Skip to main content

Are Actions Costlier Than Words? Formal Models of Protester-Police Dynamic Interactions and Evidence from Empirical Analysis

Abstract

Why do some protests face police repression while others are tolerated? This article formulates signaling game models to analyze the dynamic interaction between police and protesters in autocratic and democratic regimes. The theoretical framework of formal models suggests that low profile protester actions like peaceful marches with shouts avoid repressive police responses as opposed to high profile actions like close proximity contacts with the police. The equilibrium outcomes of our games are analyzed with empirical specifications that draw on two rich datasets. The empirical results of multinomial linear regression (MLR) models support the claim that police are likelier to repress protests which use aggressive actions. These results are one explanation for the Law of Coercive Responsiveness [14]. The robustness analysis of longitudinal data on protest events in the USA from 1960 to 1995 provides additional insights into the mechanism that protests relying on aggressive actions are more likely to face police violence.

Introduction

The scholarly literature investigating the relationship between contentious politics and regime response has gained significant pace in recent years. Some scholars study this nexus through quantitative formal models that explain interactions between dissidents and regime elites [2, 15, 26, 36]. Others rely on empirical data to focus on the role of nonviolent actions in toppling regimes and enforcing policy changes [47, 48]. Yet, others combine the strengths from both of these methods to deliver a better understanding about the roles of poor and middle classes in antiregime activities [22, 29, 39, 41]. However, little attention has been devoted to the variations of protesters’ tactics at the onset of nonviolent activities. This article attempts to fill that gap by analyzing how different levels of protester actions generate various responses by the police.

We demonstrate that low-profile nonviolent behavior such as use of words, chants, and statements produce more accommodating responses by the regime forces, while high-profile behavior such as physical aggression (bashing, throwing objects, close range contact with the police) increase the likelihood of more sever responses.Footnote 1 Put differently, when protesters face state protecting forces, the choice for using a discourse as a persuasion tactic lowers the propensity of violent police response as opposed to using actions to make claims. The motivations for this argument are twofold. First, protesters face less coordination problems when crowds march and shout without taking any other austere actions. Protesting crowds are easier to control when the protest leader uses slogans and motivating speeches in contrast to having activists take actions that can potentially go out of control. Second, the police have more tangible signals to activate repressive methods such as tear gas and arrests when protesters adopt aggressive physical actions even when the overall movement is nonviolent in essence. Thus, the internal costs of using high profile actions during nonviolent movements can increase for protesters and moreover become violent if some protesters deviate and display ferocious behavior.

This paper enriches our understanding of contentious politics, violence in collective action, and regime change research both theoretically and empirically. While surveying the extant research on determinants of state response to collective actions, Davenport [14] argues that state authorities will employ coercive measures against the behavior that threatens the regime stability. The author labeled this framework as the Law of Coercive Responsiveness. In this article, we formulate an analytical framework which allows to empirically test some of the postulates put forward by ([14], 7). For instance, in this article, we address the notion that protests targeting a regime change or a removal of a state leader (which we refer to as antiregime protests) are more likely to receive repressive response by the police than nonantiregime protests that target some other policy improvement. We reinvestigated the dissenter-repressor node through signaling games with two actors, protesters and the police. The equilibrium is for protesters to refrain from using high profile actions which insulate police to exercise repression. The propositions based on the signaling games are then tested with empirical estimation using large-N data of nonviolent and mixed (violent-nonviolent) movements where different protester tactics are identified. We find robust support for the idea that police respond to high profile behavioral forms of protests with greater levels of repression than the low-profile ones. We also demonstrate that this finding holds after accounting for the dissenter targets, campaign objectives, and the regime type of a state where protests occurred.

Previous Research: The Dynamics of Protests and Repression

Earlier research by political scientists and economists shows that regime change protests are prone to suppression by violence from threatened authorities [35, 36, 53]. When popular standoffs directly challenge the regime, the ruling elites typically exercise some level of coercion to stop the marching crowds and opposition leaders [14, 51]. The antiregime protests are more likely to be seized with violence as leader survival is at stake. Existing studies have designed formal mechanisms that explain the interdependence between protests and state repression. Escriba-Folch [19] finds that when facing survival dilemmas rulers are likely to extend their tenure by constraining individual liberties to limit coordination and deter collective action by implementing more violent repression against powerful individuals and groups capable of ousting the ruler by leading harmful uprisings. When the job security is not a concern, the leader is unlikely to repress, however, when the leader resorts to repression, the tactics are more severe than those of a more vulnerable leader [39]. The insecure leaders are more likely to repress when they realize that the time in office is running short.

Some scholars have analyzed the relationship between protesters and repressors through predator–prey game theoretic model which is motivated by Lotka-Volterra model. In this work, the protesters are conceptualized as the prey who are repressed by the police who are perceived as the predators. Tsebelis and Sprague [53] develop a dynamic interaction which suggests that as revolutionary action fuels relative depravations it increases the potential for state repression. Francisco [20] offers empirical examination of this model while updating its assumptions by considering the changing adaptation of protester tactics to various state responses. Leventoglu and Metternich [29] build a game theoretical model to show that urban middle classes are less likely to form rebel organizations than poor rural populations, and more likely to take their discontent with the regime out to the streets in the form of protests.

Mechanisms have been modeled to illustrate the strategic decision-making process of government repression in contentious situations. Pierskalla [36] presents one such formal model which specifies a theory of repression and dissent while focusing on coercive capacity of the government and the regime type. Two extensive form games with complete and incomplete information are proposed in this work. In the game with perfect information, the opposition (a party or organization leading the activists) and the government incur the costs of challenging and repressing, respectively. The costs for government to repress the opposition are higher and more unlikely in democracies, whereas in strong autocracies, the opposition faces a greater threat of repression. The leaders of opposition groups have better incentives for compromising their position as a result of weak decentralized security system in autocracies.

Government responses that typically hinder protester actions explain the costs and risks for participants. In their novel research on this topic, Klein and Regan [27] explore the reverse of this interaction with possibilities when protesters can spawn costs for the government to reshape its reaction to a protest. The authors propose several hypothetical frameworks that take protesters’ capacity to generate concession or disruption costs into the context.Footnote 2 They hypothesize that as protest concession costs increase the probability of coercive response by the state also increases. Additionally, increases in the disruptive costs are expected to correlate with accommodation instead of coercion by the state. However, when both disruption and coercion costs are high, the state is anticipated to respond with coercion. All these expectations are validated by empirical models using data on government responses such as crowd control, accommodation, coercion, or disregard to protests from 1990 to 2014.

Another strand of research proposes formal models which make conceivable predictions about the magnitude of grievances and repressive behavior by the state. Shadmehr [41] identifies a U-shaped framework, which shows satisfying the grievances of larger groups of protesters makes concessions more costly for the state. Hence, the state resorts to repression instead of concession. Similar conclusions are found in Gartner and Regan [21] who conceptualize a nonlinear relationship between dissident violence and state repression. The authors show that the state responds with more pressure as protester demands increase.

When hostilities emerge between the government and challenger groups, the former not only concentrates on subverting existing mobilization efforts but also on directly engaging with future prospects fueling new conflicts. Using a unique data from Guatemalan National Police between 1975 and 1985, Sullivan [49] empirically analyzes the dynamic developments of resistance through mass mobilization and repression. The observations of negative binomial panel data analysis reveal that the challenged government forecasts the challenger’s development based on the types of mobilization activities. The state repression is more likely to be directed towards radical claim-making activities.

More recent works have investigated the protest violence by synthesizing the state capacity and coercive capabilities. Using micro-level data of protests and responses in Mexico, Sullivan [50] examined how two distinct types of state capacities — coercive and authoritative, influence collective violence during protests. The results of this study reveal that the presence of weak state authority and strong coercive capacity (number of police per 100,000 people) increase the possibility of protest management through violence. Gause [22] develops a signaling game to examine the response of legislators to low- and high-resource protester groups. The author proposes two arguments. First, legislators are more attentive to the demands of protesters than those who abstain from raising their voices on the street. Second, once participants are out on the street, the legislators are more likely to support protesters from racial and ethnic minority, low income, and grassroots groups. The legislators are also more likely to support preferences of costly protests than non-costly ones as the former generates greater media attention than the latter. These claims are tested with legislative roll-call vote 102nd, 103rd, and 104th US congress data as well as daily protest activities from Dynamics of Collective Action dataset. The results based on logistic regression mostly support these arguments further suggesting that legislators are more responsive to when protest activities occur in districts where roll-call vote takes place.

One dominant view in contentious politics literature is that nonviolent resistance more than violent resistance is likely to succeed [3, 6,7,8, 16, 38, 42, 45, 48, 56]. Despite of these arguments, others have found that nonviolent protests still encounter some level of repression by the state. Examining Burma, Zimbabwe, and Burkina Faso as case studies, Tolstrup et al. [54] show that autocrats demonstrate repressive behavior against nonviolent movements based on signals of support from the regime’s great power patrons. Analyzing patron signals such as actions and statements during United Nations Security Council meetings suggests that consistent support by the regime patrons amplifies violent responses against peaceful protesters domestically.

Scholars have posited various theories and arguments to explain why nonviolent resistance works better than violent ones [4811]. However, the past research does not offer sufficient explanations to specify motivations for specific outcomes in confrontations between the police and protesters. In this article, we situate our theory to provide analytical framework that can be examined empirically. We argue that unlike violent dissenters, the protesters simulating nonviolent behavior are less likely to receive harsh responses by the police but with clear difference in democratic and autocratic regimes. We further break down protester actions to interpret how the police could respond to each type given movement objectives, targets and the regime type against which the people compete.

Theoretical Framework

The subsequent body of social movements research has largely focused on the role of performative dimension attempting to provide feasible interpretations on outcomes of various kinds of activist actions during the onset of a dissent [34, 43, 52, 11, 8, 38]. This article hinges on the idea that dissenter performances and tactics during collective action episodes create interpretable platforms for different audiences. Imperative among those audiences is the police who interact with dissenters to protect the social order. We identify two general types of signals transmitted by protesters through their tactics which determine the resolve of the police. The first type is categorized as a high-profile action, which substantively includes actions converging to violence during protests and physical contacts with the police. The second type is perceived as low-profile actions which are generally characterized as shouts and verbal expressions. Similarly, there are low and high levels of repression by the police. Figure 1 illustrates the theoretically expected relationship between different levels of protester performances and police repression. The regime type plays a central role in determining this hypothesized relationship [11, 12]. Following an established line of research, we expect that low profile protest actions are expected to receive less state repression in more democratic regimes, whereas in repressive states, the expectation fluctuates [18, 27, 38, 48, 54]. To remain in office, autocrats can infiltrate widespread violence against online activists, peaceful protesters, and marching crowds alike.

Fig. 1
figure1

Theoretical interaction between levels of protest performance and police repression in autocracies and democracies. The dotted line represents the hypothetically expected relationship in democratic societies, whereas the solid line resembles the relationship in autocratic societies. Low-level performances range from petition signing, online participation, verbal statements, and other types of use of words during gatherings. High-level performances include bashing, physical contact with the police, and property damage. These levels also range for repression exercised by the government. Low levels of repression may range from the interne censoring, verbal warnings, nonviolent arrests, hold-ups, etc. High levels of repression are material and physical repression, use of teargases, violent arrests, and protester killings

Combinations of many complex patterns drive the relationship between protesters and police in contentious actions. The objective of this theoretical framework is to simplify some of the decision-making scenarios when protesting crowds encounter resistance by the police. To make dynamics of protester-police interaction more discernable, a signaling game with two actors is outlined in this section. The actors represent two populations, protesters and the state protecting police forces. Among the previous articles offering formal theories of state repression and popular dissent, the one that develops a signaling game model defining interactions between dissenters and repressors is offered by Pierskalla [36].

The present article speaks to the theoretical framework of Pierskalla [36] in a way that it also includes two actors, dissatisfied protesters and responding government forces. However, we broaden the understanding of dynamic interactions of contentious episodes in Pierskalla [36] by interpreting the regime challenge from the onset of nonviolent protests. Put differently, while Pierskalla grants government opposition a flexibility to either organize a protest or remain silent, in this article, we analyze dynamics of protest and repression given that protesters are on the streets facing resistance (or toleration) from the state security. This allows examination of important driving factors of protests such as the tactics, methods, and responses which determine costs and utilities for actors in contentious environments. The empirical analysis of this article is also closely related to the work by Klein and Regan [27] in the vein of examination of government responses to various protester tactics, demands, and objectives. We extend their empirical models by (1) observing how tactical defections by protesters change police response and (2) focusing on more granular aspects of state responses, namely, those of police forces.

Next, in this theoretical framework, we expand the current state of knowledge by presenting scenarios where the magnitude of protester actions is decisive for the level of repression imposed by the government. In other words, while the previous literature has narrowly analyzed the importance of distinct types of protest actions on state repression [34, 43, 52],Chenoweth and Cunningham, 2013), we develop a theory that also brings forth testable hypotheses for empirical examinations. Borrowing from Davenport’s Law of Coercive Responsiveness, our theory suggests that the regime type is imperative for producing certain types of repressive outcomes in some societies but not in others. The cornerstone of the theory is that when examining various levels of protester actions, one should expect low-profile actions such as use of words to generate less violent encounters with the police, whereas higher-profile protester performances such as physical contact, sit-ins in restricted areas, and marching closer to government buildings to inflict more severe repressive practices.

Alternative Theories of Police-Protester Contentious Interactions

In almost all political systems, mass protests are supervised by the police or military in more severe cases. That condition allows generalization of the analytical framework of this project across many dimensions of politics. However, the role and response of police officers to protests starkly differs within each regime. In some cases, police exercise various levels of violence ranging from tear gases, bashing, and shooting on unarmed protestors in autocracies, hybrid regimes, and democracies. Although police and protesters are generally distinguishable during mass collective actions by their camouflage, positioning, and actions, yet, we must clarify that often police are infiltrated within protester groups, and likewise protesters have their agents within the ranks of police. However, to avoid this overcomplication, we note that we interpret police and protesters as separate entities in all our models.

Police violence can be prompted directly by the state leader or Chief of the police. The previous research on military interaction with demonstrators offers several explanations for regime survival or failure. Examining global surveys of forty military responses to nonviolent protests, Croissant, Kuehn, and Eschenauer [12] find that the type of authoritarian regime and past human rights violations by the military determine whether the police will adopt a violent or nonviolent response.

The objectives propagated by protest movements can define the extent and harshness of police oppression. Earlier works found that police may repress protest participants who challenge officers’ reputation based on claims of police brutality [38]. In democratic societies like the USA, police are likely to repress protests that directly threaten public order as opposed to protests that challenge the political survival of regime elites [17, 18]. Whereas in more authoritarian regimes, police repression is more likely to diffuse over dissenters that directly challenge regime elites [9]. The past research also shows that larger protests with confrontational tactics are more likely to experience police brutality [13]. Revolutionary protests more than protests with other types of campaign goals are likely to experience repression by the police [17, 54]. The arguments articulated by the previous research suggest that police response varies based on the objectives, methods, and the size of protest movements [46]. Although the existing research has offered sufficient empirical evidence attesting for effectiveness of nonviolent protest tactics over violent methods, the question of which nonviolent protest methods are more successful at reducing repressive backlash still remains unanswered. The next section develops formal models to answer this question. Specifically, we model the interaction between the police and dissenters discerning costs and payoffs for low- and high-profile actions in autocratic and democratic countries. We empirically test the implications of the models with datasets that identify variety of protester actions and police responses.

Signaling Game Models of Nonviolent Protests and Repression in Autocracy and Democracy

There are two key purposes of our modeling. First is to design scenarios with the conditions illustrating the Law of Coercive Responsiveness [14] to better understand police-protester interactive dynamics. Second is to improve on the existing similar models which have elucidated some variations of interaction between the government and protester groups [22, 26, 36]. To illustrate the police-protester dynamic interfaces in autocratic and democratic societies which can escalate to more violent actions, we model two signaling games with two broadly conceived types of players: the police denoted as \(P\) and protesters that hereon we will be denoted as \(D\) for dissenters. In this strategic interaction, the protesters are assumed to be nonviolent at the initial stage. The nonviolence can intensify depending on the protesters’ choice of actions and the police response.

The game begins when the Nature determines the types for \(D\), \({t}_{A}\)= antigovernment with probability \(\pi\), in autocracy and \(\rho\) in democracy, and \({t}_{N}\)= nonantigovernment with probability (1 − \(\pi\)) in autocracy and (1 − \(\rho\)) in democracy. The type \({t}_{N}\) is interpreted as protests of all other types that do not demand the resignation of the state leader.Footnote 3 The types are unknown for \(P\) until \(D\) makes the first move. The protesters are privately informed about their type from the state of nature. \(P\) forms some beliefs regarding the types of \(D\). After the types are established, \(D\) must decide whether to use low-profile actions or high-profile actions. Since we are analyzing how nonviolent protest tactics instigate repression or achieve accommodation, we generalize those tactics into two profile categories, low and high. In order to extend this generalization one step further, we interpret low profiles as verbal interactions such as slogans and shouts during marches or legal sit-ins, and high profiles as physical contact with the police officers such as bashing, throwing objects, or even property damage, e.g., any behavior that signals a credible reason for the police to implement force. The nonviolent protesters have two broad types, anti-regime and nonantiregime. The game ends and payoffs are realized after \(P\) responds with either repression \((R)\) or toleration \((T)\) to low- and high-profile protest actions. Figures 2 and 3 illustrate this game in autocratic and democratic settings, respectively. Table 1 below provides details on the symbols and mathematical notations.

Fig. 2
figure2

Extensive form signaling game between protesters and police with incomplete information in autocratic regimes

Fig. 3
figure3

Extensive form signaling game between protesters and police with incomplete information in democratic regimes

Table 1 Description of symbols and notations

Assumptions on Strategies and Payoffs of Each Player

Several important assumptions must be explained before simplifying the abstraction of payoffs for each actor. A general assumption that holds true in both types of regimes is that low-profile actions will always be less costly than high-profile actions for the protesters. That is, \({\mathrm{\rm X}}_{H}\) > \({\mathrm{\rm X}}_{L}\) and \({\Upsilon }_{H}\) > \({\Upsilon }_{L}\) in autocracies, and \({\mathrm{\rm A}}_{H}\) > \({\mathrm{A}}_{L}\) and \({\mathrm{N}}_{H}\) > \({\mathrm{N}}_{L}\) in democracies. Similarly, the police will sustain greater costs while repressing protesters with high profile actions as opposed to the low-profile actions. Also, coercing antiregime type with high-profile actions will be more expensive for the police than repressing low-profile protests of the same type. Specifically, \({\mathrm{\rm Z}}_{AH}\) > \({\mathrm{\rm Z}}_{AL}\) and \({\mathrm{\rm Z}}_{NH}\) > \({\mathrm{\rm Z}}_{NL}\) in autocracies, and \({\Lambda }_{AH}\) > \({\Lambda }_{AL}\) and \({\Lambda }_{NH}\) > \({\Lambda }_{NL}\) in democracies. The cost of toleration of antigovernment protests will be higher than nonantigovernment ones for the police in autocracies. This implies that the negative payoffs of toleration for the police are \(-{k}_{A}>\) \(-{t}_{N}\) in autocracies and \({-m}_{A}\ge\) \({-z}_{N}\) in democracies.

Further, we assume that the benefit for the police against both types of protesters differs. Particularly, the police in autocracies will benefit more from repressing antiregime protesters than nonantiregime ones as the former example could increase their reputation and effectiveness of regime protection with the autocrat. The benefits for the police thus would be,\({\{\Phi }_{AL}\), \({\Phi }_{AH}\)} > {\({\Phi }_{NL}\), \({\Phi }_{NH}\}.\) In democracies, the police would suffer from repressing nonantiregime protesters more than repressing antiregime ones as repression could undermine the credibility of the ruling government. It could also diffuse the news of violence and create greater mobilization of masses against the police and the regime itself [36]. That mentioned, as opposed to autocracies, in democracies the police will accumulate less benefits for repressing antiregime protesters with low-profile actions as opposed to the high-profile actions, \({\Gamma }_{AL}\) <\({\Gamma }_{AH}\). The police can incur the negative costs of protest diffusion denoted by \(D\) and its marginal cost denoted by\({(\alpha }_{j})\). This probability of diffusion draws on the assumption that various police responses may differently affect the society. For example, as Fig. 2 shows, in autocracies police incur \({-d\alpha }_{A}\) when tolerating antiregime protests as toleration may increase the dissenting crowd size. However, the police incur \({-d\alpha }_{N}\) when repressing nonantiregime protesters as this is interpreted to amplify dissatisfaction with the regime for not tolerating nonviolent demonstrations that do not target the regime.

Finally, in both types of regimes, the payoffs of toleration by police are always higher for antiregime type than nonantiregime type of protests. Hence, payoffs for the protesters are illustrated as \({m}_{A}>\) \({m}_{N}\) in democracies, and \({k}_{A}>\) \({t}_{N}\) in autocracies. This is intuitive because if protesters withstand the ruling government in order to change the regime and the police tolerate, then the protesters are faced with less obstacles for continuing antiregime actions. Additionally, the costs of repression will be higher for antiregime type of protesters compared to nonantiregime type in both autocracies and democracies. This leads to having \({\varphi }_{A}\)> \({\varphi }_{N}\) in autocracies, and \({\tau }_{A}\)> \({\tau }_{N}\) in democracies. The reasoning for this outcome is that antiregime protesters demand more resources, higher rates of participation, and are more prone to encountering preventive action by the police [24]. Although it is costly for the police to repress antiregime demonstrations, this will also pose considerable costs for the organizers of those protests should the police decide to exercise force.

The previous research shows that each participant in antiregime protests derives motivation from other protesters’ willingness to engage in dangerous actions [28, 30, 40]. This suggests that protesters’ observations of in-group strategies and tactics play a significant role in determining the next step for a participant. If a given protester observes that most of her group members have deviated to a different tactic, whether aggressive or nonaggressive, it is more likely that she will follow suit. We include this possibility for the entire group (that we treat as a single actor) with a strategic complementarity component \({\sigma }_{H}\) which we treat as an exogenous effect on the group deviation to another action. In other words, this exogenous effect is a parameter for switching from low to high profile actions. This probability is accompanied by a number of factors such as the failure for realization of collective intentionality or a common good during mass behavior [44]. The term \({\sigma }_{H}\) is relevant to include because even though as we have assumed high profile actions are always costlier than the low profile actions for the protesters, nevertheless, before making their move the protesters might want to take that cost and play with higher profile actions. That sort of deviation can occur when \({\sigma }_{H}>1\) and \(D\) has incentives to deviate from \(H\) to strategy action \(L\). However, when \({\sigma }_{H}\le 1,\) then \(D\) has incentives to only play \(L\).

Analysis of Police and Protester Dynamic Interactions with Incomplete Information

Which nonviolent protest tactics are more likely to yield successful outcomes? When does switching to violent resistance benefit protesters? To answer these questions, we must review the results of both games. The key contribution of our models compared to the previous signaling game models is the strategic complementarity component for protesters captured by the parameter of \({\sigma }_{H}.\) For example, when \({\sigma }_{H}>1,\) a separating profile can be supported by strategies where \(D\) plays \(H\) as type \({t}_{A}\) and \(L\) when being of type \({t}_{N}\). If \({\sigma }_{H}\le 1\) then \(D\) has incentives to deviate from its equilibrium strategy. This component more elastically controls for outcomes that were produced by deviation strategies. In this article, we rely on sequential equilibrium using Bayes’ rule to derive solution concepts.

Following the models in existing research, all separating and pooling equilibria on government and opposition interactions are considered [4, 36]. These equilibrium solutions are examined for games in both autocratic and democratic settings. To derive the solution concepts, we find all weak sequential equilibria which differ from standard sequential equilibria by having no restrictions on consistency of the beliefs [4]. The necessary proofs for the equilibrium outcomes are provided in Appendices B and C in Online Resources.

Results for Protester-Repression Game in Autocratic Regimes

The game in Fig. 2 illustrates that \(P\) prefers repression against \(D\) when the latter is of type \({t}_{A}\) and prefers to tolerate type \({t}_{N}\) regardless if \(D\) is playing \(L\) or \(H\). It must be mentioned that sequentially rational strategies for \(D\) significantly depend on the probability of deviation from common objective and outbreak of violent actions by protesters with low profiles, denoted by \({\sigma }_{H}.\)

Proposition 1

In the game with incomplete information between the police and protesters there exists a weak sequential separating equilibrium if \({\sigma }_{H}>1\) resulting in strategy profile with \(P=\left[\mathrm{Repress},\mathrm{ Tolerate}\right]\) and \({t}_{A}\to \mathrm{High}\) and \({t}_{N}\to \mathrm{Low}\). This strategy profile has prior beliefs \(\pi >({\Phi }_{AH}- {\rm Z}_{AH}) / ({k}_{A}+{\rm Z}_{AH})\) and updated beliefs \(\lambda =1\) of facing antiregime and \(\lambda =0\) of nonantiregime protesters.

In this equilibrium outcome, the police form a strong prior belief of encountering antiregime movements. The police then maximize their payoff by repressing against antigovernment protesters regardless of \(D^{\prime}{s}\) preferred strategy \(L\) or \(H\). In autocratic regimes, having some prior interaction or knowledge about the police behavior, the antiregime protesters form certain level of anticipation for police repression. Unlike the antiregime protests, the police are expected to be more tolerant with nonantiregime protests. This could steam from the regimes tendency to demonstrate to both domestic and international audiences a forbearance towards issues that the society is concerned about. After the updated beliefs, police are convinced that antiregime protesters prefer high profile while nonantiregime ones play low profile actions. If there are fewer collective action problems for sticking to low-profile tactics which will make,\({\sigma }_{H}<1\), then antiregime protesters will deviate from equilibrium payoff \(D\) to off equilibrium \(L\). The nonantiregime protesters also remain with their equilibrium strategy of making low profile actions as the police are more likely to tolerate them. The reverse of separating strategy profiles in proposition 1 are not sequentially rational regardless if \({\sigma }_{H}<or>1\) since nonantiregime protesters will always deviate from \(H\) to \(L\) since \({\Upsilon }_{L}\) > \({\Upsilon }_{H}\).

Proposition 2

In the game with incomplete information between the police and protesters there exists a weak sequential pooling equilibrium if \({\sigma }_{H}<1\) resulting in a strategy profile with \(P=\left[Repress, Tolerate\right]\) and \({t}_{A}\to Low\) and \({t}_{N}\to Low\). This strategy profile has prior beliefs \(\uppi >({\Phi }_{\mathrm{AH}}- {\mathrm{\rm Z}}_{\mathrm{AH}}) / ({\mathrm{k}}_{\mathrm{A}}+{\mathrm{\rm Z}}_{\mathrm{AH}})\) and updated beliefs \(\uplambda =1\) of facing antiregime and \(\uplambda =0\) of nonantiregime protesters.

If both types of \(D\) pool on \(L\), then the equilibrium strategy profile will be obtained only when \(\lambda =0\); in other words, \(P\) observes encountering nonantiregime protesters. In this case, \(P{^{\prime}}{s}\) best response against \(L\) given this belief will be \(C\). When \(\lambda =1\), then the best response is \(R\). In this case, \(D\) can deviate from equilibrium strategy when as mentioned before \({\sigma }_{H}>1\) as that condition would allow to switch from \(L\) to \(H\) for a higher payoff since \(\left({-\mathrm{X}}_{L}-{\varphi }_{A}({ \sigma }_{H}\right))< ({-\mathrm{X}}_{H}-{\varphi }_{A})\). However, when \(\lambda =0\), \(P\) prefers to respond with toleration, which is the optimal outcome for \(D\) from where there is no profitable deviation. The reverse of weak sequential pooling equilibria of proposition 3 where both types of \(D\) pool on \(H\) never holds for \(D\) as there is always a profitable deviation from \(H\) to \(L\) when police tolerates.

Results for Protester-Repression Game in Democratic Regimes

Unlike the previous model, the game illustrated in Fig. 3 suggests different dynamics of interactions in a democratic regime. In this game, the police prefer toleration against protesters with a low-profile action regardless if protesters are antiregime or nonantiregime. This suggests that even protests demanding the resignation of a state leader are tolerable in democracies if the protesters use peaceful tactics. However, the police are less tolerant when it comes to antiregime protests with high profile actions. It is crucial to note that both low and high profile nonantiregime actions receive toleration by the police as the probability of diffusion and the marginal cost associated with it are higher for police when choosing repression.

Proposition 3

In the game with incomplete information between the police and protesters there exists a weak sequential separating equilibrium if (1) \({-\Lambda }_{AH}<{-m}_{A}\) and (2) \({\sigma }_{H}>1\) resulting in strategy profile with \(P=\left[Repress, Tolerate\right]\) and \({t}_{A}\to High\) and \({t}_{N}\to Low\). This strategy profile has prior beliefs \(\rho >({\Gamma }_{AH}- {\Lambda }_{AH}) / ({m}_{A}+{\Lambda }_{AH})\) and updated beliefs \(\lambda =1\) of facing antiregime and \(\lambda =0\) of nonantiregime protesters.

When the first and second equalities of proposition 3 are true, then \(P\) prefers \(R\) against \({t}_{A}\) and \(D\) has no incentives to deviate. Put differently, when the cost of toleration is greater than the cost of repression, then the police resolve to repression against antiregime protesters. In this case, the protesters also choose high-profile action since there are high chances of deviation from the common interest of using low profile action. The police believe facing an antiregime movement, so they repress protesters with high action profiles. The police rather tolerate the nonantiregime low profile protesters which is also the optimal outcome for them.

If the equalities are reversed, then \(P\) prefers \(C\) over \(R\); however, \(D\) can deviate from \(H\) to \(L\) since its off-equilibrium payoff will be higher, \({m}_{A} {- {\rm A}}_{L}> {m}_{A} {- {\rm A}}_{H}\). Reversing the strategy profiles in proposition 5 does not hold properties of weak sequential separating equilibria.

Proposition 4

In the game with incomplete information between the police and protesters there exists a weak sequential pooling equilibrium with strategy profiles \(P=\left[Tolerate, Tolerate\right]\) and \({t}_{A}\to Low\) and \({t}_{N}\to Low\). This strategy profile has prior beliefs \(\rho >({\Gamma }_{AH}- {\Lambda }_{AH}) / ({m}_{A}+{\Lambda }_{AH})\) and updated beliefs \(\lambda =1\) of facing antiregime and \(\lambda =0\) of nonantiregime protesters.

In this equilibrium, both types of protesters use low profile actions to which police’s best response is toleration given updated beliefs when \(\lambda =0\) and \(\lambda =1\). Both types of protesters do not have a profitable deviation from the equilibrium strategy as \({m}_{A} {- {\rm A}}_{L}> {m}_{A} {- {\rm A}}_{H}\) and \({z}_{n} {- N}_{L}> {z}_{n} {- H}_{H}\). This implies that protests that adopt peaceful type of actions in democratic societies are more likely to attain toleration from the police than those that converge towards higher profile actions prone to violence. The reverse of this pooling equilibria where both types of protesters pool on \(H\) is never possible because \(D\) has an incentive to deviate to \(L\) when the police tolerate nonantiregime protests since \({z}_{n} {- {\rm N}}_{L}> {z}_{n}{- {\rm N}}_{H}\) is always true.

Discussion on the Equilibrium Results in Both Signaling Games

What have we learned from the equilibrium outcomes of signaling games in autocratic and democratic regimes? Do these outcomes convey any real-world phenomenon that can inform us about the direction of dissent mobilization and police repression as articulated by the Law of Coercive Responsiveness? We argue that our models are one explanation for the Law of Coercive Responsiveness as defined by Davenport [14] which suggests that to stop the threat to regime legitimacy, the authorities tend to engage in repressive behavior. First, interpreting the episodes of interaction between the police and protesters through a signaling game with an incomplete information resembles the real-world activities where the police does not typically get informed about the true nature of the dissent until protesters make their move. Second, the equilibrium outcomes given players’ beliefs and payoffs show that in both autocracies and democracies, the antiregime protesters who resort to high-profile actions such as bashing and physical encounters with the police are more likely to receive harsher responses. While antiregime protesters regardless of their profile type face repression in autocracies, in democracies, the police are more likely to peacefully endure antiregime low profile type actions.

Next, it would be relevant to test the broader implications of our theory that lower-profile actions are more beneficial than higher-profile ones for protesters in both autocratic and democratic settings. We accomplish this by testing the logic of formal models with a large-N analysis of protester interactions with the police in countries with different regime types using multinomial logistic regression. We also employ random and fixed effects probit and logit regression estimation models using daily event protest data from the USA across states spanning from 1960 to 1995. Additionally, narrowing down on the US protest data, we analyze specific cases of collective action episodes in New York from 1960 to 1990s. With this, we attempt to examine how our parameter of interest \({\sigma }_{H}\) explains deviations, if any, among protesters who initially used low (high) profile action, but later changed their tactics to high (low) profile after receiving some type of a police response.

Hypotheses and Empirical Design

The equilibrium results discussed above allow us to derive testable hypotheses to empirically assess our predictions. The main goal of this empirical examination is to test the implications of our models which are motivated by the Law of Coercive Responsiveness as outlined in Davenport [14]. First, the hypotheses we define are driven by equilibrium outcomes in both signaling games.

Hypothesis 1

Police are more likely to repress (tolerate) antiregime (nonantiregime) protesters as their actions intensify moving from low-profile (peaceful) to high profile (aggressive).

Hypothesis 2

Police are more likely to repress high-profile actions in both autocratic and democratic regimes.

The parameter that accounts for the deviation from initial action is of central importance in this paper. Thus, we also frame a hypothesis that allows us to synthesize some of the protest events that occurred in New York during 1960s and 1990s in order to observe trajectories of switching protester strategies and subsequent police responses to those strategies.

Hypothesis 3

Police are more likely to repress protesters who deviate from initially used low-profile (peaceful) actions to high-profile (aggressive) actions. Similarly, police will be more tolerant towards protesters who deviate from high profile to low profile actions.

We must note that assumptions in hypothesis 3 will hold only if the protesters target non-government (foreign or local) entities. As the protesters target foreign country embassies, consulates, residential or business offices of political actors, police or other military personnel then the chances of getting repressed increase.

Hypothesis 4

When targeting government buildings or the state protecting police, protesters will be repressed regardless of their action profile.

Second, to empirically test these hypotheses, we use Nonviolent and Violent Campaigns and Outcomes dataset (NAVCO) version 3.0 compiled by [10] This dataset includes actor-level information on protester tactics, movement objectives, and various state responses that offer unique opportunity to examine the dynamic interactions identified in the signaling games of this paper. The NAVCO dataset consists of event-day observations where each event is defined by variables that range from the protesters’ targeted actor to economic impact of the event. First, we filter this extensive dataset to only include events where the targeted actor is the police. This allows to isolate other actors that could potentially spoil the specific interaction between protesters and the police as described by our formal models. Second, we further exclude the missing observations which narrow down the total count of observations to 204 across different countries and time periods. Some countries like Iraq have only 3 observations, while others like China and Egypt have several dozens. To clarify, we do not estimate within or across country variations in panel data analysis fashion but instead examine the impact of overall protester strategies on police responses in the sample of interest.

Similar to several of the previous studies on contentious politics, we also rely on multinomial logistic regression (MLR) to analyze how protester tactics effect the police response [48]. This method has several advantages. First, by taking into account each independent variable and its categories, it allows to separately estimate the effect on distinct responses of the dependent variable. Second, it has a flexibility of not assuming linearity and homoscedasticity among regressors; however, it requires independence among the categories of the dependent variable [23]. Third, MLR enables including both continuous and binary regressors for estimation of predicted probability of the outcome variable.

Dependent Variable

We use the state response, more particularly the police response, to each protest event in a country on a specific day. This variable is categorical which is measured as follows:

$$\mathrm{Response}=\left\{\begin{array}{c}1= full\;accomodation\\ 2=material\;concession\\ 3=non-material\;concession\\ 4=neutral\\ 5=non-material\;and\;non-physical\;repression\\ 6=material\;and-or\;physical\;repression\;short\;of\;killing\\ 7=material\;and-or\;physical\;repression\;intended\;to\;result\;in\;death\end{array}\right.$$

After rescaling these measurements, original values of 1, 2, and 3 now equal to 0 representing police toleration, while 5, 6, and 7 are equaled to 3 to indicate police repression. The neutral value is equated to 2.Footnote 4

Key Explanatory Variables

We include three regressors to obtain the conditions specified in the signaling games as closely as possible, namely, predicting which nonviolent protests still face police repression given variations in their tactics. The first regressor is the overall category of the nonviolent protest, abbreviated for simplification as Protest, which is also a categorically measured variable where:

$$\mathrm{Protest}=\left\{\begin{array}{c}0= persuasion:expression\;of\;objection\;of\;disapproval\;by\;words\;only\\ 1=protests:expression\;of\;objection\;or\;disapproval\;by\;actions\\ 2=non\;cooperation:deliberate\;restriction, discontinuance,\;\\ with\;holding,\;or\;a\;combination\;of\;these,\;of \\\;social\;economic\;or\;political\;with\;opponent individuals,\;\\\;activities,\;institutions,\;or\;a\;government\;during\;conflict\\\;3=intervention:direct\;interference\;in\;a\;situation\;by\;nonviolent\\\;means. Most\;often \;physical,\;such as sit-in,\;occupation,\\\;or\;obstruction.\;The\;most\;militant\;nonviolent\;action\end{array}\right.$$

We rescale this variable to account for low- and high-profile protester actions by coding 0 equal to low-profile actions, and 1, 2, and 3 to equal to 1 which represents the high-profile action.Footnote 5 In order to account for antiregime and nonantiregime protests, we employ another variable which identifies the stated goals of the campaign event. This variable, shortened as Campaign, is measured in the following categories:

$$\mathrm{Campaign}=\left\{\begin{array}{c}0=regime\;change \\ 1=significant\;institutional\;reform\\ 2=policy\;change\\ 3=territorial\;secession\\\;4=greater\;autonomy\\\;5=anti-corruption\\\;6=unknown\end{array}\right.$$

We recode this variable’s original measurements by indexing the original values of 0 and 1 as antiregime which equal to 2 in the new measure, and previous values of 2, 3, 4, and 5 as nonantiregime which now are measured as 1. Lastly, we leave 6 as the unknown category which equals to 0. The third regressor is the polity scores of each country from Polity IV project [32]. This variable, shortened as Polity, ranges from − 10 to + 10 with the lowest value indicating complete autocracy, while the highest value indicating complete democracy. The autocratic and democratic regime types distinguish police and protester payoffs in the two signaling games. This variable is important to include since it will account for the effects of different regimes where protester and police interactions took place. Additionally, we include two interaction terms between Protest × Campaign and Protest × Polity to determine whether the predicted police response to protesters’ preferred actions depend on the type of the campaign and the regime type. In other words, as theorized by both signaling games, whether the police response to low- and high-profile actions depends on antiregime and nonantiregime campaigns.

The primary research objective of this paper is to provide a better understanding for factors impelling the police repression towards low- and high-profile protester actions. As such, we remove any protest events that are coded as completely violent; however, in order to tease out the tactical defections of demonstrators, we allow flexibility where mixed nonviolent and some violent tactics were exercised. Given the measurements of our explanatory variables, the very low-profile action is Protest = 0 where protester have only used words as a persuasion tactic. The highest-profile action as measured here is Protest = 3 where physical intervention such as occupying a street has been practiced by participants. As informed by our hypotheses, we expect to obtain a positive estimated coefficient on the Protest variable in relation to police repression since the repression is likelier as the protester actions intensify. We also would expect the coefficient on the Campaign variable to be positive for repression and negative for toleration as hypothesized earlier the police are more likely to repress antiregime actors than nonantiregime ones in autocratic states. In democratic states, this outcome would depend on police’s assessment of the cost for toleration as opposed to the cost for repression. Finally, we expect the coefficient on Polity to be negative for repression and positive for toleration because police are more likely to accommodate protesters as democracy levels increase.

Results of Empirical Analysis

The estimated effects of each explanatory variable along with the effect of interaction terms on the probability of police repression are reported in the Table 2. The results reveal several interesting observations that relate to the main concepts of the Law of Coercive Responsiveness. The results also provide empirical support for our hypothesized relationship between the variables. First, the police are more than 46 times likelier to repress high profile protester actions as opposed to low profile actions.Footnote 6 This result supports our general theory that when nonviolent protests rely on persuasive tactics through the use of verbal expressions the police are less likely to repress. This result also supports the assumptions of the first hypothesis. Second, as expected, the coefficient of the Campaign variable is positive showing that antiregime protests are 11 times more likely to receive police repression than nonantiregime ones. Although statistically not significant, the coefficient on Campaign is positive for police toleration. Even when considering protest objectives in Protest × Campaign interaction term, we observe that nonviolent protests are only slightly more likely to encounter police violence.Footnote 7

Table 2 Estimated effects of protester action profiles, campaign goals, and polity scores on police repression

Third, the predicted effect of interaction term Protest × Polity indicates that high profile actions in more democratic regimes are somewhat likelier to face police repression with the coefficient of 0.741 as opposed to the police toleration with coefficient of 0.683. Lastly, the coefficient on Polity informs that police are marginally more likely to tolerate protest events as democracy scores increase. Whereas, the police repression also has statistically significant coefficient which suggests that protesters still face some level of police repression in more democratic societies. In sum, these results provide some justification for the equilibrium outcomes of both signaling games. The outcomes based on MLR analysis suggest that low profile protester actions such as expressions of objections through words are less likely to face police repression. This effect also holds during nonantiregime protests where high profile actions face more repression than low profile ones. Reflecting back on propositions 1 and 3, to claim that nonantiregime protests with high profile actions face less repression than antiregime ones would be inaccurate based on these empirical results. Next, it will be relevant to further investigate how different properties of equilibrium outcomes are illustrated in specific antiregime and nonantiregime protest incidents that are not exemplified by the available data.

Sensitivity Checks

To closely observe country specific implications of the expected relationship between protesters’ preferred action types and police responses, we conduct robustness analyses. We use micro level event-day protest data ranging from 1960 to 1995 for each state in the USA that comes from Dynamic of Collective Action dataset [31]. With growing popularity in econometrics during the past few decades, the random and fixed effects allow to confront the unobserved heterogeneity bias [25, 37]. The researcher might face a problem of unobserved heterogeneity bias when together with the observed variables there are other factors that correlate with the variables of interest ([33]: 303). The practical qualities of both random and fixed effects estimators offer leverage for dealing with the challenge of endogeneity bias posed by the OLS regression on the panel data analysis [5].

Since our dataset is longitudinal, we construct both random and fixed effects logit nonlinear models that account for cross-sectional and time-series nature of the data. The dependent variable is Response, which is a binary variable where 1 = police violence against protesters and 0 = no violence. We use three explanatory variables all of which are also binaries. Violence is a binary where 1 = protester violence which we interpret as high profile actions and 0 = no violence, that is low profile actions. Activities is another binary measure where 1 = high intensity actions and 0 = low intensity actions. Form is binary variable which differs from Violence and Activities by measuring the overall form of protester actions such as 1 = riot or melee (high profile action) and 0 = symbolic display (low profile action). The last binary variable is Target which is scaled as 1 = if protesters have targeted government/state or foreign government/state and 0 = other targets such as private business, university, medical facilities. To the best of its capacity, this variable allows us to determine whether or not protesters are of anti-regime/government or nonanti-regime/government type. A detailed explanation on the original and rescaled measurements for these variables is included in the Appendix A in Online Resources.

We estimate a random effects probit as well as fixed and random effects logit models in order to determine the probability of police repression against protesters who have exerted either violent (high) or nonviolent (low) type of tactics. The fixed effects logit model is subject specific instead of population-averaged which will allow to extract the impact of daily events of protester actions on police responses that are time dependent [55]. We account for time-invariant heterogeneity between US states by employing random effects models. We also include two interaction terms, Activities × Target and Forms × Target to determine whether police response to protesters’ actions is different for targeting government entities as opposed to others.

The results of logit and probit models in both tables reveal that when protesters resort to more violent tactics, the police are more likely to implement a violent response. Again, this finding fits well with the logic for the Law of Coercive Responsiveness [14]. This affect held true given the type of activities, forms of activities, and specific actors/institutions targeted by protesters. More specifically, the probit random effects model in both tables shows that police are more likely to use violence against protesters who also displayed a violent type of a behavior. The statistically significant effect of protester tactics on police response still holds after we control for who protesters target and what type of activities they exercised during violent and nonviolent collective actions. Additionally, with the interaction terms in both tables, we also control for the effects of protesters targets (which is a proxy of whether or not the movement is antiregime or nonantiregime) that is due to their choice of activities (Table 3) and forms of protests (Table 4). The interaction terms allow us to extend the dynamic association of protester actions and police response that are intertwined with movement objectives.

Table 3 Estimated effects of protester activity types, targets, and interaction between the two on police repression in the USA, 1965–1995
Table 4 Estimated effects of protester action forms, targets, and interaction between the two on police repression in the USA, 1965–1995

Since this sensitivity analysis employs within country characteristics of protest groups in forty-nine US states, we do not have a measure for regime type. While random effects logit models suffer from omitted variable bias which in our analysis could be unmeasured variables such as state ideology, partisanship, and racism during 1960s and onwards, nevertheless, with the fixed effects logit model, we are able to control for regional characteristics by holding state effects constant [1].Footnote 8 The results in models (2) and (3) in both tables suggest that protester violence significantly predicts the probability of violence by the police overtime. Model (2) in the Table 3 shows that when protesters have switched from nonviolent action to violent ones, the police were 3.2 times more likely to respond with violence. This effect is also evident after controlling for state effects in model (3). In this model, the police are 3.3 times more likely to respond with violence to shifting tactics of protesters.

Protests that rely on switching tactics, for example, moving from flag waving to physical aggression, are 0.77 times more likely to receive violent response by the police based on both random- and fixed-effects models in Table 3. Protests that target domestic or foreign government representatives or organizational institutions are 1.17 (model 2) to 1.2 (model 3) times more likely to encounter police violence. While the variable Forms resembles no statistically significant effect on predicting police violence by itself, the interaction with protester targets enhances its impact. More specifically, activists who targeted government institutions in addition to rioting instead of exercising peaceful symbolic displays were 1.39 times more likely to receive a violent police reaction. In sum, the results shown in models 1, 2, and 3 in Tables 3 and 4 provide support for hypotheses 1 and 2 which are based on assumptions of the equilibrium outcomes. Except the forms of protests that had no significant impact in determining police violence, other variables such as protester violence, activities, and targets demonstrated a stronger relationship with our outcome variable of interest which is the police violence.

Synthesizing Deviation Strategies and Police Responses

Lastly, we analyze the impact of strategic complementarity parameter in our models to better understand protester-police interaction routines and how they are altered after the police intervention. In this effort, we extend the analysis one step further by only focusing on selected daily protest events in the state of New York from 1960 to 1995. We observe specific cases where after event participants switched tactics either from violent to nonviolent or vice versa, the police responded with either repression or toleration. The specific cases where participants changed their tactics and behavior are summarized in the Table 5 below. For ease of reference, we have numbered the protests with no particular hierarchical order except based on the date of occurrence.

Table 5 Selected collective action events in New York from 1960 to 1995 where participants change their initial action strategy after police response

The outcomes of dissent policing in New York from 1960 to 1990s have been heavily influenced by authorities’ discriminatory attitudes towards members of minority groups such as Black Americans, Latinx Americans, LGBTQ + , and immigrants from communist countries. The apparent differences in discriminatory police responses to some of these groups are illustrated in the Table 5. After a meticulous inspection of this table, we notice that in line with our expectations in hypothesis 3, police are more likely to repress protests that start off as peaceful but later escalate to demonstrations with aggressive behavior. For instance, collective action event numbers (hereon referred to as E-no for brevity) 4, 5, and 7 in Table 5 show that E-no-4 and E-no-5 were guided by low profile actions picketing and marching to which police did not respond with repressive measures, nevertheless, after deviating to high profile actions like between-group clashes the police stepped forward with a use of force. Same outcome emerged at E-no-7, where civil rights peaceful protests turned into violent encounters between participants and the police. The police responses in E-no-1, 3, 8, 14, 16, 17, 18, and 19 also support our claims in the third hypothesis. Particularly, collective actions and protests that started with strong use of high-profile actions such as clashes with police, blocking traffic, and object throwing but later turned into relatively peaceful rallies, sign-holdings, and marches attained police toleration.

Contrary to the those expected interactions, the outcomes in E-no-2, 6, 9, 10, 11, 13, and 15 reveal that in some cases, police can still repress even when protesters switch from high profile actions to low-profile ones. A closer observation of E-no-1, 2, 3, 6, 8, and 12 suggests that as hypothesized in H4, police repress activists who take a stand against government entities irrespective of their initial action choice. However, police discrimination against Black activists compared to other minority groups is evident in the type of police responses after the activists switch from high profile to low profile strategies. More specifically, when comparing E-no-1 and 2, we notice that Hungarian and Ukrainian activists protesting in front of a United Nations (U.N.) building, and Pro-Castro Cuban protesters that marched against US involvement in Cuba, faced no police repression after switching from low to high profile actions. Unlike these two ethnic groups, in E-no-3, we see that African Americans who protested against U.N. policies implemented against Congo encountered police repression even after deviating from more aggressive to peaceful actions. Another example of disproportionate targeting of Black activists by the police is found in E-no-10, 11, and 13. In these collective action events, the police employed repressive measures against students and other activists protesting against inferior education and segregation after the participants changed their actions from high profile to low profile.

Concluding Remarks

Despite the volume of research on protest movements and various state responses, a number of questions are still unanswered. Why some protest movements instigated a violent response by the police and themselves turned violent while others remained peaceful? Which protester tactics are more successful at winning police toleration? How do these dynamics of interactions vary between different regime types when considering the logic of Law of Coercive Responsiveness? To tackle these questions, the current paper formulated signaling games in the context of autocratic and democratic regimes where police have imperfect information about protester types until the latter takes an action. The equilibrium outcomes of our signaling games suggest that police are more likely to repress high profile aggressive antiregime protester actions in autocracies, while in democracies police should be more tolerant of such events. The empirical examinations of these theories based on NAVCO 3.0 dataset revealed that high profile protester actions do in fact face greater police repression. Empirical analysis revealed that, surprisingly, even though overall the police are more likely to repress protests in autocratic societies, some protests with high profile actions are prone to repressive police reaction in democratic states as well.

We also tested the propositions of our formal models by employing random and fixed effects probit and logit models for robustness check. Using the Dynamics of Collective Action dataset, our robustness check allowed to examine the expected relationship between protester tactics and police response of signaling game outcomes by focusing on longitudinal data only from the USA. This micro level analysis enabled to focus on specificities of protester actions, forms, targets, and nature of particular activities. The sensitivity check showed that after controlling for time invariant (random effects) and time-dependent (fixed effects) covariates, we still observe significant police violence against protesters who targeted government instead of nongovernment actors while relying on violence (high profile) as their preferred action type.

This paper expanded on the existing protest literature in several major ways. First, to the best of our knowledge, this work is the first to formally model a dynamic relationship between state protecting forces such as police and protesters expressing grievances to systematically examine the Law of Coercive Responsiveness proposed by Davenport [14]. Additionally, the two separate signaling games which we developed account for scenarios that could unfold in democratic as well as in more restrictive autocratic regimes. This differentiation presents a richer understanding of tactical choices by both protesters and police in societies with diverse institutional environments. Second, we derived testable hypotheses based on the signaling game propositions. We tested those hypotheses with data covering multiple countries which enabled us to control for the regime type of each country. Third, we conducted a sensitive analysis by estimating probability models with longitudinal data which confirmed the assumptions of our hypothesized association between protester actions and police responses within one ethnically diverse country with complex history of protests and racial targeting by the police, USA. Finally, our signaling games introduced a strategic complementarity component for protesters making it more flexible to control for deviation strategies. We tested this complementarity for deviation with a close examination of collective action events in the state of New York between 1960 and 1990s and found that police repressed protesters who changed from peaceful to more aggressive actions. However, the police response varied from one racial group to another while heavily disfavoring the Black activists.

This paper offers some key policy relevant insights for organizers of social, protest, and grassroot movements. First, we learned that peaceful tactics prolong the longevity of the entire agenda for protesters in democracies; however, if the movement does not specifically target overthrowing a state leader, it can also last in autocracies. Principle characteristics of nonviolent civil resistance comprise an important dimension of factors that help dissenters deliver positive changes to the state governance. Second, the availability of information on prior encounters with state protecting forces significantly impacts the strategies and action types of protesters. In particular, protesters striving to achieve a policy change or even a regime transition in repressive societies need to pay closer attention to the prior encounters between police and other protesters. If the police are approached with aggression, they are more likely to retaliate with similar violence regardless of which type of a regime the protests occur. Third, most social movements lose their relevance because of their shorter lifespan. However, by winning the hearts and minds of regime protecting coercive forces, the people power is more likely to avoid atrocities, save human lives, and achieve policy objectives. 

Data Availability

The author is willing to share the data upon publication.

Code Availability

The author is willing to share the software codes used for empirical analysis upon publication and reasonable request.

Notes

  1. 1.

    To clarify, while we provide an empirical validity for this logic in our estimation models; however, in the real world, many instances of soft tactics like words and statements do generate violent response by the police.

  2. 2.

    Disruptive costs are defined as the cumulative function of protest location, duration, and size. Concession cost index is operationalized as the aggregate function of protesters’ demands, protester violence, and recurrent demands.

  3. 3.

    Klein and Regan [27] suggest that protests which demand the resignation of the state leader are costlier than protests that present other types of demands such as resource distribution and social rights (pg. 489).

  4. 4.

    We use “Neutral” as our reference category for the multinomial logistic analysis.

  5. 5.

    In the original NAVCO dataset, there is another category 4 = political engagement: dialogue or negotiations, but we do not report it here since after filtering our dataset had no observations with this category as a tactical option for protesters.

  6. 6.

    This result of high probability can be affected by the size and specific observations in our dataset.

  7. 7.

    This result should be taken with some caution because of the lack of data on protests that seek regime change, of which there are only 26 out of 204 observations in our dataset.

  8. 8.

    The reason is that in the logistical regression context the fixed effects do not control for unobserved time-dependent covariates (see, [55].

References

  1. 1.

    Allison PD (2005) Fixed effects regression methods for longitudinal data using SAS. SAS Institute, Cary, NC

    Google Scholar 

  2. 2.

    Angeletos G-M, Hellwig C, Pavan A (2007) Dynamic global games of regime change: learning, multiplicity, and the timing of attacks. Econometrica 75(3):711–756

    Article  Google Scholar 

  3. 3.

    Asher B (1999) Nonviolent social movements: a geographical perspective. Blackwell, Malden, Mass.

    Google Scholar 

  4. 4.

    Banks JS, Sobel J (1987) Equilibrium selection in signaling games. Econometrica 50(3):647–661

    Article  Google Scholar 

  5. 5.

    Bell A, Jones K (2015) Explaining fixed effects: random effects modeling of time-series, cross sectional and panel data. Political Science Research and Methods (PSRM). 3(1):133–153

  6. 6.

    Boudreau V (2004) Resisting dictatorship: repression and protest in Southeast Asia. Cambridge University Press, New York

    Book  Google Scholar 

  7. 7.

    Burgess B, Burgess H, Peter W (1994) Justice without violence. Boulder, CO: Lynne Rienner

  8. 8.

    Butcher C, Svensson I (2016) Manufacturing dissent: modernization and the onset of major nonviolent resistance campaigns. J Conflict Resolut 60(2):311–339

    Article  Google Scholar 

  9. 9.

    Chang P, Vitale A (2013) Repressive coverage in an authoritarian context: threat, weakness, and legitimacy in South Korea’s democracy movement. Mobilization 18(1):19–39

    Article  Google Scholar 

  10. 10.

    Chenoweth E, Pinckney J, Lewis OA (2017) Nonviolent and violent campaigns and outcomes dataset, v. 3.0. University of Denver

  11. 11.

    Chenoweth E, Gallagher Cunningham K (eds) (2013) Understanding nonviolent resistance: an introduction. Special issue. J Peace Res. 50(3)

  12. 12.

    Croissant A, Kuehn D, Eschenauer T (2018) Mass protests and the military. J Democr 29(3):141–155

    Article  Google Scholar 

  13. 13.

    Davenport C, Soule S, Armstrong D (2011) Protesting while Black? The differential policing of American Activism, 1960–1990”. Am Sociol Rev 76(1):152–178

    Article  Google Scholar 

  14. 14.

    Davenport C (2007) State repression and political order. Annu Rev Polit Sci 10:1–23

    Article  Google Scholar 

  15. 15.

    Dragu T, Lupu Y (2018) Collective action and constraints on repression at the endgame. Comp Pol Stud 51(8):1042–1073

    Article  Google Scholar 

  16. 16.

    Dudouet V (2013) Dynamics and factors of transition from armed struggle to nonviolent resistance. J Peace Res 50(3):401–413

    Article  Google Scholar 

  17. 17.

    Earl J, Soule SA, McCarthy John D (2003) Protests under fire? Explaining protest policing. Am Sociol Rev 69:581–606

    Article  Google Scholar 

  18. 18.

    Earl J, Soule SA (2006) Seeing blue: a police-centered explanation of protest policing. Mobilization 11:145–164

    Article  Google Scholar 

  19. 19.

    Escriba-Folch A (2013) Repression, political threats, and survival under autocracy. Int Polit Sci Rev 34(5):543–560

    Article  Google Scholar 

  20. 20.

    Francisco RA (1995) The relationship between coercion and protest: an empirical evaluation in three coercive states. J Conflict Resolut 39(2):263–282

    Article  Google Scholar 

  21. 21.

    Gartner SS, Regan PM (1996) Threat and repression: the nonlinear relationship between government and opposition violence. J Peace Res 33(3):273

    Article  Google Scholar 

  22. 22.

    LaGina G (2020) Revealing issue salience via costly protest: how legislative behavior following protest advantages low-resource groups. Br J Polit Sci. 1–21 https://doi.org/10.1017/S0007123420000423

  23. 23.

    Greene WH (1993) Econometric Analysis, 2nd ed., Macmillan

  24. 24.

    Greitens SC (2016) Dictators and their secret police-coercive institutions and state violence. New York: Cambridge University Press

  25. 25.

    Hsiao C (2003) Analysis of panel data. Cambridge University Press

    Book  Google Scholar 

  26. 26.

    Huang H (2013) Signal left, turn right: central rhetoric and local reform in China. Polit Res Q 66:292–305

    Article  Google Scholar 

  27. 27.

    Klein GR, Regan PM (2018) Dynamics of political protests. Int Organ 72:485–521

    Article  Google Scholar 

  28. 28.

    Kuran T (1991) Now out of never: the element of surprise in the East European Revolution of 1989. World Politics 44(1):7–48

    Article  Google Scholar 

  29. 29.

    Leventoglu B, Metternich NW (2018) Born weak, growing strong: anti-government protests as a signal of rebel strength in the context of Civil Wars. American Journal of Political Science 62(3):581–596

    Article  Google Scholar 

  30. 30.

    Lohmann S (1993) A signaling model of informative and manipulative political action. American Political Science Review 87(2):319–333

    Article  Google Scholar 

  31. 31.

    McAdam D, McCarthy J Olzak S, Soule S nd (2019) Dynamics of collective action. http://www.stanford.edu/group/collectiveaction/ (accessed 14 Oct 2019)

  32. 32.

    Marshall MG, Jaggers K (2016) Polity IV project: political regime characteristics and transitions. Retrieved from http://www.systemicpeace.org/polity/polity4.htm

  33. 33.

    Mátyás L, Sevestre P (eds) (2008) The econometrics of panel data: fundamentals and recent developments in theory and practice, 3rd edn. Springer, New York

    Google Scholar 

  34. 34.

    Nepstad SE (2011) Nonviolent revolutions: civil resistance in the late 20th century. Oxford University Press, New York

    Book  Google Scholar 

  35. 35.

    North D, Wallis J, Weingast B (2009) Violence and social orders: a conceptual framework for interpreting recorded human history. Cambridge University Press, Cambridge

    Book  Google Scholar 

  36. 36.

    Pierskalla JH (2010) Protest, deterrence, and escalation: the strategic calculus of government repression. J Conflict Resolut 54(1):117–145

    Article  Google Scholar 

  37. 37.

    Raffalovich LE, Chung R (2014) Models for pooled time-series cross-section data. Int J Confl Violence 8(2):210–221

    Google Scholar 

  38. 38.

    Reynolds-Stenson H (2018) Protesting the police: anti-police brutality claims as a predictor of police repression of protest. Soc Mov Stud 17(1):14–63

    Google Scholar 

  39. 39.

    Ritter EH (2014) Policy disputes, political survival, and the onset and severity of state repression. J Conflict Resolut 58(1):143–168

    Article  Google Scholar 

  40. 40.

    Shadmehr M, Bernhardt D (2011) Collective action with uncertain payoffs: coordination, public signals, and punishment dilemmas. American Political Science Review 105(4):829–851

    Article  Google Scholar 

  41. 41.

    Shadmehr M ( 2014) Mobilization, repression, and revolutions: grievances and opportunities in contentious politics. J Pol 76(3):621–635

  42. 42.

    Schock K (2005) Unarmed insurrections: people power movements in non-democracies. University of Minnesota Press, Minneapolis

    Google Scholar 

  43. 43.

    Schock K (2013) The practice and study of civil resistance. J Peace Res 50(3):277–290

    Article  Google Scholar 

  44. 44.

    Searle JR (1990) Collective intentions and actions. Cambridge. MA: MIT Press

  45. 45.

    Sharp G (1973) The politics of nonviolent action. Porter Sargent, Boston, MA

    Google Scholar 

  46. 46.

    Sharp G (2005) Waging nonviolent struggle: 20th century practice and 21st century potential. Porter Sargent, Boston

    Google Scholar 

  47. 47.

    Solt F (2015) Economic inequality and nonviolent protest. Social Sciences Quarterly 96(5)

  48. 48.

    Stephan MJ, Chenoweth E (2008) Why civil resistance works: the strategic logic of nonviolent conflict. Int Secur 33:7–44

    Article  Google Scholar 

  49. 49.

    Sullivan CM (2016) Undermining resistance: mobilization, repression, and the enforcement of political order. J Conflict Resolut 60(7):1163–1190

    Article  Google Scholar 

  50. 50.

    Sullivan H (2019) Sticks, stones, and broken bones: protest violence and the state. J Conflict Resolut 63(3):700–726

    Article  Google Scholar 

  51. 51.

    Tansey, O, Koehler K, Alexander S (2018) Ties to the rest: autocratic linkages and regime survival. Comparative Political Studies, 1–34

  52. 52.

    Tilly C (2008) Contentious performances. Cambridge University Press, Cambridge, England

    Book  Google Scholar 

  53. 53.

    Tsebelis G, Sprague J (1989) Coercion and revolution: variations on a predator-prey model. Mathematical Computer Modeling 12(4/5):547–559

    Article  Google Scholar 

  54. 54.

    Tolstrup J, Seeberg MA, Glavind JG (2019) Signals of support from great power patrons and the use of repression during nonviolent protests. Comp Pol Stud 52(4):514–543

    Article  Google Scholar 

  55. 55.

    Wilson JR, Lorenz KA (2015) Fixed effects logistic regression model. In: Modeling binary correlated responses using SAS, SPSS and R. ICSA Book Series in Statistics, vol 9. Springer, Cham

  56. 56.

    Zunes S (1994) Unarmed insurrections against authoritarian governments in the third world: a new kind of revolution. Third World Quarterly 15:403–442

    Article  Google Scholar 

Download references

Acknowledgements

I would like to thank Jan Pierskalla, Stergios Skaperdas, David Meyer, Bernard Grofman, Samantha Vortherms, Marek Kaminski, Jeff Kopstein and the editor Panos Pardalos and annonymous reviewers for very useful comments and suggestions on earlier drafts of this paper. The current version of this work has also benefited from commentary received from students and professors alike in Experimental Economics I and II graduate courses taught by John Duffy and Michael McBride, respectively, at UC Irvine. Additionally, I would like to extend my gratitude to PhD student colleagues Avik Sanyal, Andrew Benson, Nishtha Sharma, Mark Hup and Patrick Julius in the Economics department at UCI for tremendously stimulating discussions.

Funding

This research was supported by the Jack W. Peltason Center for the Study of Democracy and the Institute of Humane Studies summer graduate research fellowship Center for Global Peace and Conflict Studies, both at UC Irvine.

Author information

Affiliations

Authors

Contributions

Not applicable.

Corresponding author

Correspondence to Sargis Karavardanyan.

Ethics declarations

Ethics Approval

Not applicable.

Consent to Participate

Not applicable.

Consent for Publication

Not applicable.

Conflict of Interest

The author declares no competing interests.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary Information

Below is the link to the electronic supplementary material.

Supplementary file1 (DOCX 29 KB)

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Karavardanyan, S. Are Actions Costlier Than Words? Formal Models of Protester-Police Dynamic Interactions and Evidence from Empirical Analysis. Oper. Res. Forum 2, 54 (2021). https://doi.org/10.1007/s43069-021-00099-4

Download citation

Keywords

  • Protest and repression
  • Signaling games
  • Risk and uncertainty
  • Collective decision making
  • Cooperation
  • Authoritarianism