Repression and Dissent: How Tit-for-Tat Leads to Violent and Nonviolent Resistance

Much research examines the state-dissident nexus by large-n studies and rational choice theories. This article contributes an analysis of dissident reasoning through a computational evaluation of ethnographic interviews. The analysis shows that dissident decision-making is based on tit-for-tat deliberations: Dissidents choose violent means primarily in response to violent repression, and nonviolent means in response to nonviolent repression. Ordinary citizens not participating in dissent consider positive state behavior or safety concerns instead. Consistent with arguments that state-dissident interactions are reciprocal, these findings reveal unexpected cognitive similarities between political dissent and cooperation, which is often associated with tit-for-tat deliberations. They also show the importance of state repression compared with other motivators of dissent, including perceived relative deprivation and social contagion. The findings identify heuristic patterns of reasoning which suggest that dissidents may be more open to change and, ultimately, cooperation with state authorities than what is argued by repressive states.

It is well known that state repression and political dissent are connected (Chenoweth, Perkoski, and Kang 2017;Combes and Fillieule 2011;Davenport 2007;Gurr 2016;Lichbach 1987;Tilly 1978).While there is a consensus in the repression-dissent literature that dissent generates repression (Chenoweth, Perkovski, and Kang 2017;Davenport 2007), it remains unclear how repression shapes dissent.Some studies suggest that repression increases dissent (e.g., Francisco 1995Francisco , 1996)), whereas others indicate that repression dampens dissent (e.g., Young 2019).Others show that repression is connected to changes in the means of dissent (Lichbach 1987;Moore 1998), while further studies suggest that the relationship between repression and dissent is reciprocal (Carey 2006), or endogenous (Ritter and Conrad 2016).
This article contributes new insight by exploring the cognitive processes of political dissidents.Based on a computational analysis of ethnographic interviews, the analysis provides new, in-depth insight into a field largely examined by large-n and gametheoretical studies.Comparing the reasoning of violent and nonviolent dissidents, the analysis shows that political dissent is based on tit-for-tat deliberations focusing on state repression.Violent dissidents focus their deliberations on increasingly violent repression, whereas nonviolent dissidents focus their reasoning on less violent and nonviolent forms of repression.Ordinary citizens who do not engage in dissent think of positive state behavior and safety instead.
These findings reveal unexpected similarities between the deliberations related to political dissent and cooperation, which has often been associated with tit-for-tat strategies (Axelrod 1981;Nowak and Sigmund 1992).Accordingly, interactions are based on the principle of reciprocity, where an actor chooses to mimic the behavior previously adopted by the other actor.A fundamental component of social systems (Gouldner 1960), reciprocity is usually linked to rational agents who reward kind but punish unkind behavior (Falk and Fischbacher 2006).While some evidence based on aggregate data at the national level suggests that the relationship between repression and dissent is reciprocal (Carey 2006(Carey , 2009)), it has not yet been demonstrated empirically that dissidents reason according to the principle of reciprocity, thus resorting to basic rules of societal conduct to challenge political regimes.
Data were constructed from ethnographic interviews with political dissidents, which were coded into cognitive maps that represent the reasoning processes underlying political behavior (Axelrod 1976).The reasoning processes contain a large number of beliefs, addressing factors as varied as structural conditions, political events, perceived deprivation, social bonds, religious convictions, or emotions.This variety exceeds what is captured by theoretical cognitive models of political dissent (e.g., Moore 1998) and provides more fine-grained insight into the dissident decision-making process.In addition, cognitive maps visualize connections between beliefs that make visible how deliberating about these diverse factors motivates decisions (belief → belief → … → decision).As such, the maps provide a very detailed, bottom-up view of the cognitive processes underlying political behavior, which adds important information to existing theoretical accounts of reasoning.
To compare the deliberations of political dissidents, the analysis builds on existing computational studies of cognitive maps (Bonham and Shapiro in Axelrod 1976;Andreou, Mateou, and Zombanakis 2005;Dornschneider and Henderson 2016;Mehryar et al. 2020;Young 1996).An investigation of belief frequencies shows that dissident deliberations focus mostly on state repression as opposed to other factors addressed by the maps.Visualizations of the maps moreover identify connections between increasingly violent repression and dissident decisions favoring violence, as opposed to less severe, and nonviolent repression and dissident decisions embracing nonviolent means.
To identify the beliefs that are key connectors to decisions, the analysis conducts a computational counterfactual experiment.The experiment systematically removes beliefs from the cognitive maps and traces the resulting changes in dissident decisionmaking (cf.Pearl's "external interventions", 2009).Interventions on beliefs addressing state repression strongly reduce decisions to engage in dissent, while interventions on other beliefs do not create similar changes.Beliefs addressing economic grievances and perceived deprivation, which are frequently associated with dissent (e.g., Gurr 2016;Salehyan and Stewart 2016;Griffin, De Jonge, and Velasco-Guachalla 2020), show no comparable role in dissident deliberations.Beliefs addressing others' participation in dissent and interaction with dissidents, which are crucial in the social contagion of dissent (Granovetter 1978;González-Bailón et al. 2011;Kuran 1991;Noelle-Neumann 1974;Saideman 2012) play an observable role, which is nevertheless not comparable to that of state repression.
Our analysis follows an inductive research design, which is increasingly applied in other disciplines to challenge existing research paradigms (Corley, Bansal, and Yu 2021) and maximize the value of the data (Jebb, Parrigon, and Woo 2017).Acknowledging the complementary roles of inductive and hypothesis-based science (Kell and Oliver 2004), we apply computational methods of data analysis to advance our understanding of dissident behavior.Constructing data from ethnographic interviews (cf.Spradley 1979), the analysis aligns with new studies of computational ethnography (Abramson et al. 2018), while responding to calls for incorporating ethnographic methods into political science (Wedeen 2010) and studies of political dissent (Fu and Simmons 2021).
The inductive analysis first manually examines each interview to identify the main components of cognitive mapsbeliefs, belief connections, and decisions.The subsequent computational analysis systematically explores connections between beliefs and decision without a priori focusing on beliefs addressing certain factors, such as rational choice calculations, state behavior, environmental factors, or religious convictions.The findings obtained from this analysis show the relative importance of beliefs addressing state repression as opposed to beliefs addressing other factors, thus underlining the value of the literature on the state-dissident nexus.The findings also identify heuristic patterns of reasoning that complement the existing literatures on both rational choice and the diffusion of dissent.
Consistent with psychology studies showing that reasoning frequently deviates from basic assumptions of rational choice (Gigerenzer and Goldstein 1996;Kahneman 2011;Simon 1985), the identified dissident deliberations show that goal-oriented cost-benefit calculations are often skipped, while positive emotions related to repression (hope to defeat the state, courage to face the state, solidarity with repression victims, pride to oppose the state) and beliefs contradictory to a chosen form of dissent (e.g., beliefs that peaceful resistance had been unsuccessful among nonviolent dissidents) are included in dissident decision-making.Moreover, the findings bridge the psychology literature on heuristics and sociological studies of the diffusion of dissent by showing that the deliberations of nonviolent dissidents may follow the "imitate the majority" heuristic (Boyd and Richerson 2001), according to which decisions are based on others' dissident behavior (Noelle-Neumann 1974;Granovetter 1978;Kuran 1991;Noelle-Neumann 1974).
The following sections present the psychological processes related to dissident behavior, with a focus on the literatures addressed by the main findings of the analysis.The article then provides descriptions of repression from ethnographic interviews, a computational analysis of the interviews, and quotes to illustrate the key findings.The online supplemental materials provide additional information on the data construction, the computational model, and findings.

The Logic of Dissent
The literature has identified two main responses available to dissidents under repression: Increasing (backlash) or decreasing their activity (dampening). 2In spite of their contradictory nature, both responses are observed in a large variety of repressive settings, 3 making it inherently difficult to differentiate between them.Insight into dissident reasoning is key to gain a better understanding. 4

Rational Choice
Typically, the literature on the state-dissident nexus infers information on dissident reasoning by adopting a rational choice perspective.Accordingly, dissident deliberations rely on strategic, utility-driven calculations that weigh the costs and benefits of dissident behavior.To understand dissident choices, repression is a major factor, as cost-benefit calculations may vary depending on differing evaluations of repression.Consistent with theories on dampening, dissidents may consider repression as a deterrent that raises the costs of their behavior (Gupta, Singh, and Sprague 1993;Lichbach 1987;Moore 1998, 852-853;Opp and Roehl 1990).In line with backlash theories, dissidents may, however, also perceive of repression as a motivator that convinces them of the necessity to oppose the state, thus increasing the benefits of their behavior (Lichbach 1987;Opp and Roehl 1990;Moore 1998; see Goodwin 2001 andHafez 2003 for related accounts of dissident behavior).

Dornschneider-Elkink and Henderson
Following rational choice models, these contradictory evaluations are linked to varying costs associated with repression magnitudes.Accordingly, dissidents and state actors increase their behavior until a certain magnitude of repression is reached and dissent becomes too costly, so that the state-dissident relationship resembles that of an inverted-u (Muller and Weede 1990; also see Lichbach and Gurr, 1981;Lichbach 1987;Regan and Henderson 2002;Gurr 2016).In this process, dissidents may attempt to curb the costs of their behavior by decentralizing their networks (Grimm and Harders 2018), dispersing their methods (e.g., from protest to strike, see Schock 2005), continuing some forms of dissent while stopping others (cf.Sullivan 2016), or changing the location of dissent (Baczko, Dorronsoro and Quesnay 2018, 133-145).Conversely, dissidents under increasing repression may also decide to increase the cost of their behavior by substituting nonviolent with violent means (Lichbach 1987;Moore 1998; also see Della Porta and Tarrow, 1986)  5 , or adjusting their activities to the most recently experienced repression (Carey 2006;Francisco 1996).These riskier decisions are assumed to be grounded in calculations of chances of success according to which the potential benefits outweigh the increased costs of dissent (Enders and Sandler 2012). 6 So far, it has not been studied empirically whether or how rational choice assumptions are represented in dissident reasoning.The following analysis adds primary data on such reasoning.Consistent with the literature, the analysis underlines the importance of state repression (Finding 1) and means (Finding 2). 7Contrasting with rational choice assumptions, according to which dissidents choose violent versus nonviolent means based on their respective chances of success, dissident deliberations focus primarily on violent versus nonviolent repression, and lead to decisions that mimic the state's chosen means (tit-for-tat, see "Heuristics").The skipping of costbenefit calculations in empirically observed dissident deliberations highlights the relevance of heuristics in reasoning, and the limits of rational choice to understand dissident behavior.

Heuristics
Research in psychology shows that basic assumptions of rational choice are frequently violated by real-world actors (Gigerenzer and Todd 1999;Kahneman 2011).Consistent with Simon's concept of bounded rationality (1997), this literature suggests that political choices are based on heuristics, cognitive shortcuts that skip calculations associated with rational choice.While the literature on the state-dissident nexus does not typically integrate these findings, some research on nonviolent dissent highlights the importance of heuristics.Contrary to rational choice models, this research typically assumes that dissidents underestimate or skip reasoning on state repression and its associated costs.
Diffusion studies show that people decide to join protest once a certain threshold has been reached and a sufficiently large number of protesters are on the streets (Granovetter 1978), making it safe to reveal their true preferences in public (Kuran 1991;Noelle-Neumann 1974).Research in psychology describes this type of deliberation as the "imitate the majority" heuristic (Boyd and Richerson 2001).Research on nonviolent dissent (Weyland 2012) moreover highlights the heuristics of availability (Tversky and Kahneman 1973) and representativeness (Kahneman and Tversky 1972), according to which dissident decisions are based on events that immediately come to mind (availability), or on events' resemblance to another (representativeness).
The following analysis shows that dissident reasoning is characterized by heuristic features.Rather than neglecting or underestimating state repression, dissident deliberations are found to rely on tit-for-tat reasoning focusing on repressive state behavior (Finding 3).Tit-for-tat is a social heuristic known to outperform standard rational choice strategies in competitive games (Raab and Gigerenzer 2005) that has not yet been known as a heuristic associated with dissident behavior.Its identification in dissident reasoning bridges rational choice theory and the psychology literature on heuristics by emphasizing the relevance of state repression, while de-emphasizing the importance of cost-benefit calculations.The analysis also finds evidence that nonviolent dissent involves the "imitate the majority" heuristic, providing new empirical support for threshold theories of protest (Finding 4).

Emotions
Heuristic-based reasoning happens faster than rational choice deliberations (Kahneman 2011), and involves emotional features (Metcalfe and Mischel 1999).Much literature has examined discrete emotions to account for the choices of political dissidents (Gurr 2016;Jasper 1998Jasper , 2014;;Pearlman 2013;Van Zomeren, Postmes, and Spears 2008), consistent with insight from cognitive science showing that emotions and cognition are interdependent (Jasper 2018;Lerner, Valdesolo and Kassam, 2015;Ochsner and Gross 2008;Young 2019).Following this literature, political dissent is related to emotions of anger and frustration as opposed to fear, which influence dissidents' assessment of the situation and translate into different choices (Pearlman 2013).
Psychological studies show consistently that anger is an important motivator of political mobilization (Van Zomeren, Postmes, and Spears 2008).Anger is also known to play an important role in violent dissent, such as terrorism (Baele 2017), civil war (Wood 2003), and intergroup aggression more generally (Spanovic et al. 2010).Frustration is considered a potential motivator of violent dissent in particular.It is assumed that interference with an actor's effort, for example through state repression, creates frustration, which may subsequently generate aggressive inclinations (Berkowitz 1989;Gurr 2016, 241).By contrast, anger is considered to contribute to positive assessments of a situation, an inclination of risk-taking (Lerner and Keltner 2001), and action tendencies.This role is especially strong related to non-structural types of disadvantages, such as state repression (Van Zomeren, Postmes, and Spears 2008).
Fear is considered an emotion deterring political dissent (Kuran 1991;Noelle-Neumann 1974;Spanovic et al. 2010;Pearlman 2013;Young 2019).In the psychology Dornschneider-Elkink and Henderson literature, the deterring effect of fear is related to pessimistic assessment of the situation, in which dissidents perceive a low sense of control and are risk adverse (Lerner and Keltner 2001;Young 2019).Rather than evaluating repression as a form of perceived injustice, fear-based appraisals perceive of repression as a type of threat to the safety of the individual or group.
Emotions are of great importance in relative deprivation theory, which highlights the importance of perceived social grievances, such as poverty or social exclusion (Best and Krueger 2011;Flam 2004;Gurr 2016;Jasper 2014;Van Zomeren, Postmes, and Spears 2008).Although this literature does not explicitly focus on emotions, it shows that anger and frustration may play an important mediating role in the relationship between perceived deprivation and dissent.
The following study adds to this literature by showing that dissident decisionmaking mimics state repression, rather than fearfully avoiding it.While the existing literature highlights the role of negative emotions triggered by repression, the following analysis shows that repression may be connected to positive emotions, namely hope to bring down the repressive regime, solidarity with the victims of repression, courage to face violent state authorities, and national pride (Finding 5).Contrasting with relative deprivation theory, the analysis finds no support for a role of poverty in dissident decision-making (Finding 6).

Data Construction
This study adopts the cognitive mapping approach.Cognitive mapping traces reasoning processes from the direct speech of political actors.Originally intended to help policy-makers improve their decision-making (Axelrod 1976), cognitive mapping has contributed great insight into subjects as varied as domestic and foreign policy (Galtung 1973;Hart 1977), belief congruence and European integration (Van Esch, Joosen, and Van Zuydam, 2016), policy-making in complex social-ecological systems (Mehryar et al. 2019), or coercive organizational politics (Voyer 1994).
Typically, cognitive mapping is applied to fields in which direct speech is readily accessible through public records (Axelrod 1976;Levi and Tetlock 1980;Van Esch, Rik, and Van Zuydam, 2016).In security studies, where public access is limited, cognitive mapping has mostly been applied to politicians, whose thought processes are documented by speeches, records of meetings, or war games (Bonham and Shapiro in Axelrod 1976;Bonham, Shapiro, Trumble 1979;Johnston 1995;Klein and Cooper 1982;Levi and Tetlock 1980;Olsson 2019;Schafer and Walker 2006).By contrast, the reasoning of political dissidents is more difficult to access, and there are few cognitive mapping applications to this field to date, with our own study of a subset of the data analyzed in this article constituting the only exception we are aware of (Dornschneider and Henderson 2016). 8 Cognitive maps consist of three main components: Beliefs, connections between beliefs (inferences), and decisions.To construct cognitive maps, this study applied qualitative analysis based on inductive coding procedures developed by Spradley (1979) and Strauss and Corbin (1990).Each sentence was coded for the main components of cognitive maps: Beliefs were identified from content words, sub-clauses or entire sentences.Inferences were identified from linguistic connectors, such as "because," "if…then," and from the temporal order in which factors were mentioned.Decisions were identified from verbs related to participation in violent or nonviolent dissent, such as "participate," "decide," "protest," or "kill." The resulting belief structures make visible detailed deliberations underlying political behavior.These deliberations consist of chains of inferences that connect beliefs to decisions (belief → belief → … → decision).An inference consists of at least one antecedent belief which triggers a consequent belief (antecedent → consequent).Inferences may include several antecedents that together trigger a consequent (antecedent + antecedent + … → consequent) or that trigger a consequent separately (antecedent a → consequent; antecedent b → consequent). 9Inference chains include a number of inferences, in which most beliefs are located in positions of both antecedents and consequents, except for the decisions, which are pure consequents, and beliefs that have no antecedents, called pure antecedents.
Constructed from inductive coding of direct speech, cognitive maps capture deliberations that both exceed and complement rational choice models, emotions, or heuristics.Rather than inferring the logic of dissident behavior from external observations, they make visible the actors' reasoning from their own descriptions.Based on this feature, they offer a unique method to explore reasoning underlying political behavior inductively.

Data Sources
The following analysis investigates dissidents from four different political settings and, to ensure that the identified deliberations are unique to political dissidents, a control group of ordinary citizens not participating in dissent (cf.Bennett 2004).The analysis shows that decision-making by violent and nonviolent dissidents in these settings is based primarily on beliefs about state repression.By contrast, ordinary citizens reason about positive state behavior and safety when deciding to refrain from participation in dissent.
The first dataset contains 27 cognitive maps by violent and nonviolent dissidents, constructed from ethnographic interviews (Dornschneider and Henderson 2016).Due to the sensitive nature of primary research on violent dissidents, data were assembled over a long time frame of 2 years.While the number of interviewees (27 dissidents) is average for ethnographic studies (Mason 2010), the sample size (27 maps) is large for cognitive mapping studies (cf.Axelrod 1976).Interviews were conducted with individuals who engaged in political dissent in Egypt and Germany between the 1970s and 1990s.These two states display very different features associated with dissent,

Dornschneider-Elkink and Henderson
including autocratic versus democratic regimes, poor versus wealthy living conditions, Muslim versus Christian majority populations, etc.Their selection ensures that the findings are not subject to specific features inherent to a particular interview setting.
Egypt is an authoritarian state well-known for repressive measures (El-Nadim 2017; Human Rights Watch 2020a).When Sadat came to power in 1970, the government applied a policy of permissiveness, easing repressive measures applied by President Nasser.This policy turned into mass repression over the following decade until Sadat's assassination in 1981 and beyond (Ansari 1987;Brownlee 2010;Hafez 2003, 84).In this climate, both violent and nonviolent forms of dissent co-existed.The main groups exercising political dissent were al-Jihad and al-Jama'et al. Islamiyya, which engaged in violent dissent, as opposed to the Muslim Brotherhood, which exercised nonviolent dissent.The data contain seven cognitive maps of members of al-Jihad and al-Jama'at al-Islamiyya as well as eight cognitive maps of members of the Muslim Brotherhood.
Germany is a democracy and not usually associated with repression.However, the cognitive maps of German dissidents contain numerous beliefs about repressive state behavior.Repression in Germany occurred after large-scale student protests broke out in the 1960s, leading to violent clashes between the protesters and the police (Della Porta 1995).The most well-known act of police violence is the murder of an unarmed protestor, Benno Ohnesorg.Various dissident groups confronted the German police with violent and nonviolent means, including two violent groups, the Red Army Faction (RAF) and Bewegung 2. Juni (the day Ohnesorg was killed).The data contain four cognitive maps of RAF members and two maps of Bewegung 2. Juni members.The most prominent nonviolent groups were the German Socialist Student Union (SDS) and Kommune 1.The data include four cognitive maps of SDS members and two maps of Kommune 1 members.
The second dataset contains 121 cognitive maps by nonviolent dissidents and ordinary citizens who did not participate in the Arab Spring in Egypt and Morocco (Dornschneider, 2021). 10The two settings were selected because they had different experiences of the Arab Spring: While the Egyptian President resigned after a few weeks, the Moroccan King stayed in power throughout the uprisings.Despite these differences, the analysis finds that, in both settings, both dissidents and ordinary people reasoned primarily about state repression when deliberating about joining the uprisings.
To confront the uprisings, many Arab states applied repression (Brownlee, Masoud, and Reynolds 2015).The Egyptian state relied on the security forces and paramilitary groups to repress the protestors through teargas, beatings, and shootings (Grimm and Harders 2018;Rudbeck, Mukherjee and Nelson 2016).The Moroccan state initially relied on selective measures, but later applied repression in the form of beatings and arrests (Abouzzohour 2019;Human Rights Watch 2020b;Lawrence 2017;Rachidi 2019).Nevertheless, unprecedented numbers of protestors participated in the uprisings in both countries.The second dataset contains 53 cognitive maps of Egyptian and Moroccan protestors and 68 maps of individuals who did not join the protests.
In spite of the different time frames and settings related to the data, the vast majority of cognitive maps contains beliefs about state repression.The following analysis shows that, despite numerous additional beliefs addressing other factors, reasoning on state repression is crucial to decisions on both violent and nonviolent dissent.

Analysis
The unit of analysis are the beliefs that constitute dissident deliberations.Dataset 1 contains 106 unique beliefs, and Dataset 2 contains 135 unique beliefs (see Figures 1  and 2 for excerpts), which translate into large numbers of belief combinations (2 106 and 2 135 respectively).These combinations are connected to decisions through hundreds of inferences.Although the large number of beliefs and inferences capture numerous aspects of dissident deliberations, thus allowing for a more fine-grained examination, they also create a difficulty for the analysis of reasoning on state repression.Building on previous cognitive mapping applications, we conduct a computational analysis that traces belief frequencies, inferential connections between beliefs and dissident decisions, and counterfactual reasoning processes.

Beliefs on State Repression
The cognitive maps contain beliefs addressing a large range of factors (see Figures 1  and 2).State repression is among the beliefs included by most maps, pointing to its relative importance in dissent deliberations. 11 Dataset 1 contains three beliefs addressing state repression, called "aggression by home state," "domination by home state," and "continuation of aggression by home state." 12Aggression by home state addresses violent forms of state repression and imprisonment.It was identified from quotes, including "Sadat arrested more than 3000 of our people," "the government arrests and tortures Islamists," "hundreds of Muslim Brothers were killed," "the police was very violent against us," "for no reason, a policeman took out his gun and shot in the air."Domination by home state addresses repressive conditions, including authoritarianism, lack of freedom, and limited opportunities for opposition.It was identified from quotes, such as "some people run this country by force," "the people did not elect the government," "the people are not free," "the government was very strict with the opposition," or "our teachers had also taught as Nazis."Continuation of aggression by home state addresses the recurrence of state repression and was identified from quotes, such as "the government attacks me more," "the government uses more violence," "we thought that the arrests would go on forever," "there was a circle of violence conducted by the state." "Aggression by home state" and "domination by home state" are among the top five beliefs in Dataset 1, occurring in 25 and 22 cognitive maps respectively (Figure 1).The remaining top beliefs are "peaceful activity against the state" 13 (24 maps), "transformatory goals" (22 maps), and "support for violence in direct environment" (18 maps)."Continuation of aggression by home state" occurs in 14 maps.By comparison, 67 beliefs occur in five or less maps.Overall, Dataset 1 includes beliefs addressing factors as varied as the state environment, the international environment, Dornschneider-Elkink and Henderson state acceptance, state strength, the opposition's strength, dissident goals, availability of means, peaceful activity against state, consequences of violence, consequences of peaceful activity, personality, private life, and decisions.
Dataset 2 contains two beliefs addressing state repression, called "state crime" and "state violence." 14"State crime" refers to repressive conditions, identified from quotes including "There is a violation of rights," "There is no democracy, freedom or popular will," "The people's dignity has been violated," "They have stolen every inch of our country," and "They are using poverty, hunger, and the reduction of education."State violence refers to violent repression and was identified from quotes including "Three armed cars of riot police, tons of hired thugs, and officers came down to terrorize us," or "I saw a woman with a bleeding head.I asked her what happened, but the police hit her." "State crime" occurs in 50 cognitive maps, which is the highest number among the beliefs contained by Dataset 2 (Figure 2)."State violence" is the 14th most frequent belief, occurring in 17 maps.Overall, Dataset 2 includes beliefs addressing factors as varied as emotions, state behavior, external conditions, events, actions, needs, preferences, attitudes towards the state and protestors, state capabilities, state personality, non-state actors, opposition capabilities, opposition personality, religion, and decisions.The online supplemental materials visualize the belief counts related to the comparative groups contained by both datasets.

Counterfactual Reasoning Without Repression
The belief frequencies show how often dissidents consider state repression in their deliberations.While this indicates the relevance of repression in dissident thinking, further analysis is needed to show that dissident decisions are actually based on state repression.This is necessary because other beliefs that occur less frequently could still be more commonly connected to dissident decisions than the beliefs addressing state repression.This is possible because of the large number of inferences connecting beliefs to decisions. 15 To account for this possibility, the analysis applies a counterfactual experiment.Following Pearl's concept of external interventions (2009), the analysis first removes beliefs from the cognitive maps.Based on the inferential connections between the remaining beliefs, the analysis then tracks changes in the dissidents' decisions.The beliefs whose removal is associated with decision change are identified as crucial to the reasoning process, as their connection to decisions cannot be replaced by the remaining beliefs.
To conduct the counterfactual experiment, the analysis assembles an experimental set of beliefs that activates decisions in each of the cognitive maps contained by the two datasets.Due to their connections with decisions, these beliefs are most likely to indicate decision changes in the counterfactual experiment. 16The experimental set consists of 17 beliefs about the world, and as such includes several beliefs that speak to major theories of dissent.In particular, it includes the mentioned beliefs addressing violent and nonviolent state repression, as well as beliefs addressing poverty, which is considered a major motivator of dissent by grievance models (Gurr 2016; also see Van Zomeren, Postmes, and Spears 2008), and social influence, which is key according to social contagion theories of dissent (e.g., Noelle-Neumann 1974;Kuran 1991;González-Bailón et al. 2011;Saideman 2012).In addition to these beliefs referring to the previously discussed literature, the experimental set includes beliefs on state behavior violating religious norms, which is an important contributor to dissent according to religious theories of violence (Canetti et al. 2010;Juergensmeyer 2017). 17The online supplemental materials list the experimental beliefs and provide detailed information on the analysis.
The analysis systematically removes all possible combinations of the experimental beliefs (2 17 ¼ 131; 072) and tracks the resulting changes in dissident decision-making.Each counterfactual combination is matched and compared with the original belief combination, so that decision changes can be linked to the particular beliefs that are removed in each run.The analysis of Dataset 1 identifies 192 of all possible combinations whose removal is related to a decision change in each cognitive map, resulting in zero decisions by both violent and nonviolent dissidents.Each of the 192 combinations contains the mentioned beliefs addressing state repression ("continuation of aggression by home state", "aggression by home state", "domination by home state") and another belief called "education of the people". 18The analysis of Dataset 2 identifies 1024 combinations whose removal results in decision change in all but three dissident maps.Each of these combinations contain beliefs on violent and nonviolent repression, as well as beliefs addressing poverty, successful protest abroad, protest by fellow citizens, and interaction with protestors.To assess the relative importance of the beliefs whose removal maximally reduces decision counts, the analysis compares the differences between the counterfactual and the original distributions, as depicted in Figures 3 and 4. Table 1 moreover shows the differences in average decision counts before and after the removal of each belief.
The main finding is that dissident decision-making is based on beliefs on state repression (Finding 1), with decisions of violent dissidents being related to violent repression, and decisions of nonviolent dissidents to nonviolent repression (Findings 2-3).Regarding Dataset 1, the removal of the belief addressing violent repression ("aggression by home state") is connected to the biggest change in the decision counts of violent dissidents (difference in means Δμ ¼ 5:25).Its removal reduces the mode from 10 to zero, and the decision range from 5-13 to 0-9.By contrast, the removal of the belief addressing nonviolent repression ("domination by home state") results in the biggest change of decision counts of nonviolent dissidents (Δμ ¼ 4:15).Its removal changes the distribution from a unimodal shape with the mode at 10 into a bimodal shape with peaks at 4 and 9.It moreover reduces the decision range from 5-14 to 0-11.Regarding violent dissidents, the removal of "domination by home state" has a weaker impact (Δμ ¼ 2:37).
The removal of the belief addressing recurring violent repression ("continuation of aggression by home state") strongly changes the decision counts of violent dissidents (Δμ ¼ 4:00), but only has a minimal impact on the decisions of nonviolent dissents (Δμ ¼ 0:50).Regarding violent dissidents, removing the belief changes the mode from eight to zero, and the decision range from 4-13 to 0-10.Regarding nonviolent dissidents, the mode is reduced from ten to nine, and the range from 1-14 to 0-14.
Regarding Dataset 2, the removal of the belief addressing nonviolent repression ("state crime") also has the strongest impact on the decision counts of nonviolent dissidents (Δμ ¼ 17:11).Its removal reduces the mode from 22 to 8, and the range from 13-53 to 3-23.By contrast, the removal of the belief addressing violent repression ("state violence") has an observably weaker impact (Δμ ¼ 3:30).It lowers the mode from 11 to 7, and the range from 4-53 to 3-48.
The decision counts of the control group of non-dissidents show no comparable impact of the removal of "state crime" (Δμ ¼ 5:60) or "state violence" (Δμ ¼ 1:60).Instead, the biggest changes are associated with the removal of beliefs called "poverty" and "safety." The analysis of Dataset 2 moreover suggests that decision-making of nonviolent dissidents is based on protest behavior of others (Finding 4).The removal of "protest by fellow citizens" has an observable impact on the decision counts of nonviolent dissidents (Δμ ¼ 6:70; see the online supplemental materials for visualizations).By contrast, the removal of "poverty" has a comparably minor impact (Finding 6).Regarding Dataset 1, neither the removal of "religious government" nor of "unreligious government" has a major impact.

Reasoning Processes About State Repression
To investigate how beliefs on violent and nonviolent repression motivate decisions, the analysis traces reasoning processes.This shows that reasoning of violent dissidents (Dataset 1) is characterized by a focus on increasingly repressive state behavior (Findings 1-2). 19Deliberations contain inferences from beliefs that the state is engaging in violence to beliefs that the state is engaging in continuous violence ("aggression by home state" → "continuations aggression by home state"), or inferences from beliefs that the state is engaging in nonviolent repression to beliefs that the state is  Aggression by home state 9.00 3.75 5.25 [5.22, 5.28] 10.00 6.62 3.38 [3.35, 3.40 -0.11, 0.11] 34.63 28.95 5.68 [5.59, 5.75] engaging in violent repression ("domination by home state" → "aggression by home state"). 20In the words of a member of the RAF: "At the beginning we went [to demonstrations] without stones.But then we made the bitter experience of always returning with bruises and black eyes.The tiniest violation of a rule was always immediately answered with the most extreme violence.So the question of not using violence did not pose itself." As illustrated by this quote, the deliberations of violent dissidents may not address strategic considerations regarding the costs and benefits of violent dissent, or the likelihood of achieving dissident goals (Finding 3). 21Moreover, they include numerous beliefs that are not captured by rational choice theory (see the online supplemental materials for an overview), and beliefs that contradict common understanding of violent resistance.As he reflected on killing President Sadat, the founder of the Egyptian al-Jama'at al-Islamiyya remembered his prospects for a successful career in the system."Of course, I had an alternative," he said."I'm a doctor."In his view, leaders of other violent groups were in a similar situation: "Look at bin Laden.He is a millionaire, but lives like a beggar." A member of the RAF said he always hated violence and tried to avoid fights in high school.When explaining his decision to blow up the German embassy in Stockholm, 22  he said: "There was an insane superpower trying to suppress the people by terror and torture.
We saw that we were lost, there was an atmosphere of death.We said we will die here [in Stockholm].We would have been destroyed anyway.I saw: either you fight and lose or you don't fight and lose.So it was obvious that you fought." By contrast, nonviolent dissident reasoning is differentiated by focusing on comparatively less repressive state behavior, namely nonviolent repression or non-recurring violent repression (Dataset 1 and 2) (Findings 1-2). 23Further underlining the limits of rational choice, more than a fourth 24 of the nonviolent dissidents believe that their peaceful resistance against the state has been unsuccessful (Dataset 1).A nonviolent interviewee said that "demonstrations did not result in anything," but that he did not switch to violent means because "whatever you gain through violence will always be violence" (Finding 3).
Other 25 nonviolent dissidents share the belief that violence is the wrong response to state repression (Dataset 1).More than half 26 believe that violent dissent provokes violent state repression (Finding 3)."If I use violence," a Muslim Brother said."The state will kill us all."A leader of the German Socialist Student Union said: have no chance to survive.In my opinion, the student movement was going to collapse if the RAF continued to push for this atmosphere of civil war." Various other 29 nonviolent dissidents interacted closely with violent groups, but objected to violent dissent, contrasting expectations from diffusion theories and research on violent networks, which suggest that they should have adopted the reasoning of violent dissidents instead (e.g., Sageman 2004).
The reasoning of nonviolent dissidents (Dataset 2) is also characterized by positive emotions, namely hope to defeat the repressive regime, solidarity with nonviolent dissidents, courage to confront the repressive regime, and pride (Finding 5). 30In the words of a dissident: "I have two girls, the oldest is four.When my girls grow up, they will boast that their father played a role in the liberation from oppression and corruption."Another dissident said: "The state was always watching us, it was very bad.Politicians could have done anythingcreating lies or arresting people."When she discovered that uprisings were happening in Tunisia, she was carried away by positive emotions: "I felt incredibly happy.My heart was with those people." The deliberations of the control group of non-dissidents include observably less beliefs on nonviolent repression, 31 and more beliefs on positive state behavior (Findings 1-3). 32In addition, they involve safety considerations. 33In the words of a non-participant: "We have the king [Mohammad VI].You can see on TV how he is with children.He is modest, he is a king of the people.We can move freely on the streets without fear."Another non-participant said: "Everyone here lives well.But they [the protestors] do what they want, asking for more salary.One demonstration leads to the next, and there is killing, war, and blood."

Conclusions
Much of the literature on the state-dissident nexus infers insight into dissident deliberations from large-n studies of external factors or rational choice models.While this research has produced great theoretical insight, empirical research in psychology has shown that the deliberations underlying social behavior frequently violate basic rational choice assumptions.Analyzing dissident reasoning identified from ethnographic interviews, this article provides new empirical evidence on dissident deliberations, deepening our understanding of the psychology underlying state-dissident interactions.
The analysis finds that dissidents mimic state behavior, adopting violent dissent in response to increasing violent repression and nonviolent dissent in the face of nonviolent repression.This tit-for-tat reasoning is based on the principle of reciprocity, and has been associated with reciprocal altruism among humans (Axelrod 1981) as well as animals (Axelrod and Hamilton 1981;Wilkinson 1984).In contrast, this article shows that tit-for-tat is also related to conflictual, non-altruistic behavior.
Tit-for-tat reasoning is known as a rational strategy which can surpass the benefits of most other strategies in competitive games (Axelrod 1981;Rapoport 2012).The identified tit-for-tat deliberations, however, are more akin to heuristic and intuitive cognitive processes.Means-driven, they often skip goal-oriented reasoning, are emotionally-laden, and include features that contradict a chosen behavior.Several nonviolent dissidents considered their peaceful activity to be unsuccessful, or interacted closely with violent groups.Some violent individuals hated violence, or believed they had the opportunity to pursue a successful career inside the state system.There is also evidence that nonviolent dissidents apply the "imitate the majority" heuristic, deciding to oppose their repressive rulers based on observing others resist.
Heuristic and emotionally-laden reasoning involves a fast cognitive "go-system", whereas rational-choice deliberations rely on the slow cognitive "know-system" (Kahneman 2011;Metcalfe and Mischel 1999).Dissident deliberations show similarities to the fast go-system, mimicking the means applied by the state.This suggests that states willing to engage in nonviolent behavior are likely to be met with similar, nonviolent dissident behavior and reduced levels of violence.When reflecting on the decision to assassinate the Egyptian President Sadat, the founder of al-Jama'at al-Islamiyya remembered the violent persecution of a thousand Islamists.Had the state not engaged in this behavior, he said, Sadat would not have been dead.engage in both violent and nonviolent dissent.In contrast, this article focuses primarily on the substantial cognitive differences between decisions to engage in violent versus nonviolent dissent, based on an analysis of the old data and an additional new sample of 121 individuals from Egypt and Morocco.The new sample, introduced in the following section, includes participants in nonviolent dissent and ordinary people refraining from participation in dissent.The analysis shows, for the first time, that violent dissent is linked to violent repression, whereas nonviolent dissent is based on nonviolent repression.This identifies tit-for-tat reasoning akin to known cognitive processes usually associated with cooperative behavior (Axelrod, 1981;Nowak and Sigmund 1992).9.These inferences are called conjunctions and disjunctions respectively (Dujmović and Larsen 2007).10. 102 were constructed from interviews, and 19 from Facebook posts.The online supplemental materials provide more information.11.Numbers represent the count of cognitive maps that contain a particular belief.12. Names were assigned based on in vivo codes reflecting the interviewees' own words (Strauss and Corbin 1990).13. "Peaceful activity against the state" is differentiated from beliefs addressing nonviolent dissent by not being preceded by beliefs that address the possibility of violent dissent.14.Names were assigned based on in vivo codes, which is why the names of Dataset 1 and Dataset 2 differ.Interviewees referred to crime (jarima) rather than domination (saytara) when discussing nonviolent forms of repression, and to violence (`unf) rather than attack (hujum) when discussing state violence.15.Based on these inferences, other beliefs could trigger multiple key antecedents of dissident decisions, bypass the beliefs about repression through cognitive shortcuts, or trigger key antecedents of the beliefs to state repression.16.Identifying the experimental set poses a challenge because of the large number of beliefs and inferences contained by the data, and the idiosyncratic nature of the individual maps.Given this data structure, large numbers of beliefs activate only small numbers of decisions, and it is not immediately obvious which beliefs activate all decisions contained by the two datasets.We identified the experimental set by examining beliefs with high frequencies, finding that both datasets contain 17 beliefs that together activate decisions in each cognitive map.17.Poverty is captured by a belief called "poverty," contained by Datasets 2. Social influence is captured by four beliefs addressing interactions with protestors, observations of protest behavior by others, observations of successful protest abroad, and external influence from foreign countries, contained by Dataset 2. State behavior violating religious norms is addressed by two beliefs called "unreligious government" and "religious government," contained by Dataset 1. 18.As shown by Table 1 and the online supplemental materials, the removal of "education of the people" does not change the decision distribution of violent dissidents (Δμ = 0), but has a minor impact on the decision distribution of nonviolent dissidents (Δμ = 1.13).19.This type of reasoning is observed in each map of violent dissidents.The online supplemental materials provide plots that show the belief counts related to violent versus nonviolent dissidents.

Figure 3 .
Figure 3. Decision counts of violent and nonviolent dissidents with (grey) and without (black) beliefs addressing state repression (Data 1).

Table 1 .
Average Decision Counts of Beliefs Examined in the Counterfactual Experiment.The Averages Before Removal Correspond to the Observed Cognitive Maps; the Averages After Removal are Based on the Counterfactual Alterations of the Observed Cognitive Maps.The Table Includes Beliefs Contained by Combinations Whose Removal Maximally Reduced Decision Counts, and Beliefs On Religious Governance (Data