The Effect of Curfews on Political Preferences

Abstract

Governments often use curfews as counterinsurgency measures. Do such state actions affect citizens’ political preferences? We argue that citizens’ responses depend on their group alignment. Citizens who are aligned with the government are more likely to interpret state actions as serving their interests and reward the governing parties as a result. Citizens who are aligned with the insurgent movement, however, are likely to interpret these same actions as targeted and disengage from politics: they have no reason to reward the governing parties and may also be fearful of expressing support for the parties affiliated with the insurgent movement. We find support for this argument with survey data from Turkey, fielded before and during curfews. When exposed to curfews, Turks generally increase their support for the governing party. However Kurds do not increase their support for the governing party or the opposition; instead, we present strong evidence that they withdraw their support from the Kurdish opposition and become more hesitant to express any political preferences.

Keywords

counterinsurgency counterterrorism elections violence

Governments often use indiscriminate or collective repression as a counterinsurgency tactic (Davenport 2007; Longo, Canetti and Hite-Rubin 2014; Hultquist 2017; Hatz 2018). As opposed to selective counterinsurgency actions, which target militants and their supporters specifically, collective repression does not discriminate between militants and civilians. Instead, it is often implemented based on geographic location, affecting everyone in those areas, regardless of whether they are combatants or not (Kalyvas 2006; Lyall 2009; Duffy Toft and Zhukov 2015). A large body of literature on political violence studies the military consequences of such indiscriminate actions on subsequent patterns of militant attacks (Benmelech, Berrebi and Klor 2014; Condra and Shapiro 2012; Enders and Sandler 1993; Kocher, Pepinsky and Kalyvas 2011; Lyall 2009; Zussman and Zussman 2006). However, we know relatively little about the political consequences of these government actions. Do indiscriminate state counterinsurgency actions shape citizens’ political preferences? Are these state actions politically costly or beneficial to the governing parties? Do they affect support for the opposition parties associated with the insurgent movement?

Answering these questions is important because prior work sees civilian attitudes and behavior as the main mechanism through which state actions lead to alternative military outcomes. That is, both militant groups and governments rely on civilian support and cooperation to succeed in outperforming one another (Balcells 2011; Berman, Shapiro and Felter 2011; Berman, Felter and Shapiro 2018; Carter 2016; Condra and Shapiro 2012; Cunningham, Gleditsch and Saleyhan 2013; Humphreys and Weinstein 2006; Kalyvas 2006; Kocher, Pepinsky and Kalyvas 2011; Thompson 1966; Tse-tung 1961). Accordingly, both actors have an interest in shaping citizens’ attitudes and behavior, and military operations are seen as a way to do so (Gambetta 1993; Kalyvas 2006). However, despite their relevance, citizens’ attitudes are difficult to study directly due to challenges of collecting survey data in conflict settings.¹ Furthermore, existing research provides conflicting expectations about the effects of indiscriminate state actions on civilian attitudes: some studies suggest that they increase civilian support for the militants at the expense of the government (Bueno de Mesquita and Dickson 2006; Kalyvas 2006; Kocher, Pepinsky and Kalyvas 2011; Rosendorff and Sandler 2004; Wood 2003), while others claim the opposite (Downes 2008; Lyall 2009; Stoll 1993).

We study individual level political attitudes directly, using monthly survey data gathered during counterinsurgency operations. We focus on the imposition of curfews as a common, important, and potentially severe form of collective repression and indiscriminate counterinsurgency tool (Brass 2006). Curfews implemented by geographic location are indiscriminate in that they apply equally to militants and civilians. Violations can be punishable by fine, imprisonment, or (in the extreme) death (Brass 2006). This type of collective repression can significantly affect the well-being of civilians in the curfew-bound areas. For example, in the case of curfews implemented by the Turkish government in response to Kurdish militant activities between 2015 and 2019, human rights organizations documented violations of the rights to life, liberty, health, education, freedom of assembly, etc. The curfews severely restricted civilian movement, and operations conducted during curfews caused significant damage; reports indicate that civilians were trapped without food or basic necessities and not allowed to seek necessary medical help.² In short, in countries where the government is in conflict with armed groups, repressive and indiscriminate counterinsurgency measures such as curfews significantly influence civilians’ lives in curfew-bound areas. In this paper we study curfews to understand the political consequences of such indiscriminate counterinsurgency actions.

Focusing on curfews, we develop a novel argument by theorizing that, despite being collectively affected, citizens’ response to indiscriminate state actions is unlikely to be uniform (Lyall, Blair and Imai 2013; see also Hatz 2018). Rather, it is likely to depend on whether they are ethnically aligned with the governing party or the insurgent movement. Citizens who are aligned with the government are more likely to interpret state actions as serving their interests by fighting militant violence and reward the governing parties as a result. Citizens who are misaligned with the government and aligned with the insurgent movement, however, are more likely to interpret these same actions as discriminatory against their group. This gives them no reason to increase their support for the government following exposure to indiscriminate counterinsurgency operations. However, they may also be reluctant to express support for the political forces associated with the insurgency out of fear of further retribution, and are therefore more likely to disengage from politics altogether.

We test the observable implications of this argument with individual-level attitudinal data and election results data from Turkey. There are several reasons for why this is a suitable research site for our purposes. First, armed conflict with militants is a long-standing and salient issue in Turkey. The conflict involves one of the most active militant organizations in the world – the Partiya Karkerên Kurdistan or PKK (Lafree 2010) – and security is a significant concern for citizens, with about 70% of them being worried about terrorism.³ Furthermore, the Turkish government has been engaged in extensive counterinsurgency operations since the mid 1980s. In short, Turkey offers a context where conflict with militants is a long-standing, salient, and central part of political reality. Second, and equally importantly, within the period under study, Turkey held competitive elections for executive recruitment,⁴ which makes studying citizens’ political preferences meaningful because these attitudes can affect electoral outcomes. Finally, unlike most conflict settings, Turkey offers unique and rich data for studying the attitudinal effects of indiscriminate state actions. Between August 2015 and January 2019 the government implemented over 350 curfews in civilian populated areas as a counterinsurgency measure. Since the curfews were implemented at different times and only in some subnational units, we can exploit the geographical and temporal variation in curfew exposure to study its effects. We also have access to unique and detailed monthly surveys which recorded respondents’ geographic location, ethnic affiliation, and political preferences, providing us with appropriate individual-level data to study how exposure to curfews influences respondents with different group affiliations.

Applying our theoretical argument to the Turkish case, we expect and show that ethnic Turks and ethnic Kurds affected by curfews react very differently to them. Turks, who are mostly ethnically aligned with the governing party and more likely than Kurds to be ideologically aligned with the governing party, can interpret this government counterinsurgency measure as a legitimate way to fight militant violence. While curfews may have negative consequences for Turks in curfew-bound areas, curfews are also a powerful and observable daily reminder that the incumbent is trying to maintain order. Because of this, they are likely to increase their support for the governing Justice and Development Party (AKP). As an out-group in Turkey, Kurds are ethnically aligned with the insurgent movement and ideologically less likely to be aligned with the government. This out-group status combined with prior experience with discrimination makes them more likely to see the government actions in a much different light: as discriminatory targeting of the Kurds. This gives them little reason to increase their support for the government. Fear of further retribution is also likely to coerce them to withdraw their support from opposition political parties with ethnic connections to the insurgent movement and not express any partisan preferences at all. Our empirical analysis largely confirms these expectations.

Our study makes important contributions to the literature. First, unlike prior work, we focus specifically on political and attitudinal rather than military consequences of indiscriminate state counterinsurgency actions, and show that these actions affect political preferences to a significant degree. Second, we theorize and show that the effects of state actions are not uniform but depend on whether an affected individual is aligned with the government. Third, we study a previously understudied indiscriminate counterinsurgency measure: curfews. While the majority of the existing studies focus on lethal violence or direct killing of civilians, states and non-state armed actors have a repertoire of forms of violence and repression at their disposal, such as torture, suicide bombings, forced displacement of populations (Gutierrez-Sanin and Wood 2007). Imposition of curfews is another understudied yet common and potentially severe form of collective repression and indiscriminate counterinsurgency tool (Brass 2006).

The Consequences of State Actions: Prior Work

Most of the existing work on the consequences of state counterinsurgency actions examines how state actions influence the volume and intensity of subsequent militant attacks (Berman, Shapiro and Felter 2011; Condra and Shapiro 2012; Enders and Sandler 1993; Kalyvas 2006; Kocher, Pepinsky and Kalyvas 2011; Lyall 2009). However, this literature sees civilian attitudes and behavior as the main mechanism through which state actions lead to alternative military outcomes. It therefore offers a potential starting point for theorizing about the effect of state actions on citizens’ political preferences. The arguments presented in the literature lead to conflicting expectations about civilian behavior and attitudes (Davenport 2007, 8).

The predominant argument in this literature is that indiscriminate state countermeasures against militant groups, which lead to civilian suffering, are often counterproductive in reducing militant violence (Lichbach 1994; Mason 2004; Bueno de Mesquita and Dickson 2006; Kalyvas 2006; Findley and Young 2007; Bennett 2008; Kocher, Pepinsky and Kalyvas 2011). Recent empirical work from Israel, Iraq, and Turkey confirms this expectation (Benmelech, Berrebi and Klor 2014; Condra and Shapiro 2012; Tezcür 2016b). According to this argument, the negative military consequences result from changing civilian attitudes and behavior. State violence and repression leaves citizens with an impression that the state is insensitive to the negative effects of its actions on civilians (Kalyvas 2006, 153–155). This, in turn, can make it rational for civilians to support the insurgents to seek protection (Lichbach 1994; Mason 2004; Kalyvas 2006; Bueno de Mesquita and Dickson 2006) or generate powerful emotional reactions (e.g. revenge, resentment, outrage) against the state that lead to identification with the insurgents (Wood 2003; Tishkov 2004; Tezcür 2016b; Parkinson 2013; Viterna 2013). In terms of political preferences, this line of argument implies that citizens are more likely to decrease their support for the governments that engage in indiscriminate counterinsurgency actions, and increase their support of the opposition parties with links to insurgent movements.

A second set of studies, however, challenges the idea that indiscriminate state actions are ineffective (Downes 2007 2008; Lyall 2009; Stoll 1993). Instead, Stoll (1993) shows that indiscriminate reprisals by Guatemalan military forces helped the government in its fight against the leftist guerrillas. Lyall (2009) finds that Russian artillery strikes on civilian populated localities in Chechnya effectively reduced insurgent attacks. These studies suggest a very different citizen response by arguing that indiscriminate state actions can “drive a wedge between the civilians and insurgents” because civilians start blaming insurgents for instability in their locality (Lyall 2009, 337). That is, rather than feeling resentment toward the government, according to this argument, citizens feel resentment toward the militants for having to suffer the consequences of government actions. Citizens are then less likely to support the militants, who are seen as the cause of their suffering, and potentially reward the government for attempting to eliminate that cause.

Note that similar effects can result from a different mechanism: rather than supporting the government (and opposing militants) out of gratitude, civilians may be compelled to do so out of fear. That is, they can show support for the government despite the suffering that its actions caused (and withdraw support from the militants) because by doing so they hope to prevent further harm from the state (Zhukov 2013). This does not require shifting the blame for their suffering from the government to the militants. Rather, it suggests that citizens can be threatened into supporting the perpetrator, i.e. that indiscriminate government actions can coerce and intimidate civilians into compliance and loyalty (Downes 2008; Lyall 2009; Stoll 1993).

In terms of political preferences, these arguments imply that indiscriminate counterinsurgency actions by the state should increase citizens’ political support for the government and decrease their support for the political forces affiliated with the insurgency. This effect can occur for one of the following reasons: (1) because citizens sincerely believe that insurgents are the culprits and the state acts as a protector, or (2) because citizens are intimidated and coerced to support the governing parties and withdraw support from the insurgent movement out of fear and a desire to stop further suffering from the state’s indiscriminate actions.

In sum, existing work on the effects of indiscriminate counterinsurgency actions offers conflicting expectations and several arguments about how state actions might affect citizens’ political preferences. We see this as an opportunity for further theorizing. We start from a basic premise that, rather than having a uniform response from all citizens, different groups of citizens may interpret and respond to government actions differently (Condra and Shapiro 2012; Hatz 2018; Lyall, Blair and Imai 2013), with some of the above arguments applying to some groups but not others. We then borrow from social identity theory and related literature to flesh out our argument about the political effects of indiscriminate state actions.

Theory and Expectations

Indiscriminate counterinsurgency measures do not distinguish between civilians and militants and thus can put civilians in harm’s way. This is certainly the case with curfews, which are imposed on everyone in a curfew-bound locality. Prior work on state violence and repression often assumes that affected civilians respond to such indiscriminate state actions uniformly (Kalyvas 2006; Berman, Shapiro and Felter 2011; Condra and Shapiro 2012; Stoll 1993). Recent work in Afghanistan, however, draws attention to group-based biases in response to violence: Lyall, Blair and Imai (2013) emphasize that individuals’ inclination to view the actions of one’s in-group members more favorably than those of out-group members plays an important role in determining civilian response. This study shows that civilians punish an out-group perpetrator for harm caused to civilians, yet they do not punish an in-group perpetrator for causing similar harm.

We build on this insight and argue that group affiliation conditions how individuals interpret state actions and perceive the government’s intentions. It is this interpretation, rather than the indiscriminate nature of the action itself, that determines their subsequent response (see also Bueno de Mesquita and Dickson 2006; Hatz 2018; Kalyvas 2006). Group affiliation matters because actions by in-group members are interpreted differently than those by out-group members (Lyall, Blair and Imai 2013; Tajfel and Turner 1979). A key finding of the social identity theory (SIT) is that group members are motivated to affirm and protect the positive distinction of their in-group (Hewstone, Rubin and Willis 2002; Tajfel and Turner 1979). Because of this, they are likely to see even negative actions by in-group members in a positive light. We argue that this favorability bias leads individuals who are aligned with the government (in-group members) to interpret the potentially harmful government actions – such as indiscriminate counterinsurgency measures – not in a derogatory manner, but through the lens of in-group favoritism. Thus, even though their lives may be negatively influenced by government actions, individuals aligned with the government are more likely to believe that these actions are necessary to curb violence. Instead of attributing blame for potential harm, they are more likely to assign credit to the government for efforts to increase security. In terms of political preferences, literature on retrospective performance-based voting (Healy and Malhotra 2013; Lewis-Beck and Stegmaier 2000) strongly suggests that such an interpretation should lead individuals to increase their support for the government. Moreover, for in-group members, the government’s counterinsurgency measures serve as a powerful and observable signal of government’s resolve to maintain order. This signal is strongest in curfew-bound areas where civilians regularly observe government forces and their operations. Thus, curfew exposure should increase the likelihood that in-group members support the government. To sum, we expect individuals from the government’s in-group who experience indiscriminate counterinsurgency measures to be more likely to express political support for the governing parties than in-group members who did not experience such measures. In-group status makes them likely to interpret government actions in a positive light, and exposure to these measures clearly show them the government’s resolve to maintain order.

Hypothesis 1. Individuals from the government’s in-group who experience indiscriminate counterinsurgency measures are more likely to express political support for the governing parties than in-group members who did not experience such measures.

Individuals whose identity is aligned with the insurgent movement (out-group members), however, are likely to interpret the same state actions differently. For them, government is an out-group. According to SIT, that alone provides no reason to interpret potentially harmful actions by the state in a positive light. If anything, the bias runs in the opposite direction and such actions may be interpreted as malicious attacks (Brewer 1999; Hewstone, Rubin and Willis 2002; Lyall, Blair and Imai 2013). Several arguments lead to this expectation. First, SIT suggests that, in addition to affecting beliefs about the in-group, social identities also impact perceptions of out-groups, who are usually viewed unfavorably, with prejudice and vilification. Negative interpretation of potentially harmful behavior is in line with this desire to denigrate the out-group (Brewer 1999). Second, beyond SIT, literature on race relations in the U.S. suggests that the state’s past actions toward the individual or her group affect the interpretation of its current actions. According to this literature, blacks in the U.S., who have historically faced discrimination by white-dominated law enforcement agencies, are more likely than whites to see police actions as unfair and discriminatory (Gibson and Nelson 2018; Hurwitz and Peffley 2005; Johnson and Kuhns 2009). In more general terms, this implies that groups that have faced discrimination by an out-group dominated government are skeptical of state’s actions. Rather than seeing these actions as legitimately curbing militant violence, these individuals are more likely to see them as part of a pattern of discrimination.

How does such interpretation of state actions affect political preferences? On the one hand, due to the negative bias described above, one might expect that individuals who are aligned with insurgent movement are likely to punish the governing parties for what they see as discriminatory, and therefore inferior, performance. One might also expect that these individuals heighten their support for the opposition parties associated with the insurgent movement because, according to SIT, the threatening behavior by the out-group should strengthen their in-group bias (Brewer 1999).

However, the considerations of the individuals who are aligned with insurgent movement might be more complex than the literature on inter-group bias suggests. Even if identity-based forces push these individuals to sympathize with political forces of the insurgency, strategic considerations may prevent them from expressing these preferences. That is, if individuals believe that they are targeted because of the alignment of their identity with insurgents, they may choose to strategically downplay and withdraw political support from opposition political parties with links to insurgent movements. Fearful individuals may see this as a way to escape further retributions by the state. Note that this argument is in line with the second set of studies reviewed above, predicting that a state’s indiscriminate actions may coerce its population into loyalty.

This discussion generates several hypotheses about the political responses of affected out-group members. First, because of inter-group bias and past discrimination, out-group members should not be more likely to express political support for the government as a result of indiscriminate counterinsurgency actions.

Hypothesis 2. Out-group members who experience indiscriminate counterinsurgency measures are not more likely to express political support for governing parties than out-group members who did not experience such measures.

Second, in the hopes of preventing further government actions, these individuals should also not be more likely to express political support for their in-group political actors, i.e. political parties associated with the insurgent movement. If anything, they should be less likely to express support for such parties because this distancing serves as a stronger signal of their disassociation from the movement.

Hypothesis 3. Out-group members who experience indiscriminate counterinsurgency measures are less likely to express political support for insurgent-associated parties than out-group members who did not experience such measures.

Note that, combined, Hypotheses 2 and 3 suggest that these individuals may be reluctant to express any partisan preferences at all: they may see expressing political support for their in-group parties as costly because it might trigger further government action, yet they have no reason to express political support for the governing parties either. As a result, these individuals should be more likely to express ambivalence, effectively disengaging from politics altogether.

Hypothesis 4. Out-group members who experience indiscriminate counterinsurgency measures are more likely to express political ambivalence than out-group members who did not experience such measures.

Research Design

We test our hypotheses with individual-level attitudinal data from Turkey. We explained in the introduction why Turkey is an appropriate case for our purposes. In this section, we provide some more context to the conflict in Turkey and describe our data.

Background on Turkey

The Turkish government has been in armed conflict against Kurdish militants mainly organized under the Partiya Karkerên Kurdistan (PKK) for over 35 years. The group was established in 1978 and conducted its first attack in August 1984 with the goal of establishing an independent Kurdistan for ethnic Kurds, the largest ethnic minority in the country (Kıbrıs 2014). Historically, the deadliest period of the conflict was in 1990s with number of casualties reaching up to 30,000 by the end of the decade (Tezcür 2010; McDowall 2000). The capture of the PKK leader Abdullah Öcalan in 1999 temporarily led to a slow-down in the group’s operations for the next 5 years (Marcus 2007; Tezcür 2016b; Tezcür and Gürses 2017). After 2005, the conflict was reignited until the beginning of secret negotiations with the group’s leadership in early 2013. Around the same time, a tentative truce was observed and fatalities decreased to lowest levels since the early 2000s (Tezcür 2016a, 67). With the end of the ceasefire in July 2015 political violence became the single most important issue in the country (Sözen 2016, 201).

The conflict following the end of the ceasefire is considered to be the most violent episode since the 1990s (Aktürk 2016, 59). August 2015 marked the beginning of increased militant activity and intensive government operations in the southeast. Extended, around the clock curfews became an important part of the government operations (Aydin and Emrence 2016).⁵ The curfews restricted civilians’ freedom of movement, and the clashes between the government forces and militants in localities under curfews caused serious material damage and civilian suffering.⁶

Since the curfews were implemented at different times and only in some subnational units, we exploit the geographical and temporal variation in curfew exposure to study their effects. We combine the information on the timing and location of the curfews with detailed monthly surveys, fielded before and during the curfews with information on respondents’ location, ethnic affiliation, and political preferences. This provides us with appropriate individual-level data to study how exposure to curfews influences respondents with different group affiliations.

Data and Variables

We rely on the nationally representative KONDA Barometer surveys conducted face-to-face every month by a prominent independent research company in Turkey.⁷ The surveys were fielded between March 2010-October 2017 with every wave including around 2500 respondents.⁸

Since conflict is regionally concentrated, we restrict our dataset to provinces with significant Kurdish populations.⁹ Figure 1 illustrates the location of the provinces and the geographical distributions of curfews across the districts within the provinces. Each area in the map is a district, and the shaded districts are within the sample of provinces we study. Dark shaded districts experienced curfews while light shaded districts did not. After restricting our data to provinces with sizable Kurdish populations we have 37,601 total respondents.¹⁰ The surveys include the province and district names in which each respondent resides. This helps us determine whether a respondent resides in a district that experienced a curfew and whether they took the survey before or after the curfew. We describe our variables below and provide further details in the online appendix (OA) section 1.

Figure 1.

Geographic Distribution of Curfews Across Districts. Notes: Light and dark shaded districts are located within the provinces with significant Kurdish populations: Adıyaman, Ağrı, Ardahan, Batman, Bingöl, Bitlis, Diyarbakır, Elazığ, Erzincan, Erzurum, Gaziantep, Hakkâri, Iğdır, Kahramanmaraş, Kars, Malatya, Mardin, Muş, Şanlıurfa, Siirt, Şırnak, Sivas, Tunceli, and Van. Light shaded districts did not experience curfews, while dark shaded districts experienced curfews.

Our dependent variable, measuring the respondents’ political preferences, is based on the following survey question: “If elections were held today, which party would you vote for?” We generate a categorical dependent variable (Vote) with four categories: Vote AKP, Vote HDP, Abstain, and Vote Other. Vote Other (coded 1) serves as the baseline category and identifies respondents who would vote for any opposition party except the Kurdish opposition party, HDP. Vote AKP (coded 2) identifies the respondents who would vote for the governing party and serves as the outcome category for testing Hypotheses 1 and 2. Vote HDP (coded 3) identifies the respondents who would vote for the Kurdish opposition party, HDP, and serves as the outcome category of interest for Hypothesis 3. Abstain (coded 4) identifies respondents who (a) reported an intention to abstain from voting, (b) stated that they are undecided, or (c) refused to answer the vote choice question.¹¹ This category is used for testing Hypothesis 4.

Our main independent variables measure respondents’ curfew exposure based on the information on the location and timing of curfews collected by the Human Rights Foundation of Turkey (TIHV 2017). The curfews were implemented in 45 districts within 11 provinces.¹² Exactly 4500 of our survey respondents live in one of these 45 districts, and roughly 26% of them took the survey after experiencing a curfew.¹³ We measure curfew exposure in two ways. First, we code a dummy variable for curfew exposure (Curfew), coded 1 if government implemented at least one curfew in a respondent’s district prior to the month in which the respondent took the survey. Second, we measure the duration of the curfews each respondent experienced based on the total number of hours under curfew in the respondent’s district (Curfew Duration).

To test whether curfews affected respondents differently depending on their group affiliations, we record the ethnicity of each respondent (Kurdish), coded 1 for respondents who self-identified as Kurdish and 0 otherwise. Throughout our analysis, we work with the assumption that Kurdish respondents are less likely than Turkish respondents to be ethnically and ideologically aligned with the government. While ethic affiliation is not a perfect indicator of partisan preferences in Turkey, this is a reasonable assumption supported by descriptive patterns in our dataset. In the provinces we study, around 43% of the respondents identify as Kurdish. Out of the Kurdish respondents around 34% express that they have previously voted for the AKP; similarly around 29% mention that they would vote for AKP in the upcoming election. However, out of the Turkish respondents around 59% express that they have previously voted for AKP, and around 54% express intention to vote for the AKP. Thus, the assumption that Turkish respondents are more likely to be aligned with the government than Kurdish respondents is reasonable. To uncover the effect of curfews for Turkish and Kurdish respondents, we include interaction terms between our primary independent variables, (Curfew, Curfew Duration) and the ethnic identification variable (Kurdish) in our models.

We control for several variables that may influence partisan preferences. At the individual level, we control for respondents’ gender (Male), Age, Education, Religiosity, and Income (Esmer 2002; Çarkoğlu and Toprak 2000).¹⁴ At the district level, we control for socioeconomic conditions: Unemployment rate, Population (logged), Literacy rate, and High School completion rate (Livny 2020).¹⁵ While available district level variables are time invariant, they are relatively good proxies for the socioeconomic conditions in a respondent’s district at the time they took the survey since district-level socioeconomic conditions are unlikely to fluctuate during the relatively short period under study.

Finally, we also control for the volume and severity of militant attacks in a respondent’s district. The curfews were ostensibly adopted in response to militant activity. Therefore, any relationship between curfew exposure and political attitudes may be due to militant activity. Moreover, militant attacks can influence support for the governing parties (Bali 2003; Berrebi and Klor 2006; Gassebner, Jong-A-Pin and Mierau 2008) as well as the insurgent-affiliated ones (Birnir and Gohdes 2018; de la Calle and Sánchez-Cuenca 2013). We obtained information on the number of militant attacks and the number of killed and wounded individuals for each district from the Global Terrorism Dataset (GTD) (START 2008).¹⁶ Based on this information, we coded the number of militant Attacks in each respondent’s district in the last 6 months and the number of people Killed and Wounded in these attacks. Additional models in OA 3.2 include alternative violence measures from the Turkish State-PKK Conflict Event Database (Kibris 2021).

Once observations with missing data are excluded, our final dataset contains information on 24,705 respondents between March 2010 and October 2017.¹⁷ We consider it unlikely that curfews were implemented randomly. The government’s stated reason for curfews was to allow for targeted counterinsurgency operations in specific areas. Furthermore, the government may have been motivated by non-military concerns, such as to influence HDP support, making the selection issue relevant. Thus, it is important to compare districts that experienced curfews to those that did not, and to use statistical techniques to account for possible confounding variables due to selection bias. In Table 1, we present mean values, standard deviations, and numbers of observations of our variables for districts with and without curfews. We also present statistics from cluster-adjusted t-tests to examine the statistical significance of differences in means between curfew and no-curfew districts.

Table 1.

Descriptive statistics for district curfew selection.

	Districts selectedfor curfew			Districts notselected			Differencein means
Variable	Mean	S.D.	Obs.	Mean	S.D.	Obs.	Diff.	p-value*
Vote AKP	0.212	0.409	4500	0.478	0.500	33,085	−0.266	<0.001
Vote HDP	0.535	0.499	4500	0.159	0.366	33,085	+0.376	<0.001
Abstain	0.207	0.405	4500	0.195	0.396	33,085	+0.012	=0.531
Kurdish	0.817	0.386	4487	0.383	0.486	32,925	+0.435	<0.001
Male	0.510	0.500	4496	0.518	0.500	33,051	−0.008	=0.465
Age	37.67	13.76	3639	38.85	14.48	26,357	−1.18	=0.067
Education	3.32	1.67	4485	3.62	1.52	32,948	−0.30	=0.030
Religiosity	2.84	0.69	4483	2.87	0.65	32,934	−0.03	=0.488
Income	3.15	1.29	3452	3.52	1.25	25,075	−0.37	=0.030
Population^a	11.70	0.93	4500	11.89	1.33	32,625	−0.19	=0.705
Unemployed^a	18.15	6.47	4500	14.32	6.53	32,734	+3.84	=0.136
Literacy^a	90.25	2.45	4500	91.76	3.35	29,446	−1.51	=0.240
High school^a	20.66	7.83	4500	22.27	9.68	29,446	−1.61	=0.667
Attacks^b	1.22	2.72	4500	0.26	0.82	33,085	+0.96	<0.001
Killed^b	1.19	3.13	4500	0.47	4.10	33,085	+0.72	=0.112
Wounded^b	6.92	22.00	4500	1.35	9.00	33,085	+5.57	=0.001

Notes: *Difference in means statistics are from a variance cluster-adjusted t-test using 136 district clusters (Herrin 2012; Donner and Klar 2000).

^aMeasures from Livny SES data. Population is logged; others are percentages.

^bMeasures from GTD violence dataset aggregated for the previous 6 months prior to the survey response.

Table 1 shows that the districts that were targeted for curfews had several characteristics that differentiate them from districts that never experienced curfews. Targeted districts had lower AKP support and higher HDP support than non-targeted districts. This is largely a function of their ethnic composition; targeted districts have a higher proportion of Kurdish residents. There are also some socio-economic differences: targeted districts have slightly lower average education and income levels. Finally, targeted districts experienced more militant attacks.

The statistically significant differences between curfew-exposed and curfew-free districts raise the concern of confounding through backdoor pathways between a respondent experiencing a curfew and their political preferences due to selection effects. These effects are a problem for regression models if the factors affecting selection into the treatment group (in our case, experiencing a curfew) are correlated with the dependent variables. We implement several strategies to reduce the possibility of bias from selection effects. First, we include several covariates that can present backdoor pathways between curfew selection and political outcomes, including district level violence and socio-economic measures, as well as individual characteristics. Second, we use regression models that include random and fixed effects for both the time of the survey and for respondents’ district. These fixed and random effect specifications help us control for potential unmeasured differences on covariates between respondents that vary by time or by geography, including factors that could have led to districts being selected for curfews, such as district-level ethnic composition. Using all the surveys between 2010 and 2017 allows us to have enough respondents from the Kurdish-populated provinces over a long period of time to compare their responses before and after curfew exposure while controlling for unmeasured factors that vary by time or geography. Third, we conduct sensitivity analyses (see OA5) to examine how large the impact of unmeasured confounders needs to be to invalidate our inferences. Based on these, we are confident that our results are robust: for almost all of our findings, a potential confounding variable would have to be as influential as Kurdish identity is on curfew exposure and voting behavior to invalidate our findings.

Results

Our data have a hierarchical structure with survey respondents located in districts nested within provinces, and we have a categorical dependent variable. Thus, we employ multi-level multinomial logistic models.¹⁸ To account for the geographical clustering of observations and to control for potential district-level unmeasured confounders, we include random intercepts for districts. To account for time-relevant factors that might influence political preferences across all respondents within each year, we include survey year dummy variables.¹⁹

We run two sets of models to examine the relationship between curfew exposure and political preferences. In the first set of models we use the binary curfew exposure variable (Curfew) and an interaction term between Curfew and Kurdish. The interaction term helps us determine whether curfews affect Kurdish and Turkish respondents differently. In a second set of models we use curfew duration variable (Curfew Duration) as the main independent variable, and include a squared measure of curfew duration (Duration Squared) and interaction terms between Kurdish and Curfew Duration and Duration Squared. The squared terms allow the model to estimate the impact of duration in a non-linear way so that the effect of additional hours can be stronger or weaker as the total hours accumulate. The interaction terms allow the effect of the curfew duration to differ by ethnicity.

In Table 2, we present the results from our first set of models. Recall that our dependent variable Vote takes four values between 1 and 4. Vote Other is the baseline category, and the coefficients for the other three categories represent changes in probability relative to baseline. Because the coefficients for each category in a multinomial logistic model do not directly represent how variables affect the probability of the category of interest relative to all other categories, the statistical significance, direction, and magnitude of effects cannot be interpreted directly from Table 2. Thus, we graphically illustrate the substantive effects of Curfew in Figure 2. The graphs display point estimates for changes in predicted probabilities if we change Curfew from 0 to 1, and the corresponding 95% confidence intervals for each ethnic group. All control variables are held at their mean values rounded to the nearest unit. We also calculate the difference between these substantive effects, with the effect for Kurds subtracted from the effect for Turks. Online appendix OA 2.2 displays these differences graphically along with their 95% confidence intervals.²⁰

Table 2.

Effect of curfews on partisan preferences.

Variables	Vote AKP	Vote HDP	Abstain
Curfew	0.82**	0.36	0.56
	(0.38)	(0.43)	(0.38)
Kurdish	0.60***	3.14***	0.91***
	(0.07)	(0.09)	(0.07)
Curfew * Kurdish	−0.44	−0.09	0.71
	(0.50)	(0.53)	(0.50)
Male	−0.18***	−0.03	−0.48***
	(0.04)	(0.06)	(0.05)
Age	−0.01***	−0.01***	−0.01***
	(0.00)	(0.00)	(0.00)
Education	−0.25***	−0.13***	−0.08***
	(0.02)	(0.02)	(0.02)
Religiosity	0.86***	0.13***	0.39***
	(0.03)	(0.05)	(0.04)
Income	−0.00	−0.09***	−0.07***
	(0.02)	(0.03)	(0.02)
High school	−0.05***	−0.09***	−0.04***
	(0.01)	(0.02)	(0.01)
Unemployment	0.02	0.08***	0.05***
	(0.02)	(0.03)	(0.02)
Literacy	−0.00	−0.14*	−0.06
	(0.04)	(0.07)	(0.04)
Population	0.26**	0.26	0.14
	(0.11)	(0.21)	(0.12)
Attacks	0.09**	0.13***	0.14***
	(0.04)	(0.05)	(0.05)
Killed	−0.02**	−0.02	−0.01*
	(0.01)	(0.01)	(0.01)
Wounded	−0.00	0.00	0.01
	(0.00)	(0.01)	(0.00)
Constant	−1.43	9.23	4.67
	(3.24)	(6.10)	(3.46)
N: Respondents	24,877	24,877	24,877
District random effects	Yes	Yes	Yes
Year dummy variables	Yes	Yes	Yes

Notes: Table entries are unstandardized multinomial logistic regression coefficients with standard errors in parentheses. The baseline category is Vote Other.

***p $<$ 0.01, **p $<$ 0.05, *p $<$ 0.1.

Figure 2.

Effect of curfews on partisan preferences.

Figure 2(a) shows the change in predicted probability of voting for the governing party (Vote AKP) when we change the Curfew variable from 0 to 1. The probability that a Turkish respondent expresses an intention to vote for the governing party increases by 11% after curfew exposure. However, curfew exposure does not have a statistically significant impact on the Kurdish respondents, as the change in probability of voting for the governing party is negative, but close to zero and the confidence intervals cross the zero line. The −16% difference between these substantive effects is statistically significant (Figure OA 2.2).

These results provide support for Hypotheses 1 and 2. The former suggested that support for the governing party should increase among Turks exposed to curfews since they are more likely to be aligned with the government and more likely to perceive counterinsurgency operations as increasing security. For Turks, this is exactly what we see: the effect of curfews is estimated with a large and positive point estimate. Hypothesis 2 suggested that Kurdish respondents should not increase their support for the government following curfew exposure. If anything, they should decrease this support because as out-group members they are more likely to perceive government actions as unjustified. Although we do not observe a significant negative effect on Vote AKP for Kurds using 95% confidence intervals, we see an estimated effect for Kurds that is negative, indicating that curfews do not increase their support for the governing party like they do for Turks.

Figure 2(b) illustrates the effect of curfew exposure on support for the Kurdish opposition and provides support for Hypothesis 3. We expected curfew exposure to decrease the probability of HDP support among the Kurdish respondents. In line with expectations, the likelihood that a Kurdish respondent expresses an intention to vote for the HDP decreases by just over 10% following curfew exposure. The effect of curfews on Turks’ support for the HDP, which is very low already, is statistically insignificant. The 9% difference between these substantive effects is statistically significant (Figure OA 2.2).

Figure 2(c) illustrates the substantive effect of Curfew on Abstain. We find that curfew exposure increases the likelihood that a Kurdish respondent will abstain by 17%. This effect is statistically significant. Curfew exposure does not have a statistically significant influence on the Turkish respondents’ likelihood of abstaining. The 19% difference in the substantive effect between Turks and Kurds is statistically significant (Figure OA 2.2). This finding supports Hypothesis 4, according to which we expected Kurdish respondents to become more likely to report political ambivalence following curfews.

Overall, the findings from the empirical models examining the influence of Curfew on political preferences confirm our theoretical expectations. We now discuss the findings from models examining the influence of Curfew Duration on political preferences. We graphically present the findings in Figures 3 to 5. The solid black lines in graphs display point estimates for changes in predicted probabilities as we change Curfew Duration from 0 (no curfew) to 500 h.²¹ The shaded regions show the corresponding 95% confidence intervals for each substantive effect. We hold all control variables at their mean values rounded to the nearest unit for generating substantive effects.

Figure 3.

Effect of curfew duration on AKP support.

Figure 4.

Effect of curfew duration on HDP support.

Figure 5.

Effect of curfew duration on abstaining.

Figure 3 illustrates the impact of curfew exposure measured in duration (Curfew Duration) on support for the governing party (Vote AKP) for Turkish and Kurdish respondents. Curfew exposure does not have a statistically significant impact on either group of respondents: the point estimates are minuscule and the confidence intervals cross zero. These results fail to support Hypothesis 1 but provide some support for Hypothesis 2 by illustrating that Kurds do not increase their support for the AKP following curfews.

Figure 4 shows the effect of curfew duration on support for the Kurdish opposition party Vote HDP. Curfew exposure decreases the likelihood of Kurdish respondents voting for the HDP. For example, when we increase curfew duration from 0 to 214 h (mean duration in the data), the predicted probability of a Kurdish respondent voting for the HDP decreases by 11%. When we increase curfew duration from 0 to 500 h, the predicted probability decreases by around 20%. Overall, the estimated drop in HDP support becomes substantively large as curfew duration increases. These findings provide strong support for Hypothesis 3.

Figure 5 illustrates the effect of curfew duration on abstaining. Curfew duration has a statistically significant positive effect on abstention only for the Kurdish respondents. The effect is relatively small for those who experienced a short curfew. For instance, when we increase curfew duration from 0 to 100 h, the change in predicted probability is close to 0. However, the effect increases in magnitude in a relatively linear manner as curfew duration increases. For example, if we increase curfew duration from 0 to 214 h (average duration in the data), Kurdish respondents become around 9% more likely to abstain from supporting a party or intending to vote. These findings strongly support Hypothesis 4. Overall, the findings from the empirical models examining the influence of Curfew Duration on political preferences are largely in line with our expectations. They support all of our hypotheses except for Hypothesis 1.

Electoral Outcomes

One might be concerned that our findings based on surveys do not match the actual voting behavior of respondents. Due to fear of government retaliation, respondents in curfew-affected areas may be afraid to express their true voting preferences. Respondents could also face a social desirability bias against expressing support for HDP, and this bias may have been larger in districts with curfews. Analyzing electoral data helps us address this concern. While we cannot examine differences in actual voting behavior of different ethnic groups (Turks vs Kurds) due to ballot secrecy and aggregation of election results, we can test some implications of our argument using election data.

Two sets of elections were held in Turkey in 2015: one in June, before curfew implementation, and another in November, by which time 15 districts had experienced curfews. We use district-level voter turnout and party support from these two elections. Recall that the districts that experienced curfews are overwhelmingly Kurdish (see Table 1), thus district-level election results should help us uncover the curfews’ effects on Kurdish voters. Since our survey results show that Kurdish respondents become less likely to support the HDP following curfews, we expect to find that the vote share of the HDP decreased in curfew-affected districts. Additionally, since the survey results show that Kurds become more likely to abstain following curfews, we expect to find that in curfew-affected districts invalid votes increase and voter turnout decreases.

We employ a difference-in-difference research design and compare the change in party support and turnout across June and November 2015 elections in curfew-affected districts to all other districts that did not experience a curfew. Difference-in-difference designs do not involve controlling for covariates, but instead rely on the “parallel trends” assumption. This assumption states that aside from the impact of the “treatment” (curfews), all districts should have similar increases or decreases in voting behaviors over time. Another way of thinking about this assumption is that the influence of potential confounding factors can vary over time if it is the same for all geographic units, or across geographies if it is the same for these units over time, but not by a combination of both time and geography. If this assumption is met, the model estimates the impact of the treatment (curfews) on observed voting behavior independently of potential confounding factors that influence both curfew implementation and voting outcomes. Since these two elections were held very close to each other, we are fairly confident that aside from the curfews, districts did not separately experience large changes that would invalidate this assumption. Table 3 shows the results of the difference-in-difference models, which were calculated from linear regressions using fixed effects for districts and a fixed effect for the November election along with a dummy variable for curfew exposure.

Table 3.

Effect of curfews on election results (Diff-in-Diff models).

	(1)	(2)	(3)	(4)
Variables	DV: AKP	DV: HDP	DV: Turnout	DV: Invalid
Curfew	−3.884***	−2.019**	−3.124***	1.039***
	(1.049)	(0.836)	(0.621)	(0.220)
Nov. Election	8.890***	−1.927***	0.452***	−0.632***
	(0.125)	(0.100)	(0.074)	(0.026)
Constant	42.977***	12.386***	85.630***	2.187***
	(0.088)	(0.070)	(0.052)	(0.018)
R-squared	0.828	0.273	0.051	0.357
Observations	2110	2110	2110	2110
Number of districts	1055	1055	1055	1055
District FE	Yes	Yes	Yes	Yes

Notes: Table entries are coefficient estimates from two-way fixed effect models, which are equivalent to difference-in-difference models for balanced panel data. The fixed effect coefficient for the November election is shown, while the fixed effects for districts are not shown. Standard errors are in parentheses. ***p $<$ 0.01, **p $<$ 0.05, *p $<$ 0.1.

Table 3 shows that curfew-affected districts experienced a decrease in AKP vote share of almost 4 percentage points relative to other districts, which experienced an almost 9-point increase in AKP support on average. Recall that these districts are overwhelmingly Kurdish. Thus, this finding is in line with our first hypothesis which suggested that Kurdish respondents should not be more likely to express political support for the governing parties as a result of indiscriminate counterinsurgency actions. Additionally, we find that curfew-affected districts experienced a decrease in HDP vote share of around 2 percentage points relative to other districts, which is consistent with the survey results from Figure 2(b) (Hypothesis 3).

Our survey results on the Abstain dependent variable imply that we should observe either lower voter turnout or more invalid votes in curfew-affected districts. Consistent with the survey findings, we find that curfew-affected districts experienced a statistically significant decrease in voter turnout of around 3 percentage points relative to other districts. Furthermore, these areas experienced a significant but small (1%) increase in invalid votes relative to other districts, which is also consistent with the findings in Figure 2(c) (Hypothesis 4). Overall, the analysis of the election data help alleviate the concern that our survey findings are simply the result of social desirability bias or fear of individual targeting; instead, the patterns that we observe with the attitudinal data are reflected in actual voting behavior.

Additional Robustness Checks

We conducted several additional analyses as robustness tests. The results of these analyses are reported in online appendices and confirm that our results are robust to different model specifications, including those which trade a multi-category dependent variable for stronger controls for unmeasured factors over time and by geography. More specifically, our conclusions remain the same when we use three separate binary dependent variables, Vote AKP, Vote HDP, and Abstain, and estimate linear probability models with fixed effects for both districts and survey months. In the appendix we also show that our results are mostly robust when we use an alternative measures of conflict in respondents’ districts based on the Turkish State-PKK Conflict Event Database (TPCONED); see OA 3.2 (Kibris 2021). The only finding that loses statistical significance is the influence of curfew exposure on Turkish respondents support for the AKP (Hypothesis 1).

Another potential concern is that, rather than reflecting changes in respondents’ political party support, our results are instead detecting changes in the number or type of respondents that the survey company was able to contact as a result of curfews. In the appendix we present an analysis of the number and types of respondents contacted per month in areas affected by curfews versus other areas. There were no significant differences across these areas before and after a curfew in either the number of respondents contacted or their demographic profiles.

We have shown that our main findings are robust to a variety of different model specifications and subsets of data. Still, we cannot fully eliminate the possibility of confounding from omitted variable bias. However, we can perform sensitivity analysis on the linear probability models used in OA 3.1 to estimate how strong this confounding would have to be in order to eliminate our statistically significant findings. We provide the details of the sensitivity analysis in OA5 and find that, for almost all of the analyses, an omitted variable would need to have a relatively large impact – larger than all observed variables except for Kurdish ethnicity – to make the observed results insignificant.

Conclusion

We set out to explore the effect of indiscriminate state counterinsurgency actions on citizens’ political preferences using curfews as an example of such actions. We argued that affected citizens’ responses to such state actions depend on their group alignment. Citizens whose group identity is aligned with the government tend to interpret the potentially harmful government actions through the lens of in-group favoritism – as legitimate counterinsurgency operations intended to root out militants – and reward the government with further political support as a result. Citizens who are part of the out-group and therefore less likely to be aligned with the government, however, interpret these same actions as targeted and discriminatory against their group. Rather than shifting their support from the government to the political parties affiliated with the insurgent movement, these respondents become less likely to vote or to express a certain voting preference. These citizens become more likely to disengage from politics because, while they have no reason to increase support for the governing parties, they may also be reluctant to express support for the political forces associated with the insurgent movement out of fear of further retribution.

We find support for this argument with survey data from Turkey. Specifically, our empirical analysis shows three main findings. First, we find evidence that exposure to curfews increases the likelihood that Turks (i.e., individuals who are more likely to be aligned with the government) support the governing party, while we see no increase in AKP support among Kurds. However, these results depend on the model specification, with the binary exposure model showing a significant increase for Turks while the duration model does not. Second, we find that exposure to curfews decreases Kurdish respondents’ likelihood of voting for the Kurdish opposition party. This is in line with our argument that Kurds, i.e., the out-group members, interpret government actions as discriminatory against their group, which (a) motivates them to not reward the governing party and (b) discourages them from showing support for the Kurdish opposition party. Finally, the analysis shows that Kurds also become more likely to express political disengagement following curfew exposure. This is also in line with our argument according to which the out-group members, who have no reason to support the government and are fearful of supporting the opposition, become politically ambivalent or disenfranchised instead.

These findings make several important contributions to the existing literature. First, we expand the study of indiscriminate state counterinsurgency actions beyond their military consequences, and show that these actions (at least in the form of curfews) affect political preferences. This is important because prior work considers public attitudes as the primary mechanism by which counterinsurgency affects military outcomes but rarely studies them directly. Understanding the effect of indiscriminate repression and counterinsurgency on political preferences is also important in its own right, especially in democracies, where these preferences influence electoral outcomes.

Overall, our results bear some similarity to the extant studies which find that indiscriminate state actions that cause civilian suffering can contribute to the formation of more radical views among the civilian populations (Hatz 2018; Jaeger et al. 2012; Longo, Canetti and Hite-Rubin 2014). While our analysis does not focus on radicalization, our results indicate that indiscriminate state actions can sway citizens away from peaceful political processes. The findings that the Kurdish respondents become more likely to abstain from voting and less likely to vote for Kurdish opposition following curfew exposure suggest that peaceful political participation can potentially lose its allure due to indiscriminate state actions.

Second, we show that the effects of indiscriminate state actions are not uniform but differ significantly depending on whether an individual is ethnically aligned with the government or the insurgent movement. Reactions to indiscriminate counterinsurgency actions are subject to in-group favoritism, out-group animosity, and fear of retribution. Detailing exactly how group dynamics play out in this context contributes to theory development in understanding the consequences of indiscriminate state repression, where these dynamics are often overlooked. More specifically, we add to studies that emphasize the role of in-group favoritism in determining civilian response to state actions, suggesting that an in-group perpetrator is less likely to be punished (Lyall, Blair and Imai 2013). In line with this, we find that Turks do not punish the government for curfews and may even reward it instead.

Our findings also have important real world implications. They suggest that the use of indiscriminate counterinsurgency measures such as curfews are not politically neutral and can significantly affect citizens’ political preferences and electoral choices. Most concerningly, such government actions may frighten individuals who are ethnically and potentially ideologically misaligned with the government into not declaring any political allegiance, effectively disengaging and demobilizing them politically. Thus, governments may be able to use indiscriminate counterinsurgency measures such as curfews to build their support among in-group members and thwart the political strength of the out-group insurgent movements.

Supplemental Material

sj-pdf-1-jcr-10.1177_00220027221109788 – Supplemental Material for The Effect of Curfews on Political Preferences

Supplemental Material, sj-pdf-1-jcr-10.1177_00220027221109788 for The Effect of Curfews on Political Preferences by Deniz Aksoy, Andrew Menger and Margit Tavits in Journal of Conflict Resolution

Footnotes

Acknowledgments

Authors would like to thank William Winston for excellent support with GIS data, Aydin Erdem, Eren Pultar and KONDA research and consultancy for sharing the monthly survey data. Authors also thank Carly Wayne, and the participants of the 9th Annual Eurasian Peace Science Conference at Koc University in Istanbul for their helpful feedback.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The authors gratefully acknowledge financial support for this study from the Weidenbaum Center on the Economy, Government and Public Policy.

ORCID iD

Deniz Aksoy

Supplemental Material

Supplemental material for this article is available online.

Notes

References

Aktürk

Sener

. 2016. “Why Did the PKK Declare Revolutionary People’s War in July 2015?” Contemporary Turkish Politics 22: 59‐63.

Aydin

Aysegul

Emrence

Cem

. 2016. “Politics of Confinement: Curfews and Civilian Control in Turkish Counterinsurgency.” Contemporary Turkish Politics 22: 63‐66.

Balcells

Laia

. 2011. “Continuation of Politics by Two Means: Direct and Indirect Violence in Civil War.” Journal of Conflict Resolution 55, no. 3: 397‐422.

Bali

Valentina A.

2003. “Terror and Elections: Lessons from Spain.” Electoral Studies 26, no. 3: 6691‐7687.

Başlevent

Cem

Kirmanoğlu

Hasan

Şenatalar

Burhan

. 2005. “Empirical Investigation of Party Preferences and Economic Voting in Turkey.” European Journal of Political Research 44, no. 4: 547‐562.

Benmelech

Efraim

Claude

Berrebi

Klor

Esteban

. 2014. “Counter-Suicide-Terrorism: Evidence from House Demolitions.” The Journal of Politics 77, no. 1: 27‐43.

Bennett

D. Scott

. 2008. “Governments, Civilians, and the Evolution of Insurgency: Modeling the Early Dynamics of Insurgencies.” Journal of Artificial Societies and Social Simulation 11, no. 47.

Berman

Eli

Shapiro

Jacob N

Felter

Joseph H

. 2011. “Can Hearts and Minds be Bought? The Economics of Counterinsurgency in Iraq.” Journal of Political Economy 119, no. 4: 766‐819.

Berman

Eli

Felter

Joseph H.

Shapiro

Jacob N.

. 2018. Small Wars Big Data: The Information Revolution in Modern Conflict. Princeton, NJ: Princeton University Press.

10.

Berrebi

Claude

Klor

Esteban

. 2006. “On Terrorism and Electoral Outcomes: Theory and Evidence from the Israeli-Palestinian Conflict.” The Journal of Conflict Resolution 50: 899‐925.

11.

Birnir

Jóhanna

Gohdes

Anita

. 2018. “Voting in the Shadow of Violence: Eelctoral Politics and Conflict in Peru.” Journal of Global Security Studies 3, no. 2: 181‐197.

12.

Brass

Paul R

. 2006. “Collective Violence, Human Rights, and the Politics of Curfew.” Journal of Human Rights 5: 323‐340.

13.

Brewer

Marilynn B.

1999. “The Psychology of Prejudice: Ingroup Love and Outgroup Hate?” Journal of Social Issues 55, no. 3: 429‐444.

14.

Bueno de Mesquita

Ethan

Dickson

Eric S.

. 2006. “The Propaganda of the Deed: Terrorism, Counterterrorism, and Mobilization.” American Journal of Political Science 51, no. 2: 364‐381.

15.

Çarkoğlu

Ali

Toprak

Binnaz

. 2000. Türkiye’de Din, Toplum ve Siyaset. Türkiye Ekonomik ve Sosyal Etüdler Vakfı.

16.

Carter

David B.

2016. “Provocation and the Strategy of Terrorist and Guerrilla Attacks.” International Organization 70, no. 1: 133‐173.

17.

Cesur

Resul

Mocan

Naci

2018. “Education, Religion, and Voter Preference in a Muslim Country.” Journal of Population Economics 31, no. 1: 1‐44.

18.

Çınar

Kursat

2016. “Local Determinants of an Emerging Electoral Hegemony: The Case of Justice and Development Party (AKP) in Turkey.” Democratization 23, no. 7: 1216‐1235.

19.

Condra

Luke N.

Shapiro

Jacob N.

2012. “Who Takes the Blame? The Strategic Effects of Collateral Damage.” American Journal of Political Science 56, no. 1: 167‐187.

20.

Cunningham

David E.

Idean

Salehyan

Gleditsch

Kristian Skrede

Saleyhan

Idean

2013. “Non-State Actors in Civil Wars: A New Dataset.” Conflict Management and Peace Science 30, no. 5: 516‐531.

21.

Davenport

Christian

2007. “State Repression and Political Order.” Annual Review of Political Science 10: 1‐23.

22.

de la Calle

Luis and Ignacio Sánchez-Cuenca

Sanchez-Cuenca

Ignacio

2013. “Killing and Voting in the Basque Country: An Exploration of the Electoral Link Between ETA and its Political Branch.” Terrorism and Political Violence 25, no. 1: 94‐112.

23.

Donner

Allan

Klar

Neil

2000. Design and Analysis of Cluster Randomization Trials in Health Research. London: Arnold.

24.

Downes

Alexander.

2007. “Draining the Sea by Filling the Graves: Investigating the Effectiveness of Indiscriminate Violence as a Counterinsurgency Strategy.” Civil Wars 9, no. 4: 420‐444.

25.

Downes

Alexander.

2008. Targeting Civilians in War. Ithaca: Cornell University Press.

26.

Duffy Toft

Monica

Zhukov

Yuri

2015. “Islamists and Nationalists: Rebel Motivation and Counterinsurgency in Russia’s North Caucasus.” American Political Science Review 209, no. 2: 222‐238.

27.

Enders

Walter

Sandler

Todd

1993. “The Effectiveness of Antiterrorism Policies: A Vector-Autoregression-Intervention Analysis.” American Political Science Review 87, no. 4: 829‐844.

28.

Esmer

Yılmaz

2002. At the Ballot Box: Determinants of Voting Behaviour in Turkey. Boulder, CO: Lynn Rienner. Chapter in Politics, Parties and Elections in Turkey edited by Esmer

Yılmaz

Sayarı

Sabri

29.

Findley

Michael G.

Young

Joseph K.

2007. “Fighting Fire with Fire? How (Not) to Neutralize an Insurgency.” Civil Wars 9, no. 4: 378‐401.

30.

Gambetta

Diego

1993. The Sicilian Mafia: Business of Private Protection. Harvard University Press.

31.

Gassebner

Martin

Jong-A-Pin

Richard

OMierau

Jochen

2008. “Terrorism and Electoral Accountability: One Strike, You’re Out!” Economics Letters 100, no. 1: 126‐129.

32.

Gibson

James L

Nelson

Michael J

2018. Blacks and Blue: How and Why African Americans Judge the American Legal System. USA: Oxford University Press.

33.

Gutierrez-Sanin

Francisco

Wood

Elisabeth Jean

2007. “What Should We Mean by “Pattern of Political Violence”? Repertoire, Targeting, Frequency, and Technique.” Perspectives on Politics 141, no. 2: 1281‐1301.

34.

Hatz

Sophia

2018. Selective or Collective? Palestinian Perceptions of Targeting in House Demolition, 1‐21. Conflict Management and Peace Science.

35.

Healy

Andrew

Malhotra

Neil

2013. “Retrospective Voting Reconsidered.” Annual Review of Political Science 16: 285‐306.

36.

Herrin

Jeph

2012. “CLTEST: Stata modules for performing cluster-adjusted chi-square and t-tests.” URL: https://EconPapers.repec.org/RePEc:boc:bocode:s424901

37.

Hewstone

Miles

Rubin

Mark

Willis

Hazel

2002. “Intergroup Bias.” Annual Review of Psychology 53, no. 1: 575‐604.

38.

Hultquist

Philip

2017. “Is Collective Repression an Effective Counterinsurgency Technique? Unpacking the Cyclical Relationship Between Repression and Civil Conflict.” Conflict Management and Peace Science 34, no. 5: 507‐525.

39.

Humphreys

Macartan

Weinstein

Jeremy M.

2006. “Handling and Manhandling Civilians in Civil War.” American Political Science Review 100, no. 3: 429‐447.

40.

Hurwitz

Jon

Peffley

Mark

2005. “Explaining the Great Racial Divide: Perceptions of Fairness in the US Criminal Justice System.” The Journal of Politics 67, no. 3: 762‐783.

41.

Jaeger

David A.

Klor

Esteban F.

Miaari

Sami H.

Paserman

Daniele

2012. “The Struggle for Palestinian Hearts and Minds: Violence and Public Opinion in the Second Intifada.” Journal of Public Economics 96: 354‐368.

42.

Johnson

Devon

Kuhns

Joseph B.

2009. “Striking Out: Race and Support for Police Use of Force.” Justice Quarterly 26, no. 3: 592‐623.

43.

Kalyvas

Stathis

2006. The Logic of Violence in Civil War.

44.

Kayaoğlu

Ayşegül

2017. “Voting Behavior of the Youth in Turkey: What Drives Involvement in or Causes Alienation from Conventional Political Participation?” Turkish Studies 18, no. 1: 32‐55.

45.

Kibris

Arzu

2014. “The Polarization Trap.” Studies in Conflict and Terrorism 37: 492‐522.

46.

Kibris

Arzu

2021. “The Geo-Temporal Evolution of Violence in Civil Conflicts: A Micro Analysis of Conflict Diffusion on a New Event Data Set.” Journal of Peace Research 58: 885‐899.

47.

Kocher

Matthew Adam

Pepinsky

Thomas

Kalyvas

Stathis

2011. “Aerial Bombing and Counterinsurgency in the Vietnam War.” American Journal of Political Science 55, no. 2: 201‐218.

48.

Lafree

Gary

2010. “The Global Terrorism Database: Accomplishments and Challenges.” Perspectives on Terrorism 4: 24‐46.

49.

Lewis-Beck

Michael S

Stegmaier

Mary

2000. “Economic Determinants of Electoral Outcomes.” Annual Review of Political Science 3, no. 1: 183‐219.

50.

Lichbach

Mark

1994. “What Makes Rational Peasants Revolutionary? Dilemma, Paradox, and Irony in Peasant Collective Action.” World Politics 36, no. 3: 383‐418.

51.

Livny

Avital

2020. Trust and the Islamic Advantage. Cambridge University Press.

52.

Longo

Matthew

Canetti

Daphna

Hite-Rubin

Nancy

2014. “A Checkpoint Effect? Evidence from a Natural Experiment on Travel Restrictions in the West Bank.” American Journal of Political Science 58: 1006‐1023.

53.

Lyall

Jason

2009. “Does Indiscriminate Violence Incite Insurgent Attacks?” Journal of Conflict Resolution 53, no. 3: 331‐362.

54.

Lyall

Jason

Blair

Graeme

Imai

Kosuke

2013. “Explaining Support for Combatants during Wartime: A Survey Experiment in Afghanistan.” American Political Science Review 107, no. 4: 679‐705.

55.

Marcus

Aliza

2007. Blood and Belief: The PKK and the Kurdish Fight For Independence. New York, NY: New York University Press.

56.

Marschall

Melissa

Aydoğan

Abdullah

Bulut

Alper

2016. “Does Housing Create Votes? Explaining the Electoral Success of the AKP in Turkey.” Electoral Studies 42: 201‐212.

57.

Marshall

Monty G.

Gurr

Ted Robert

Jaggers

Keith

2018. Polity IV dataset version 2017. www.systemicpeace.org

58.

Mason

T. David

2004. Caught in the Crossfire: Revolution, Repression, and the Rational Peasant.

59.

McDowall

David.

2000. A Modern History of the Kurds. London: I.B. Tauris.

60.

Parkinson

Sarah Elizabeth

2013. “Organizing Rebellion: Rethinking High-Risk Mobilization and Social Networks in War.” American Political Science Review 107, no. 3: 418‐432.

61.

Rosendorff

B Peter

Sandler

Todd

2004. “Too Much of a Good Thing? The Proactive Response Dilemma.” Journal of Conflict Resolution 48, no. 5: 657‐671.

62.

Sözen

Yunus

2016. “The Parliamentary Elections in Turkey, November 2015.” Electoral Studies 43: 200‐203.

63.

START

2008. National Consortium for the Study of Terrorism and Responses to Terrorism. http://www.start.umd.edu

64.

Stoll

David.

1993. Between Two Armies in the Ixil Towns of Guatemala. New York: Columbia University Press.

65.

Tajfel

Henri

Turner

John C.

1979. “An Integrative Theory of Intergroup Conflict.” In The Social Psychology of Intergroup Relations, edited by Austin

William G.

Worchel

Stephen

, eds. Monterey, CA: Brooks/Cole.

66.

Tezcür

Gunes Murat

2010. “When Democratization Radicalizes: The Kurdish Nationalist Movement in Turkey.” Journal of Peace Research 47, no. 6: 775‐789.

67.

Tezcür

Gunes Murat

2016a. “Foreign Fighters from Turkey.” Contemporary Turkish Politics 22: 67‐71.

68.

Tezcür

Gunes Murat

2016b. “Ordinary People, Extraordinary Risks: Participation in an Ethnic Rebellion.” American Political Science Review 110, no. 2: 247‐264.

69.

Tezcür

Gunes Murat

Gürses

Mehmet

2017. “Ethnic Exclusion and Mobilization: The Kurdish Conflict in Turkey.” Comparative Politics 49, no. 2: 212‐234.

70.

Thompson

Robert

1966. Defeating Communist Insurgency. London: Chatto & Windus.

71.

TIHV

2017. “TIHV D’okumantasyon Merkezi Verilerine Gore 2 Yılda Ilan Edilen Sokaga Cikma Yasaklari.” https://tihv.org.tr

72.

Tishkov

Valery

2004. Chechnya: Life in a War. Torn Society.

73.

Tomz

Michael

Wittenberg

Jason

King

Gary

2003. “CLARIFY: Software for Interpreting and Presenting Statistical Results.” Journal of Statistical Software 8.

74.

Tse-tung

Mao

1961. On Guerrilla Warfare. Chicago: Praeger.

75.

2017. “Report on the Human Rights Situation in South-East Turkey.”. https://www.ohchr.org

76.

Viterna

Jocelyn

2013. “Women in War: The Micro-Processes of Mobilization in El Salvador.”

77.

Wood

Elizabeth

2003. Insurgent Collective Action and Civil War in El Salvador. Cambridge, UK: Cambridge University Press.

78.

Zhukov

Yuri M.

2013. “An Epidemic Model of Violence and Public Support in Civil War.” Conflict Management and Peace Science 30, no. 1: 24‐52.

79.

Zussman

Asaf

Zussman

Noam

2006. “Assassinations: Evaluating the Effectiveness of an Israeli Counterterrorism Policy using Stock Market Data.” Journal of Economic Perspectives 20, no. 2: 193‐206.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.49 MB