wealth redistribution meant to generate popular support. While these tactics did not rise to the level of the full-fledged systems of governance Shining Path instituted in ‘liberated zones,’ they encompass much of the group’s governance repertoire.

Analysis

Local elections were held in November 1983, three months after these events. If Shining Path’s activities generated popular support and political mobilization, we would expect to see evidence in these elections: an increase in protest ballots relative to the 1980 elections. To estimate the effect of this governance on protest voting while accounting for time trends (i.e. the change in protest voting that would have occurred in Carrión Province in the absence of rebel governance), I need a baseline to compare against.

Since no natural baseline exists, I use the synthetic control method (Abadie & Gardeazabal, 2003; Abadie, Diamond & Hainmueller, 2010) to construct a counterfactual version of Carrión Province – one that was not ‘treated’ by Shining Path governance. The synthetic control method builds an artificial version of the treated unit as a weighted combination of untreated units, optimized to closely mirror the trajectory of the treated unit’s covariates and outcomes prior to treatment. It then predicts the post-treatment outcome of that artificial, untreated unit in order to serve as a baseline to the treated unit.

Using election and census data, I build a panel of province-election cycles leading up to 1983. Since Peru was ruled by a military dictatorship from 1968 to 1980, I use results from the 1963, 1966, and 1980 local elections and the 1978 constitutional assembly election to establish the trajectory of null-and-spoiled votes across provinces. I gather this data from Tuesta Soldevilla (2001) and from original collection. In order for the synthetic Carrión Province to approximate the real one along the key sociopolitical covariates that may affect protest voting, I interpolate several statistics from the 1961, 1972, and 1981 censuses and from Albertus (2020).

To control for previous political orientation, which may correlate with likelihood of protest voting, I include a measure of leftist party voteshare (available for 1978 and 1980). I record the provincial illiteracy rate and the proportion of residents speaking Spanish as a first language to account for two potential causes of non-protest blank-or-spoiled ballots (ballots were printed in Spanish). I also incorporate a count of past social movements as a broad measure of capacity for social mobilization, either within or against electoral politics. Finally, one might expect higher levels of protest balloting where economic conditions are poor and the state is absent or ineffectual. I include province population, as well as mean elevation, road density, and proportion of cultivable land as proxies for economic opportunity, and the (log) count of government employees as a measure of local state capacity.

To establish the donor pool for the synthetic control, I subset the sample to the 74 provinces that did not experience civil war violence through 1983 and for which I have complete information. ⁹ This donor pool allows me to construct a synthetic control province that shares the social and political characteristics of Carrión that may affect protest voting, but that is not treated by Shining Path governance or electoral violence. The pre-treatment characteristics of the synthetic Carrión Province generated by this approach are summarized in Table II. Table III shows which donor provinces combined to create the synthetic province. It appears to be composed of provinces from around the country, from the south to the central highlands and Amazonian interior.

The results of the synthetic control analysis are depicted in Figure 1. The protest voteshare of the synthetic Carrión approximates the true province tightly until the treatment period, at which point it diverges sharply from the true value. This disjuncture suggests a

OPEN IN VIEWER

Table II.

Pre-treatment characteristics

Covariate	True Carrión	Synthetic Carrión	Sample mean
Protest voteshare	24.995	24.995	18.230
Leftist voteshare	35.070	35.070	22.372
Illiteracy	0.439	0.487	0.356
Prop. Spanish	0.667	0.567	0.530
Population	34562	34780	66973
Cultivable land	0.068	0.048	0.052
Elevation	3343	3438	2409
Social movements	0.000	0.156	0.699
ln(State employees)	5.342	5.342	5.849
Road density	0.000	16.771	34.336

OPEN IN VIEWER

Table III.

Synthetic Carrión province

Province (Department)	Weight
Paruro (Cusco)	0.289
Caylloma (Arequipa)	0.223
Chumbivilcas (Cusco)	0.156
Bongará (Amazonas)	0.145
Mariscal Nieto (Moquegua)	0.101
Huamalíes (Huánuco)	0.065
Tahuamanu (Madre de Dios)	0.022

substantial treatment effect: an impressive 54% of votes cast in the real Carrión Province in 1983 were blank or spoiled, compared to 34% in the synthetic control province. This treatment effect – a 20 percentage point increase – is large, equaling two standard deviations of the full sample of protest voteshares. It is also substantively significant: the fact that more than half of votes cast in Carrión Province were invalid carries a clear political message and approaches the threshold for annulling the election.

Is this effect statistically robust, or could it be spurious? I follow Abadie & Gardeazabal (2003) in evaluating statistical confidence through placebo-style tests. I conduct placebo tests for each province in the donor pool. Each of these tests fits a synthetic control for the placebo unit (excluding Carrión from the donor pool) and estimates the treatment effect. If a spurious or unobserved factor were driving the treatment effect for Carrión, it would likely show up in the results of these control units as well.

Figure 2 graphs the results of these tests. In both plots, the y-axis shows the gap between the true and synthetic protest voteshare for each placebo unit; the magnitude of that gap in the period following 1980 is the treatment effect. The top pane shows the results of the full set of placebo tests, with Carrión’s trajectory bolded. CarriónOPEN IN VIEWER

Figure 1.

Synthetic control results

OPEN IN VIEWER

Figure 2.

Placebo tests

appears to have a large but not wholly unparalleled treatment effect. However, many of the placebo units depicted suffer from poor model fit (hence the comparatively large gaps in the pre-treatment period). The bottom plot considers only placebo units for which good synthetic controls could be found (provinces with up to 10 times the mean squared prediction error (MSPE) of the treated unit). Among this more credible population of placebos, Carrión’s treatment effect appears to be substantially larger than an effect one might observe by chance. Following Abadie, Diamond, and Hainmueller (2010), I use randomization inference of the distribution of the placebos to estimate a p-value of 0.04, underlining the improbability of observing an effect of this size by chance.

One concern may be that the treatment voters in Carrión Province experienced was compound: their political behavior may have been affected by the group’s presence or use of violence to non-governance ends. Data limitations make disentangling these factors from governance and electoral coercion difficult. I nonetheless attempt to do so by narrowing the pool of donor provinces to those that, according to the CVR data and testimonies, reported Shining Path presence and use of violence but no acts of governance or electoral coercion from 1980 to 1983. These provinces should resemble Carrión in all aspects of treatment except for their exposure to governance. The results of this more conservative synthetic control analysis, presented in Online appendix 3, closely mirror the main model.

Finally, I conduct two placebo tests in time by rolling the treatment date back. If prior factors explain the results I attribute to governance by Shining Path, they may generate comparable effects in periods prior to treatment. The results of these tests are presented in Online appendix 3. In neither case does the placebo treatment approach the size and statistical significance of the true treatment, providing further evidence that the effect we observe in 1983 is due to Shining Path governance.

Does the effect of governance extend beyond 1983? Estimating the long-run effect of governance poses challenges: the national expansion of the conflict after 1983 risks spillover effects if Shining Path activity affected the donor provinces that make up the synthetic Carrión Providence in my analysis. In Online appendix 3, I nonetheless extend the analysis to the 1986, 1989, and 1993 local elections; the treatment effect remains large throughout.

Discussion

This result suggests that, consistent with a strategy of leveraging ties of authority to shape political behavior, Shining Path governance led to a substantial increase in protest voting in Carrión Province. To be sure, challenges to inference may remain: despite the presence of government troops in the province, Shining Path may have had the coercive capacity to force residents into protest voting; despite ballot secrecy, the threat of reprisals may have motivated protest votes. The covariates I was able to collect to build the synthetic control model may have failed to capture non-conflict factors driving blank-and-null voting.

But the close fit of the synthetic control prior to treatment and the large effect after suggest that something substantially shifted voting behavior in 1983. This effect may be unusually large: a uniquely effective episode of governance at a fortuitous moment of public support for Shining Path. But it may well underestimate the potential effect of governance on political behavior. Communities where Shining Path was able to fully displace state authorities and implement sophisticated, lasting political institutions likely received a higher ‘dose’ of governance than that delivered in Carrión Province. We might expect larger and longer-lasting effects in these places.

Two limitations are worth highlighting here. First, I am unable to empirically identify which aspect of Shining Path’s activity in Carrión Province may have swayed voting behavior. Voters may have responded to Shining Path’s propaganda and political mobilization, dismantling of traditional structures of authority, or redistribution of wealth. Support for Shining Path’s political agenda may have been genuine or the product of resignation and an urge for self-preservation in the new political order; in practice it was surely a combination. Better disentangling these mechanisms requires further research.

Second, these findings might not apply across the universe of armed actors. Some armed actors have parochial goals and little incentive to sway or disrupt democratic politics. Militant organizations with little interest in popular support and little fear of government reprisal may prefer direct coercion to the indirect work of building credible governance structures. In other cases, intervention in electoral politics may be unnecessary; local voters may already be sympathetic to the militants’ goals. Future work could help delineate the scope conditions for this argument.

Implications

Shining Path’s attempts to overthrow the Peruvian state were not ultimately successful. As security forces ramped up counterinsurgency efforts after 1983 and Shining Path began to struggle militarily and economically, the insurgents increasingly turned their use of violence against the rural poor in an effort to prevent collaboration with state authorities and to extract food, supplies, and conscripts. Governance institutions that had been popular became more repressive and abusive (Del Pino H, 1998; McClintock, 2001). This use of coercion against Shining Path’s putative support base was deeply ineffective. Shining Path rapidly lost popular support and was effectively marginalized after Abimael Guzmán was captured in 1992.

Nonetheless, Shining Path’s ability to influence political behavior by leveraging relations of political authority suggests important implications for scholarship on governance and conflict. First, it provides evidence of the effects of governance by armed actors on political behavior. Research on governance by non-state actors has largely focused on explaining variation in which groups govern and what forms that governance takes. This article contributes to an emerging agenda that explores the consequences of long-term exposure to governance for the attitudes, beliefs, and behaviors of civilians (Huang, 2016; Martin, Piccolino & Speight, 2022). It suggests that even short-term, indirect signals of governance by armed actors have significant effects: passing experiences of redistribution and propaganda, such as those experienced by residents of Carrión Province in 1983, are sufficient to drive political behavior. This result underlines the importance of analyzing governance in models of conflict and civilian behavior.

Second, Shining Path’s attempts to undermine Peruvian democracy suggest that, for aspiring state-builders, coercion and governance may be complements rather than substitutes. Shining Path’s efforts to sabotage Peru’s democracy were part of a broader strategy of building support for its political project at the expense of the state’s. While governance succeeded in swaying voters, coercion against candidates and bureaucrats was also highly effective – dozens of candidates dropped out of races in the face of Shining Path’s threats and elections across northern Ayacucho were canceled. Mirroring canonical models of state formation, state-building by armed actors seems to rely on a mixture of governance and coercion. Future research exploring how armed groups choose when to coerce and when to govern – and what the consequences of those choices are for political outcomes – would be fruitful.