Rally post-terrorism

Chen, Shuai Article — Published Version Rally post-terrorism Journal of Population Economics Provided in Cooperation with: Springer Nature Suggested Citation: Chen, Shuai (2025) : Rally post-terrorism, Journal of Population Economics, ISSN 1432-1475, Springer, Berlin, Heidelberg, Vol. 38, Iss. 2, https://doi.org/10.1007/s00148-025-01106-2 This Version is available at: https://hdl.handle.net/10419/323186 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. http://creativecommons.org/licenses/by/4.0/ Journal of Population Economics (2025) 38:50 https://doi.org/10.1007/s00148-025-01106-2 ORIGINAL PAPER Rally post-terrorism Shuai Chen1,2,3 Received: 13 July 2022 / Accepted: 16 May 2025 / Published online: 2 June 2025 © The Author(s) 2025 Abstract This study examines whether the “rally ’round the flag” phenomenon is present after terrorist attacks, and investigates explanations for this increase in confidence in national political institutions and approval of the country leader’s job performance. I exploit variations in terrorist occurrences and results (success or failure) across subnational EU regions where at least one attack took place during the data period. I show empirically that both terrorist occurrences and results are plausibly exogenous to the prior political and economic climate. Conducting a difference-in-differences analysis, I compare changes in political confidence and approval among individuals who were exposed to an attack in their region with those who were not. Utilizing another more sophisticated identification, I also compare such political changes after successful attacks with those after failed attacks. I find that post-terrorism, individual political confidence and support significantly increased by more than 10 percentage points, and that this political increment was over 5 percentage points after successful attacks relative to failed ones. Such rally effects were temporary and faded away within a year. Furthermore, I explore various potential channels suggesting patriotism and civic engagement as mechanisms while rejecting perceived economic capture and political acquisition as alternative explanations. Keywords Rally ’round the flag effect ·Terrorism ·Confidence in institutions · Political approval ·Patriotism ·Civic engagement ·Economic capture · Political acquisition JEL Classification D74 ·H12 ·P16 Responsible editor: Alfonso Flores-Lagunes BShuai Chen [email protected] 1University of Leicester, University Road, Leicester LE1 7RH, UK 2Institute of Labor Economics (IZA), Bonn, Germany 3Global Labor Organization (GLO), Essen, Germany 123 50 Page 2 of 29 S. Chen 1 Introduction In the political sciences literature, the “rally ’round the flag” effect refers to the phenomenonofariseinpopularsupportforthenationalleadershipfollowinginternational crises (Mueller 1970; Hetherington and Nelson 2003). After the 9/11 attacks and despite the increased counterterrorism spending worldwide, both transnational and domestic terrorist incidents have resurged, and the consequences of transnational terrorism have become more disastrous than those of domestic terrorism, especially since 2008 (see Section 2 of Gaibulloev and Sandler 2019). Thus, terrorism is likely to have become an important potential source of the rally phenomenon. Such a rally effect following terrorist attacks has important economic implications. For immediate crisis aid or management after terrorism, a government often needs to reallocate economicresourcesfromproductiveinvestmentstowardgovernmentspending(Blomberg et al. 2004). Moreover, making counterinsurgency efforts, launching reconstruction and community projects, providing social services and taking large numbers of indiscriminate conciliatory acts are effective measures the government can take in the shortto mid-term to reduce terrorism (Berman 2009;Bermanetal.2011; Dugan and Chenoweth 2012). The success of these measures relies on the civilian rally: popular support for the national leadership and/or civic participation in the projects and services. The current study examines the possibility of the rally effect by investigating the influences of localized terrorism at the subnational level on individual confidence in national political institutions and political support for the country’s leader in the European Union. Moreover, I explore various potential mechanisms, including patriotism, civic engagement, perceived economic capture and perceived political acquisition by the government. This investigation may facilitate understanding whether such political solidarity is more likely to be voluntary or to be elicited by the political incumbent. Specifically, I manually assemble a comprehensive survey at the individual level with the universe of terrorist attacks in the EU at the first-order subnational (i.e., NUTS 1) regional level. This is possible through utilizing survey information about interview dates and locations. In this way, I connect the experience of terrorism with individual confidence in political institutions and political approval. I examine both confidence in the national government and approval of the country leader’s performance. I apply a difference-in-differences (DID) approach to compare changes in political confidence and approval among individuals who were exposed to a terrorist attack in their region with those who were not. I restrict the sample to individuals residing in regions that were hit by at least one attack during the data period of 2008–2016. I impose this restriction because targeted regions may differ from non-targeted ones in various aspects, including the political landscape. Such geographic heterogeneity might also intertwine with local political shocks to trigger terrorist occurrences. Inspired by Brodeur (2018), I take two steps to verify my identification strategy. First, I show that the occurrence of terrorism is orthogonal to various potential confounding attitudes to, and perceptions of, law and order, politics and economic conditions for 1 year, 6 months, or 3 months before a terrorist incident. Thus, I argue that the attack is plausibly exogenous to the prior political and economic climate that may be correlated with individual political confidence and approval. 123 Rally post-terrorism Page 3 of 29 50 Second, in the same manner, I again provide evidence that the success or failure of terrorism (conditional on its occurrence) is also plausibly exogenous to the perceived political and economic situation before an attack. The success or failure of terrorist attacks is defined based on the type of attack that took place, and thus is dependent on the latter factor.1The identification of terrorism effects then comes from the “inherent randomness” of the success or failure of an attack (Brodeur 2018). With this more sophisticated strategy, I then compare changes in political confidence and approval after successful attacks with those after failed ones. By these two means, I document a consistent rally effect after terrorism. I find that in the subsequent months up to 1 year post-attacks, the likelihood of having confidence in the national government increased by 11 percentage points. The likelihood of approval of the country leader’s performance rose by more than 13 percentage points after a terrorist incident. Both effects are sizable, in comparison to around 40% rate of political confidence and approval prior to terrorism. Moreover, the increment was 6.5 percentage points for confidence in the national government, and more than 5 percentage points for approval of the country leader’s performance after successful attacks relative to failed ones. These increments are not trivial, compared with about 50% rate of confidence and 45% rate of approval in the case of failed terrorism. The rally effect is thus exhibited post-terrorism. However, such an effect is short-term and fades away two quarters after the terrorist incident, and one quarter after the successful attack in comparison to the failed attack. In a second step, I explore potential channels through which terrorism affects individual political confidence and approval. First, I provide suggestive evidence for the mechanisms of patriotism and civic engagement. In regions hit by terrorist attacks, stronger patriotism may prevail among civilians. With proxies for patriotism, I document a significant improvement in nationalist sentiment following terrorism. I also find that the increases in political confidence and approval after attacks launched by domestic perpetrators were only half as large as those after attacks launched by nondomestic terrorists. This evidence may also reflect an element of nationalism. Another mechanism of civic engagement may come into effect if residents participate in activities of community rebuilding and need aid or support from the government. Their impression of the government may improve through mutual cooperation with these activities. I find evidence for enhanced civic engagement and reduced dissent from officials after terrorist incidents, which implies this channel of civic engagement. I then investigate other alternative explanations related to perceived economic captureandpoliticalacquisitionbythegovernment.Afteraterroristattack,thegovernment may allocate economic resources to the affected region to pacify local civilians and maintain their political support. Without direct measurements of resource allocation, I instead analyze the before/after terrorism variation in a composite index of satisfaction with various aspects of the residents’ area, including infrastructure and housing, as well as variations in confidence in the police and perceived safety in the local region. I do not find any significant effects on the subjective or perceived reallocation of economic resources in the targeted areas. Given the subjective nature of 1Appendix C reports the types of terrorist attacks and their corresponding definitions of success in the Global Terrorism Database. If an attack is not considered successful, it is regarded as failed. 123 50 Page 4 of 29 S. Chen political confidence and approval, perceived economic capture does not seem to play an important role in explaining the political effects of terrorism. In addition, a government may grant more political freedom in order to acquire local support. Examining the perceived freedom of expressing political views and the perceived freedom of the media, I do not find evidence for any corresponding improvement after terrorism. Given that the European Union is already an advanced democracy, there may not be much room in terms of franchise extension for the government to manipulate. This study addresses several strands of literature. Principally, it adds to the growing academic discussion on the political consequences of terrorism. Godefroidt (2023) conducted a meta-analysis on this literature, concluding that terrorism was associated with out-group hostility, political conservatism and the rally phenomenon, though the effect sizes depended on contexts. The first focus of the discussion is on support for and attitudes toward the incumbent. Gassebner et al. (2011) analyzed more than two thousand cabinets in over 150 countries between 1970 and 2002, and found that on average, terrorism shortened cabinet duration. Using the timing of the terrorist attacks in Madrid on 11 March 2004, only 3 days before the congressional election, Montalvo (2011) compared individuals voting before the terrorist acts with those voting afterwards. He documented a negative impact of the attacks on electoral support for the ruling conservative party. Kibris (2011) used the security force casualties in the fight against terrorists as a measurement of the terrorist intensity in Turkey, and found that Turkish voters were sensitive to these incidents; they blamed the government for the losses. Chowanietz (2011) analyzed the responses of political elites and parties to terrorist acts, and found a rally effect among them. Chanley (2002) and Perrin and Smolek (2009) investigated the aftermath of the 9/11 attacks in the USA and documented an increase in trust in the national government. Similarly, Dinesen and Jæger (2013) examined changes in trust in institutions after the 3/11 Madrid terrorism, and expanded the types of institutions to politics, media and justice. Nussio (2020) studied the attitudinal and emotional consequences of the 2016 Berlin Christmas market Islamist terrorism, and documented unchanged political attitudes but escalated sadness and anger. Somewhat less related is the work of Bozzoli and Müller (2011), who used the London terrorist attacks in 2005 as a shock to examine individual risk perceptions. They found that terrorist-induced risk assessment enhanced the willingness to trade off security for liberties. Linking changes in individual democratic attitudes to regional variations in terrorist attacks in Pakistan, Rehman and Vanin (2017) identified a positive association between persistent exposure to terrorism and lower support for democratic values. Closest to my work are the papers of Peri et al. (2020,2021) and Amarasinghe (2023). Peri et al. (2020,2021) independently studied the link between terrorism and political opinions using European Social Survey data. The authors documented increased trust in the country’s parliament and satisfaction with the national government, but little evidence for changes in attitudes to immigration. Amarasinghe (2023) examined how a constructed composite index of public discontent changes after terrorism, with data at the country level for 135 countries. The author found a short-term increase in public discontent following successful attacks, and provided suggestive evidence that such discontent is aimed at the government. Compared with these studies and other prior work, my study adds to the literature in the following aspects: first, 123 Rally post-terrorism Page 5 of 29 50 the rich Gallup survey data in my study help to better delineate potential mechanisms of the terrorism-induced rally effect. I am able to simultaneously explore different channels by analyzing proxy variables for patriotism, civic engagement, perceived economic capture and political acquisition by the government. Thus, this study offers the first attempt to empirically disentangle voluntary solidarity from economically or politically elicited solidarity in the context of terrorism. Second, in the empirical analysis, I aim to combine micro-based data at the individual level with specific characteristics of terrorism such as attack types, logistics and salience. This approach facilitates the comparison of attacks of the same class. Given the empirical evidence provided in the paper for plausible exogeneity of the occurrence and result of terrorism, this approach can hence improve the identification of causal effects. Moreover, I exploit two identification strategies using terrorist occurrence and result, respectively, to verify the robustness of my results and conclusions. Finally, leveraging event data at the more precise subnational administrative (i.e., NUTS 1) regional level, I establish the post-localized-terrorism rally effect at the individual level, in terms of both national political institutions and the country’s leader. Following localized terrorism in the EU, not only did individual confidence in the national government increase, but individual support for the country leader’s performance also improved in the short run. Another focus of the political discussion is on the corresponding shift of the political landscape post-terrorism. Gould and Klor (2010) relied on geographic variations in terrorism from 1988 to 2006 in Israel to investigate political responses of Israelis. They found that Israelis became more willing to make territorial concessions and shifted their political landscape to the left following terrorist incidents. Even though Israelis tended to vote for right-wing parties, these parties moved to the left after terrorism.2 Using a German individual panel from 1999 to 2003, Schüller (2015) studied the effect of the 9/11 terrorist attacks on political mobilization and detected a resulting shift in a conservative direction. As for the same incident, Kaushal et al. (2007) found a rise in hate crimes and a reduction in intrastate migration of Arab and Muslim men, and Åslund and Rooth (2005) documented deteriorating attitudes toward certain ethnic minorities in Sweden after the 9/11 attacks in the U.S. Alesina et al. (2019)developed a model showing that voting for parties against criminal organizations decreased after violence conducted by these organizations. They found empirical evidence by analyzing large data in Sicily and across Italian regions. Combining long-term individual panels from several Western countries with global terrorism information over 20 years, Akay et al. (2020) found that global terrorism reduced subjective well-being and raised the intention to vote for conservative parties. Giavazzi et al. (2020) utilized a machine learning method to analyze texts of Twitter users in Germany. They discovered that after terrorism and criminal events, Twitter texts on average became similar to those of the far-right party AfD. They also found an association between shifts in textual similarity and changes in the vote shares in federal elections. The vast majority of the above studies presented a voting inclination toward the right-wing or conservative parties in the midor longer-term. By contrast, the current paper deals with the prompt political attitudinal responses to terrorist attacks. I document new 2Berrebi and Klor (2006,2008) and Getmansky and Zeitzoff (2014) also documented an increase in voting for ring-wing parties in response to terrorism in Israel. 123 50 Page 6 of 29 S. Chen evidence for the transitory attitudinal reactions concerning political confidence and approval directly after terrorist incidents. The corresponding rally effect in terms of these political attitudes became nonsignificant within a year following terrorism. Moreover, my paper complements the literature on the economic impact of conflicts or unrest. With regard to terrorism, Abadie and Gardeazabal (2003) documented a 10 percentage points reduction of GDP per capita in the Basque Country following the outbreak of terrorism in the late 1960s compared with a synthetic control region. Bloom et al. (2007) showed that during periods of high uncertainty, such as after 9/11, the responsiveness of investment by firms to any given policy stimulus may be much weaker. Kaushal et al. (2007) discovered an association between 9/11 and a 10 percentage points temporary decline in the real wage and weekly earnings of Arab and Muslim men. Similarly, Elsayed and De Grip (2018) found that following Islamic terrorist attacks across Europe, Muslim immigrants’ attitudes toward integration worsened, and that their (perceived) integration related to segregation and unemployment also deteriorated in the Netherlands. Analyzing Israelis’ reactions to terrorism during the “Al-Aqsa” Intifada, Becker and Rubinstein (2011) found that occasional users reduced their demand for goods and services subject to terrorist attacks as a response to fears of future dangers. Brodeur (2018) noted that successful terrorist attacks reduced jobs and total earnings in affected regions relative to failed attacks. The successful terrorist attacks especially diminished housing prices and raised consumer pessimism about business and buying conditions. Terrorism leads to a reallocation of economic resources and an increase in counterterrorism expenditure. For instance, Di Tella and Schargrodsky (2004) and Draca et al. (2011) both documented a notable increase in police deployment immediately after terrorist incidents in targeted regions, and a resulting reduction in local crime rates. According to the calculation by Mueller and Stewart (2014), domestic counterterrorism spending per year was 25 billion US dollars in 2010 before the 9/11 terrorist attacks, where the expenditure increased three-fold in the subsequent decade. In addition to counterterrorism measures, an incumbent may also try to accommodate the crisis in a political way by extending the franchise (Acemoglu and Robinson 2000a,b,2001; Aidt and Franck 2015; Besley and Persson 2011), or may attempt to acquire support in an economic way by reallocating resources to affected regions (Maffioli 2020;Zhu2018).3In the current study, I not only investigate both channels of perceived economic capture and perceived political acquisition by the government post-crisis, but also analyze whether these strategies work. I document no significant improvements in confidence in the police or perceived safety, no significant perceived reallocation of economic resources and no increase in perceived political freedom in the targeted regions from the perspective of citizens. Last but not least, the current study is closely related to research on changes in political trust and citizen cooperation corresponding to changes in institutional performance. Some studies in political science, for example, Mishler and Rose (2001), Rothstein and Stolle (2008) and Zmerli and Newton (2008), argued that political trust was endogenous to the performance of political institutions; that is, confidence in political institutions was a consequence rather than a cause of their performance. This 3Political connections with firms further facilitate such resource allocations (Faccio 2006; Fisman 2001; Hodler and Raschky 2014; Shleifer and Vishny 1994). 123 Rally post-terrorism Page 7 of 29 50 idea was modeled as “consensually strong state equilibrium” by Acemoglu (2005). Among empirical studies in economics, Aksoy et al. (2020) found a persistent negative impact of epidemic exposure in people’s “impressionable years” on confidence in political leadership and institutions. Such results were mainly driven by individuals in weak states incapable of acting against epidemics, thereby disappointing citizens. Dell et al. (2015) documented that in Northern Vietnam, where there was a stronger bureaucratic capacity, citizens were more able to organize and cooperate with the local government. Furthermore, using high-stakes lab-in-the-field experiments in Pakistan, Acemoglu et al. (2020) found that receiving information on the improved service delivery of state courts increased citizens’ willingness to interact with, and trust in, the courts. People with updated information also reduced their interactions with nonstateactors.Mystudy, instead,documents increased confidence in political institutions and political approval among citizens in adversely affected regions, who did not even perceive improvements in local community basics following terrorist incidents. This was even the case if a government failed to prevent terrorism (i.e., an attack was successful). However, additional analyses in the current study also suggest that the rally effects are not independent of a government’s performance against terrorism: indeed, an increase in political trust and support diminished if terrorist acts occurred repeatedly in a region over time. Additionally, I provide suggestive evidence for reinforced citizens’ civic engagement, as well as reduced dissent from officials after terrorist incidents. 2 Data In my analysis, I combine the Gallup Worldwide Research survey data with the Global Terrorism Database (GTD). The Gallup Worldwide Research draws a nationally representative random sample of individuals annually from more than 150 countries around the world. Every year it typically surveys at least 1000 individuals in every country, with either a1-hface-to-faceinterviewora 30-min telephoneinterview.These cross sectional data include respondents’ personal characteristics, such as gender, age, highest completed level of education, marital status, number of children living in the household, whether the respondent was born in the surveyed country and employment status. In addition, the survey asks a set of core questions covering topics including, for example, well-being, political opinions and religious opinions. Importantly, the data record the date and the first-order subnational administrative (i.e., NUTS 1) region for when and where the individual interview was conducted.4I exploit this information later to merge the Gallup data with the terrorism data. Data are available for the period 2008–2016. The Global Terrorism Database (GTD) documents over 190,000 international and domestic terrorist attacks that took place around the world from 1970 to 2018. The data define terrorist attacks as “the threatened or actual use of illegal force and violence by a non-state actor to attain a political, economic, religious, or social goal through fear, coercion, or intimidation.” To be included in the GTD, an incident has to present three 4Please refer to Appendix C2 for more detailed information about the Gallup Worldwide Research Survey. 123 50 Page 8 of 29 S. Chen Table 1 Means of terrorism characteristics by attack type, weapon type and logistics Success (%) Injury Death No. obs (%) Attack type Assassination 23.08 1.38 0.92 13 (2.41) Armed assault 90.67 7.75 1.75 75 (13.89) Bombing/explosion 45.91 1.54 0.19 257 (47.59) Faculty/infrastructure 94.58 0.18 0.01 166 (30.74) Unarmed assault 80.95 2.90 0.14 21 (3.89) Other and unknown 100.00 1.00 0.88 8 (1.48) Weapon type Firearms 86.67 11.56 2.89 45 (8.33) Explosives/bombs/dynamite 45.45 1.50 0.19 264 (48.89) Incendiary 93.14 0.32 0.01 175 (32.41) Melee 93.02 2.02 0.33 43 (7.96) Other and unknown 69.23 2.83 0.54 13 (2.41) International logistics 77.14 10.44 1.54 35 (6.48) Domestic logistics 80.52 0.77 0.04 77 (14.26) Unknown 65.89 1.63 0.34 428 (79.26) Total 68.70 2.04 0.38 540 (100) attributes: it must be intentional, it must entail some level of violence or immediate threat of violence, and the perpetrators of the incident must be subnational actors. In addition, the incident must meet at least two of the three following criteria: first, the act must be aimed at attaining a political, economic, religious, or social goal; second, there must be evidence of an intention to coerce, intimidate, or convey some other message to a larger audience (or audiences) than the immediate victims; and third, the action must be outside the context of legitimate warfare activities. In the current study, I follow the above definition and rules. For the vast majority of terrorist incidents filed, the data include information about thedateandlocation,thetypeoftheattack,whethertheattackwassuccessfulornot,the weapons used, the nature of the target, the number of casualties, and the identity of the perpetrator(s). The success or failure of terrorist attacks is defined based on the attack type that took place, thus is dependent on the latter factor.5For instance, a bombing is regarded as successful if the explosive device detonates, and as failed if it does not detonate. In order to alleviate concerns about the correlation between terrorism results and differences in counterterrorism capacities, I include country×year fixed effects and region-specific time trends, in addition to region fixed effects, to account for both time-varying and time-invariant geographical variations in counterterrorism capacities. Table 1displays the means of some important characteristics of the attacks in my sample. Among the 540 terrorist incidents occurring in EU countries from 2008 to 5See Appendix C for the types of terrorist attacks and their corresponding definitions of success in the Global Terrorism Database. 123 Rally post-terrorism Page 15 of 29 50 potential confounders are reported in Table B2 of Appendix B. None of them are significant even marginally at the 10% level, whether for the terrorism occurrence or for the terrorism success. Therefore, according to the corrected individual coefficient tests, both the terrorism occurrence and the terrorism success are plausibly exogenous to the political and economic climate prior to terrorism. The estimates suggest that this large set of potential confounders is considerably balanced between individuals before they were confronted with a terrorist incident and individuals who were not exposed to terrorism, as well as between individuals before they experienced a successful terrorist attack and individuals before they faced a failed one. Therefore, neither the occurrence nor the result of terrorism can be predicted by the perceived political and economic conditions before the incident. I exploit both strategies based on the occurrence and result of terrorist incidents in the empirical analyses. The conclusions drawn from the estimates with the two approaches are in accord, though the rally effects identified based on terrorist results appear more transitory. According to the above statistical tests, terrorist occurrences seem more plausibly exogenous than terrorist results to the politico-economic climate prior to attacks. Furthermore, Eq.1based on terrorism occurrences is parsimonious, and the coefficient of successful terrorism relative to failed terrorism in Eq.2based on terrorism results underestimates the effect of terrorism which is of interest (reported later in Table 2), estimates based on Eq.1are preferred and hence regarded as the baseline. Table 2 Effects of terrorism on confidence in government and approval of country leader’s performance (1) (2) (3) (4) A. Confidence in government Post-attack 0.081 (0.032)** 0.081 (0.031)*** 0.110 (0.013)*** 0.045 (0.030) Success×post 0.065 (0.026)** B. Approval of leader’s performance Post-attack 0.057 (0.031)* 0.057 (0.030)* 0.134 (0.021)*** 0.084 (0.037)** Success×post 0.051 (0.030)* Ind. covariates Y Y Y Terror. covariates Y Y Region FE Y Y Y Y Year×month FE Y Y Y Y Note: Based on 61,718 observations. Individual covariates contain a male dummy variable, the quadratic of age, dummy variables of completed educational levels, dummy variables of marital status, number of children living in the household, a dummy variable of being born in the country surveyed and dummy variables of unemployment and out of the labor force. Terrorism covariates include attack-type fixed effects, a dummy variable of logistically international attack, a dummy variable with the value of 1 if the attack is unknown to be logistically international or not and another dummy variable of the major attack. A constant is also included in every model, but not shown for parsimony. Standard errors clustered at the regional level in parentheses; *p<0.1; **p<0.05; ***p<0.01 123 50 Page 16 of 29 S. Chen Fig. 4 Confidence in government and approval of country leader’s performance across time around the terrorism occurrence. Note: Scale in the horizontal axis is in quarters. The segments denote 90% (dark) and 95% (light) confidence intervals, respectively, of estimated coefficients of interactions of terrorism and time periods around the occurrence of terrorism (the vertical dashed line) 4 Parameter estimates 4.1 Parallel pre-terrorism trends To examine the trends of the outcome variables for individuals before they were exposed to terrorism versus those who did not experience such events, I add a set of dummy variables for one quarter, two quarters and three quarters to 1 year before an attack, respectively, in Eq.1. If the coefficients of these quarterly dummy variables pre-terrorism are not significantly distinguishable from zero, it is evidence of parallel pre-trends. The corresponding estimates are visualized in Fig.4.16 The figure shows that all estimates pre-attacks are not significantly different from zero at the 5% level. Moreover, the coefficients for 3 months and 6 months after a terrorist incident are sizable and significant for both outcome variables, while estimates 9 months after attacks shrink and already lose significance at the 5% level. Hence, the results imply thatthe effects of terrorism onconfidence in politicalinstitutions and politicalapproval are merely short term. Similarly, Figs.4and 5display the coefficient estimates in the same type of event study, by separating successful terrorist attacks from failed ones.17 Following the occurrence of terrorist incidents (the vertical dashed line), the red circle estimates with thick segments denote successful attacks and the green square estimates with thin segments refer to failed ones. The rally phenomenon was evident after successful terrorist attacks within one quarter, and disappeared by the second quarter. It materialized more slowly after failed terrorism: the rally effect gradually emerged, and then became significant in the second quarter after the incidents, but vanished in the third quarter as well. Hence, irrespective of the attack result, terrorism triggered a temporary rally phenomenon which faded away within a year, though it operated significantly for one quarter at a different time depending on the result. 16 The estimates are reported in Table B3 of Appendix B2. 17 Likewise, the relevant estimates are also reported in Table B4 of Appendix B2. 123 Rally post-terrorism Page 17 of 29 50 Fig. 5 Confidence in government and approval of country leader’s performance across time; separating successful terrorism from failed attacks. Note: After terrorist occurrences (the vertical dashed line), the red circle estimates with thick segments refer to successful terrorism and the green square estimates with thin segments denote failed terrorism; see also note under Fig.4 4.2 Effects of terrorism on political confidence and approval Table 2provides an overview of the main parameter estimates for the effects of terrorism on confidence in the national government and on political approval of the country leader’s performance.18 The first column includes region fixed effects and time period fixed effects, without other controls. Column 2 adds individual characteristics, and column 3 further includes terrorism covariates. Column 4 displays the results of the more sophisticated model—the effects of successful terrorist attacks in comparison to failed ones. Panel A reports estimates for confidence in the national government. In the European Union after terrorist attacks, the probability of having confidence in the national government significantly increased by 8 to 11 percentage points across different model specifications in columns 1 to 3. Despite their qualitative similarity, the divergence in the sizes of the effect indicates the importance of the terrorism covariates. Without taking them into account, the political effects could be neutralized by different classes of attacks. In column 4, the coefficient in the second row shows the additive effect of successful terrorism to that of failed terrorism. Thus, compared with failed attacks, successful terrorist incidents significantly increased this probability by 6.5 percentage points; the occurrence of terrorism (also the failed events compared with no terrorism exposure) increased the probability by 4.5 percentage points, though not significantly. If we sum up the two coefficients in column 4, we obtain the effect of successful terrorist attacks relative to no terrorism exposure. The magnitude of the sum is the same as the estimate in column 3 since successful attacks dominate not only the number but also the impact of terrorism (see the descriptives in Table 1). Importantly, I find that the effects and their sizes are consistent between specifications in columns 3 and 4. Hence, both identification strategies in column 3 based on terrorism occurrences and in column 4 based on their results yield the same conclusions. Estimates in columns 3 and 4 may be both interpreted as causal effects. As discussed in the previous section, 18 Extended baseline estimates are reported in Appendix B3. If not specified otherwise, the analyses later are all based on the same estimation sample as in Table 2. 123 50 Page 18 of 29 S. Chen terrorism occurrences seem more plausibly exogenous than terrorism results to the politico-economic climate prior to terrorist attacks. Additionally, the model specification in column 3 is parsimonious, and the coefficient of successful terrorism relative to failed terrorism in column 4 underestimates the effect of terrorism, i.e., the effect relative to no terrorism exposure, estimates in column 3 are preferred and thus regarded as the baseline. Likewise, panel B presents the effects on approval of the country leader’s performance. The occurrence of terrorist attacks significantly raised the probability of approval of the country leader’s performanceby6to13.5 percentage points, depending on the model specification. Successful attacks significantly increased the performance approval by about 5 percentage points relative to failed ones (column 4). The effects of terrorism occurrences in column 3 of both panels are sizable, in comparison to around 40% rate of political confidence and approval prior to terrorism. The relative effects of successful attacks to failed ones, displayed in column 4 of both panels, are also not trivial, compared with about 50% rate of confidence and 45% rate of approval in the case of failed terrorism, respectively. Moreover, the rally effects of successful attacks in comparison to failed attacks, that is, the coefficients of Success ×post, are smaller and less statistically significant than their counterparts following (general/any) terrorist incidents, namely the coefficients in column 3. This makessenseinthat failedterrorismdrovea(milder)rallyphenomenon(thecoefficients of Post-attack in column 4). At the end of Section 3, I have presented empirical evidence that both the occurrence of terrorism and the success of attacks may be plausibly exogenous. In this section, I show that both approaches based on terrorism occurrencesandresultsdeliverconsistentconclusions,eventhoughtherallyeffects are more transitory when distinguishing successful attacks from failed ones (see Fig.5). 4.3 Potential mechanisms Inthissubsection,Iexplorepotentialmechanismsthroughwhichterroristattackscould have affected confidence in national political institutions and political approval of the country’s leader. I provide suggestive evidence that terrorism could exert influence by triggering stronger patriotism and civic engagement among citizens. However, I do not find evidence for the alternative channels of perceived economic capture or political acquisition by the government.19 4.3.1 Patriotism Patriotismisonemechanismthatmaydrivetheriseinpoliticalconfidenceandapproval post-terrorism. This proposed mechanism is inspired by the literature dealing with the rally ’round the flag effect, in which the Patriotism School argues that crisisinduced nationalist sentiment translates into political support for the leadership after (international) crises (Brody and Shapiro 1989). 19 More discussion on alternative explanations for the mechanisms explored in this section can be found in Appendix B4. 123 Rally post-terrorism Page 19 of 29 50 Table 3 Mechanisms of patriotism, civic engagement, economic capture and political acquisition (1) (2) (3) (4) A. Patriotism Move abroad Move within year Post-attack −0.087 (0.012)*** −0.074 (0.019)*** No. obs. 61,718 7333 B. Civic engagement Donation Volunteer Help strangers Voice opinion Post-attack 0.031 (0.016)* 0.039 (0.034) 0.064 (0.030)** −0.041 (0.015)*** No. obs. 59,078 43,650 C. Economic capture Index Conf. police Feel. safe Post-attack 0.017 (0.012) 0.022 (0.015) 0.004 (0.015) No. obs. 59,078 D. Political acquisition Poli. view free. Media free. Post-attack −0.189 (0.011)*** −0.086 (0.016)*** No. obs. 36,451 57,565 Note: Individual covariates contain a male dummy variable, the quadratic of age, dummy variables of completed educational levels, dummy variables of marital status, number of children living inthe household, a dummy variable of being born in the country surveyed and dummy variables of unemployment and out of the labor force. Terrorism covariates include attack-type fixed effects, a dummy variable of logistically international attack, a dummy variable with the value of 1 if the attack is unknown to be logistically international or not and another dummy variable of the major attack. A constant is also included in every model, but not shown for parsimony. Standard errors clustered at the regional level in parentheses; *p< 0.1; **p<0.05; ***p<0.01 In panel A of Table 3, I use unwillingness to move permanently abroad ideally (after a terrorist attack) as one proxy for patriotism.20 Dreher et al. (2011) showed that terrorist incidents boosted skilled emigration due to deterioration in local living and working conditions. If, even in an ideal situation, individuals would be less likely to move abroad after a terrorist attack, it may reflect a strong nationalistically attached sentiment. After attacks, the probability of being willing to permanently move abroad in an ideal situation was significantly reduced by 8.7 percentage points. Among the potential emigrants, the likelihood of planning to leave their country within the following year significantly decreased by 7.4 percentage points post-attacks. These results are consistent with the patriotism mechanism: terrorist incidents happening in a person’s region of residence improved their country attachment and hence decreased their willingness to move abroad and to move soon. Later, as a heterogeneity analysis, I show that political confidence and approval improved two-fold after attacks launched 20 The table note means that the same sample and the same covariates as for the baseline model specification in column 3 of Table 2are used for estimation. 123 50 Page 20 of 29 S. Chen by non-domestic terrorists compared with after attacks launched by domestic perpetrators. This difference is in accord with the idea of the nationalist attachment. 4.3.2 Civic engagement Next, I study the potential channel of civic engagement. After the occurrence of terrorist incidents in their region, residents may be motivated to participate in rebuilding and enhancing their own local community. Needing support from the government and/or cooperating with the government in reconstruction activities, individuals may have greater political confidence and approval. I examine different dimensions of civic engagement: charity donation, volunteer participation in an organization, provision of help to a stranger and opinion voiced to a public official. Panel B of Table 3reports the results. Post-terrorism, the probabilities of charity donation and helping strangers significantly rose by around 3 percentage points and 6.5 percentage points, respectively. However, the likelihood of people voicing their opinion to a public official was significantly reduced by about 4 percentage points. This last result makes sense, in that after a terrorist attack, citizens engage in civic events and cooperate with institutions, and hence are less likely to voice dissenting opinions to public officials.21 The estimates in panel B provide evidence for enhanced civic engagement among residents in adversely affected regions, and thus support the channel of civic engagement. 4.3.3 Economic capture After a terrorist incident occurs, in addition to immediate responses such as rescuing victims and hunting for and arresting perpetrators, the government may take strategic measurestopacifycivilians,especiallyintheimpactedregion,inordertopreservetheir confidence and support. By allocating economic resources to the targeted region and improving (or at least maintaining the pre-terrorist incident) perceived local economic climate, the government will gain approval from civilians affected by terrorism. I term this strategy “economic capture.” The economic capture in this case goes beyond the necessary measures of crisis aid or management, such as reconstruction of the facilities damaged in an attack. InpanelCofTable3,Ipresenttheestimatesforthepossiblemechanismofperceived economic capture by the government. The dependent variable in column 1 is an economiccaptureindexofareabasics.Thisindexisconstructedbasedondummyvariables of satisfaction with a series of elements in the respondents’ area of residence. These elements, or “basics,” include infrastructure, such as roads and highways, the public transportation system, housing availability, the educational system or schools and healthcare availability.22 If the government allocates economic resources to targeted 21 Anotherpiece ofevidenceforlesspolitical dissentfollowingterrorismisillustrated inFig.4:immediately after attacks (i.e., in the first quarter post-terrorism), not only did the average probabilities of political confidence and approval improve, but their standard errors and hence variances also shrank. This suggests that people became more united with political institutions. 22 Details of specific questions about these elements and how the index is created can be found in Appendix C. 123 Rally post-terrorism Page 21 of 29 50 regions, these basics may improve. Lacking an objective measurement for improvement of these elements, the index constructed of satisfaction with basics provides a perceived composite alternative. Column 1 in panel C displays the effects of terrorism on the economic capture index. There was a nonsignificant increase in the index of 0.017 (in a range of 0–1) after a terrorist incident. That is to say, on average, civilians’ satisfaction with the overall area basics improved by around 1.7 percentage points after their region of residence was hit by an attack; however, the improvement was not significant. Thus, the perceived economic capture mechanism does not seem important in explaining the effects of terrorism on confidence in the government and approval of the country leader’s performance. Thegovernmentmayredeploymorepoliceofficerstoimproveprotectionintargeted regions after the occurrence of terrorism (Di Tella and Schargrodsky 2004; Draca et al. 2011), and this could be regarded as part of a reallocation of economic resources. Hence, in columns 2 and 3 of panel C, I examine the influence of terrorism on the dummy variables of having confidence in the police and feeling safe while walking alone at night. However, I find neither significant nor sizable effects on confidence in the police or perceived safety. In this subsection, I do not attempt to argue that a terrorism-besieged government hasnottriedtoeconomicallycapturepoliticalsupportorimprovethesecuritymeasures in targeted areas following terrorism. Instead, I document that citizens in impacted regions did not perceive significant improvements in these community basics. Considering the subjective and perceived characteristics of political confidence and approval, I do not find evidence that the increased levels of political trust and support postterrorism can be attributed to the perceived reallocation of economic resources to the relevant areas. 4.3.4 Political acquisition Along with economic resource allocation, a government may extend the franchise to acquire local political support after terrorist attacks. In authoritarian regimes, democratization can become a concessionary option under the threat of widespread unrest (Acemoglu and Robinson 2000b; Aidt and Franck 2015). Nevertheless, the room for extending the franchise is limited in advanced democracies such as the EU. Thus, I examine the changes in the (perceived) freedom of expressing political views and the (perceived) freedom of the media, rather than changes related to voting rights. In the last panel, the negative coefficient estimates of Post-attack are significant for both (perceived) freedom indicators. If the government’s strategy of franchise extension worked, we would see a significant rise in these measures of political freedom post-terrorism. Hence, political acquisition does not seem to be a mechanism.23 The significant negative impacts of terrorism occurrences on perceived freedom could 23 One potential concern is that if implemented, the franchise would be extended at the national level rather than at the regional level. Thus, I replicate this analysis at the country level and obtain a marginally significant negative effect (−0.048) on the freedom of expressing political views, and a nonsignificant effect (0.026) on the freedom of the media. 123 50 Page 22 of 29 S. Chen be interpreted as either the restrictions imposed by the authorities (especially in lessliberal EU member states) or the fears of future revenge and further attacks in the affected regions. Neither of the explanations validates the channel of political acquisition. Therefore, the exploration of potential mechanisms suggests that patriotism and civicengagementofciviliansintargetedregionswerelikelytobuildlinksfromterrorist incidents to political confidence and approval. Nonetheless, I do not find evidence that the increase in political trust and support after terrorism was due to the perceived allocation of economic resources or perceived political acquisition by the government. 4.4 Heterogeneity investigation In this subsection, I examine heterogeneous effects in terms of various individual characteristics such as gender, employment status, educational attainment and age cohort and terrorism characteristics including logistics and geography/institution within the EU. The aim is to explore which specific subgroups of people and countries and which classes of attacks drove this post-terrorism rally effect. For ease of comparison, in panel A of Table 4, I reproduce the baseline estimates from column 3 of Table 2. First, I explore whether there are differentials between men and women. Perrin and Smolek (2009) documented a stronger rally phenomenon among men than among women; however, my estimates in panel B do not accord with their findings. Consistently, increases in both political confidence and approval after terrorist incidents were significant for both genders and of a slightly larger magnitude among women than men. In the last row of the panel, I test whether the gender difference is significant or not. The corresponding p-values indicate a nonsignificant gap between men and women in the rally effect.24 Second, terrorism adversely affects business conditions and leads to job losses (Brodeur 2018). People laid off after a terrorist incident may blame the government for a lack of capacity to deal with the crisis, and hence their own unemployment. Nonetheless, they may also be angry with terrorists because of their own job loss, and thus support the government’s counterterrorism measures. The results in panel C seem to support the second hypothesis. Both the employed and the non-employed present significant rally syndrome after terrorism. Among the non-employed, improvements in political confidence and approval were more sizable than those of the employed. However, such a divergence in magnitude is onlysignificantforapprovalofthecountry leader’s performance, as shown by the p-value of the difference test. Third, in panel D of Table 4, I examine the differences between college degree holders and individuals without such a degree. Surprisingly, the effects of terrorism on political confidence and approval for college degree holders are significantly negative 24 Heterogeneous effects in terms of gender are significant for neither political confidence nor political approval, and heterogeneous effects regarding labor market status and age cohort are only partially significant. This might not necessarily exclude the possibility of heterogeneities concerning these characteristics. Statistical non-significance of these heterogeneities could also be due to fewer observations with terrorism exposure after sample divisions and hence decreased variations that can be leveraged to compare estimated effects between sub-samples. Unfortunately, it seems infeasible to verify this possibility by increasing the proportion of observations with terrorism exposure. 123 Rally post-terrorism Page 23 of 29 50 Table 4 Effects of terrorism on political confidence and approval–heterogeneity investigation (1) Confidence govt. (2) Approval leader No. obs A. Baseline 0.110 (0.013)*** 0.134 (0.021)*** 61,718 B. Gender Men (βm) 0.108 (0.024)*** 0.140 (0.025)*** 26,801 Women (βw) 0.125 (0.022)*** 0.155 (0.022)*** 34,917 pvalue (βm=βw) 0.677 0.438 C. Labor market status Employed (βep) 0.106 (0.023)*** 0.107 (0.017)*** 33,635 Non-employed (βnep) 0.125 (0.025)*** 0.211 (0.036)*** 28,083 pvalue (βep =βnep) 0.645 0.000*** D. Educational attainment College degree (βcg) -0.247 (0.035)*** -0.300 (0.027)*** 15,930 No college degree (βncg) 0.135 (0.014)*** 0.165 (0.022)*** 45,788 pvalue (βcg =βncg) 0.000*** 0.000*** E. Age cohort 50 & younger (β50−) 0.082 (0.022)*** 0.123 (0.018)*** 32,636 Older than 50 (β50+) 0.270 (0.052)*** 0.187 (0.054)*** 29,082 pvalue (β50−=β50+) 0.006*** 0.160 F. Logistics Domestic (βdm) 0.111 (0.013)*** 0.135 (0.021)*** 60,742 Non-domestic (βndm) 0.265 (0.039)*** 0.246 (0.030)*** 61,563 pvalue (βdm =βndm) 0.000*** 0.002*** G. Geography/institution Central and Eastern Europe (βceu) 0.184 (0.022)*** 0.072 (0.039)* 19,782 Western Europe (βweu) 0.114 (0.012)*** 0.138 (0.022)*** 41,936 pvalue (βceu =βweu) 0.030** 0.016** Note: Only the coefficients of Post-attack are presented. Individual covariates contain a male dummy variable, the quadratic of age, dummy variables of completed educational levels, dummy variables of marital status, number of children living in the household, a dummy variable of being born in the country surveyed and dummy variables of unemployment and out of the labor force. Terrorism covariates include attack-type fixed effects, a dummy variable of logistically international attack, a dummy variable with the value of 1 if the attack is unknown to be logistically international or not and another dummy variable of the major attack. A constant is also included in every model, but not shown for parsimony. Standard errors clustered at the regional level in parentheses; *p<0.1; **p<0.05; ***p<0.01 and very large. Apparently, the rally phenomenon after terrorism was driven by lower educated people without a college degree while such political trust and support among higher educated individuals declined substantially. The difference tests in the last row of panel D confirm such a significant divergence. Furthermore, I separate the sample into two age groups: one group of 50 years old and below, and the other older than 50. For both outcome variables, the coefficients are significant for the two groups. However, the terrorism-induced rally syndrome was 123 50 Page 24 of 29 S. Chen more sizable among the older cohort than the younger one. The p-values show that such age cohort differentials are significant for confidence in the national government, especially. In addition to individual characteristics, I also examine heterogeneity regarding terrorism logistics. In panel F, I distinguish attacks with domestic logistics from those with international or unknown logistics. The coefficients are significant for both types of attacks, but roughly twice as large for non-domestic terrorism as for domestic terrorism. Hence, residents in affected regions seemed to exhibit a considerably stronger rally syndrome after attacks that were launched by non-domestic perpetrators. This substantial gap is further confirmed by the test conducted in the last row. Such a difference is consistent with the patriotism mechanism: people increased their support for the government and approval of their country leader’s performance, particularly when attacked by foreign terrorist groups. Finally, I investigate differentials between EU member states in Western Europe, and those in Central and Eastern Europe, as the latter are traditionally perceived as less liberal or less maturely democratic.25 The estimates in panel G show that confidence in the national government improved following terrorist attacks more notably in Central and Eastern Europe, while approval of the country leader’s performance in Western Europe increased almost twice as much as in Central and Eastern Europe. The tests in the last row confirm such significant geographic and institutional gaps. These results imply that the rally effects seemed to be exhibited more notably in the form of political institutions—the national government in the current context—than the individual leadership in Central and Eastern Europe. Nevertheless, the rally phenomenon was significantly present regarding both political institutions and individual leadership in Western Europe. In this subsection, I study heterogeneous rally effects in different subgroups and types of terrorism. I find that after attacks, political confidence and approval increased more substantially among the non-employed, individuals without a college degree and older people. However, political confidence and approval reduced significantly, both economically and statistically, among college degree holders. Moreover, attacks launchedby domestic perpetrators hadonlyhalftheextentof political effectsasattacks launched by non-domestic terrorists. This may reflect a nationalistically attached sentiment; a mechanism of the rally syndrome. The rally phenomenon was exhibited significantly in terms of both political institutions and individual leadership in Western Europe, while in Central and Eastern Europe, it was presented more strongly in the form of political institutions. 4.5 Sensitivity analyses To check the robustness of my estimates, I conduct a range of sensitivity analyses. I extend the post-terrorism window period to explore the persistence of the rally effect. I also include country×year fixed effects and region-specific time trends. 25 According to the definition from the OECD, Central and Eastern European member states of the EU include Bulgaria, Croatia, the Czech Republic, Hungary, Poland, Romania, the Slovak Republic, Slovenia and the three Baltic States: Estonia, Latvia and Lithuania. 123