Everybody s a Victim? Global Terror and Well-Being

Transcription

1 Everybody s a Victim? Global Terror and Well-Being Alpaslan Akay, Olivier Bargain, Ahmed Elsayed March 2018 Abstract Terror has become a global issue. Terror acts perpetuated by religious, nationalist or political groups at any corner of the globe can propagate rapidly through the media or through economic channels (such as stock markets). The cumulated welfare loss experienced by world citizens may be large. We suggest an original evaluation of this cost by estimating the e ect of the 84; 000 terror events that took place over in the world on the well-being of citizens in six Western countries (representing around 640 million people). We combine large panel datasets for Germany, the UK, Switzerland, Russia, Australian and the US. Individual well-being information for 750; 000 time-individual observations are recorded on precise dates that are matched with daily information about terror events worldwide. We nd a signi cantly negative and internationally stable e ect of terror intensity on individual well-being. People recover from terror events quickly but the cumulated cost is high: about 2% of GDP. We provide suggestive evidence that the main channel is fear, and discuss the subsequent political consequences. We nd substantial heterogeneity depending on the level of exposure to the event proxied by the variation in media use or by the physical and genetic distance to the victims. Key Words : Terror, Subjective Well-Being, Adaptation, Media. JEL Classi cation : C35, C90, D60 Acknowledgements: Akay is a liated with the University of Gothenburg and IZA, Bargain with University of Bordeaux, IZA and the Institut Universitaire de France, Elsayed with IZA. Corresponding author: Olivier Bargain, LAREFI, Bordeaux Université, rue Léon Duguit, Pessac, France, olivier.bargain@u-bordeaux.fr

2 1 Introduction Intentionally and indiscriminated acts of violence have been perpetuated by individuals, groups or organisations since ancient times as a means to create fear to achieve a political, religious or ideological aim. Traditionally, the material and human costs of religious fanatism, political radicalism or independentist movements have be born by local populations and states. In the recent years, however, terror has become a global issue. Violent acts are increasingly transnational in the sense that they rely on funding and infrastructures from di erent countries, pursue international objectives or target groups beyond national borders. In an integrated world economy, terror events also reverberate through economic channels, chie y through stocks (Chen and Siems, 2004, Straetmans et al., 2008) and commodity prices (Guidolin and La Ferrara, 2010). New information and communication technologies also make that a local event can reach people worldwide. Researchers have commented on the symbiotic relationship between mass media and terrorism (Rohner and Frey, 2007), especially now that the audience is global and that terror groups seek international exposure to exert pressure. Social media have accelerated this trend and are increasingly used as a way to propagate fear at a very broad scale. The "theater of terror", as a modern psychological warfare, tends to reach far beyond the immediate victims of terrorism (Weimann, 2005). Against this background, a comprehensive evaluation of the global impact of terror on human welfare is required. To date, most economic evaluations focus on local economic consequences. 1 A few studies go further by assessing the social and psychological impact of terror using selfreported measures of well-being. 2 In equivalent income terms, they nd that the loss usually exceeds the purely economic consequences of terror. Yet, the bulk of the evidence focuses on the local experience of terror rather than on global welfare implications. An exception is the study of Metcalfe et al. (2011) who estimate the impact of 9/11 on mental health in the UK. 3 Other 1 Valiño et al. (2010) survey the evaluations of direct and indirect costs. Speci c studies assess the impact of terror on growth (Blomberg et al, 2004, Crain and Crain, 2006) and scal revenue and policy (Gupta et al., 2004, Chernick and Haughwout, 2006), particularly in conjunction with its e ect on investment (Fielding, 2003), rms pro ts and their business prospects (Frey et al., 2007) or rms location and town planning (Glaeser and Shapiro, 2002). Long-term impacts on health as measured by birth weight are studied in Camacho (2008). 2 For instance, Frey et al. (2009) study the impact of terror events in France and Northern Ireland. Romanov et al. (2012) explore the e ect of Palestinian attacks on the well-being of Israelis in the early 2000s. Clark et al. (2017) investigate the impact of the 2013 Boston marathon bombing on Americans well-being. In psychology and medical sciences, many studies have explored the local impact of attacks like 9/11 in New York (Silver et al., 2012, Galea et al, 2002, Schlenger et al, 2002), July 2005 terrorist attacks in London (Rubin et al., 2005) or March 2004 attacks in Madrid (Salguero et al., 2011). Some of these studies also address the intangible e ects felt elsewhere in the country (Silver et al, 2002, Schlenger et al, 2002, Krueger, 2007). 3 Evidence also stems from macro regressions in Vorsina et al. (2015) and Farzanegan et al. (2016), using average self-reported levels of life satisfaction by country and year from the World Happiness Database ( ) and the World Values Survey ( ), respectively. We argue that our approach based on daily match 1

3 studies have also shown that terror attacks have an impact on fear and public opinion, even if the terror attacks happen far away for instance Finseraas and Listhaug (2013) on the 2008 terror attacks in Mumbai, India. Rather than focusing particular events, how big they may be, we suggest exploiting the multitude of daily events that happen in the world over a long period of time. In this way, we estimate how the global tension of terror may a ect the well-being of world citizens. This welfare impact is not only important per se, as motivated above, but may in turn change people s behavior on many accounts including voting choices, public opinion (on defense or immigration policies), charity giving, medication, criminal activity or fertility. 4 Precisely, we rely on a bit more than 84; 000 terrors events that took place over in the world. We match information on these events with subjective well-being data (SWB hereafter) from citizens of six Western countries, representing around 640 million people. For that purpose, we assemble six large panel datasets from Germany, the UK, Switzerland, the Russian Federation, Australia and the US. 5 While these panels have been extensively used individually, pooling them into a large international longitudinal dataset is unique. The total sample comprises almost 750; 000 individual-time observations. They all contain a common SWB indicator of live satisfaction, commonly used as a proxy for utility (Kahneman and Sugden, 2005). Terror events are recorded daily in the Global Terror Dataset (GTD hereafter), which we match to our six-country sample using interview days. We regress life satisfaction on a detailed set of controls and on several measures of terror, notably the total incidence of terror events and the total number of subsequent fatalities each day. Thus, we identify the impact of global terror on Western citizens from the daily variation in terror intensity occurring in the world. If the time of interview can be treated as random, comparing the well-being of individuals in our sample before and after the events occurring on a certain day, and comparing this to their well-being a year before, provides a convincing quasi-experiment. Two important aspects matter for inferrence. The rst one is that terror events can be treated as exogenous shocks. Studies that assess the impact of terror events on individual well-being locally may su er from the fact that confounding factors generate both between events and welfare response is more precise. Note also that some studies address the global implications of terror through its e ect on trade (Blomberg and Hess, 2006, Egger and Gassebner, 2005), FDI (Enders and Sandler, 1996, Abadie and Gardeazabal, 2008) and speci c sectors like tourism (Enders et al., 1992, Fleischer and Buccola, 2002), transportation companies (Drakos, 2004, Blunk et al., 2006), insurances (Cummins et al., 2003, Brown et al., 2004) and stock markets (Karolyi and Martell, 2010, Abadie and Gardeazabal, 2003). 4 Evidence has been gather on several of these outcomes, especially when looking at the Israeli-Palestinian con icts (Gould and Klor, 2010, Berrebi and Klor, 2006, 2010, Berrebi and Yonah, 2016), other local terror events in Turkey (Kibris, 2011) or Madrid (Montalvo, 2011), or variation in the timing of terror events and aggregate political measures (Berrebi and Oswald, 2015, Jones and Olken, 2009). 5 Our baseline exclude immigrants and focus only on the native population. Further work could address speci c issues related to migrants. 2

4 terror events and depressed social and economic conditions a ecting well-being. 6 In contrast, we mainly consider the e ect of events taking place abroad, so we can more easily assume that there are no omitted variables codetermining terror and local well-being. The second aspect is that individuals are interviewed on a day that is uncorrelated with the day when terror events occurred. Shock waves of events like 9/11 are global, which may a ect response rate of interviews planned to take place the following days. Yet this does apply to only a very few of the 84; 000 terror events used for identi cation. Note that our repeated di erence-in-di ernce estimations also control for individual xed e ects. Results can be summarized as follows. Fixed-e ects model speci cations suggest a negative, highly statistically signi cant, and sizeable global e ect terror related fatalities and terror incidences on the well-being of individuals. The relative size of fatalities and incidences are highly similar to each other. The e ect shows a striking regularity across countries with respect to size and signi cance of the e ect. One striking result is people quickly adapt to negative shocks of terror. Adaptation is full and takes about a week (5 to 7) days to bounce back to the level prior to the terror event. The adaptation speed for fatalities is 1-3 days slower than that of incidences. There is a substantial monetary global cost of terror. Using marginal rate of substitution between terror measures and household income our baseline xed-e ect models suggest that the per capita cost of terror for a year is 30.4$ for fatalities and 254.4$ for incidences (83.9$ for fatalities and 733.6$ for incidences without Russia Federation). We interpret the strong negative e ect of global scale terror as fear and anxiety, which negatively relate to well-being (Becker and Rubinstein, 2011). The fact is the odds of experiencing a terror attack in the regions under investigation or get hurt due to terrors occurring in distant regions is very low. The results suggests that people might be overreacting or even have irrational fears towards terror. Using alterative sets of proxies we investigate a large set of mediating factors pertaining to the relationship between global terror and well-being. We mainly aim to disentangle actual threat or reasons for fear from terror from irrational fear. People might have beliefs and attitudes supporting their reasoning (e.g., fears and worries). They might perceive a real threat due to distance and exposure (e.g., physical and genetic distance, media and internet). Immigration (e.g., Muslim immigrants) and other local characteristics (e.g., living in populated areas) in the immediate living environment might trigger overreactions to terror. An alternative explanation of the negative e ect of terror might not be the fear but economic damages or other natural and terror related socio-political global shocks (e.g., stock markets or natural disasters). Our result favors the interpretation of fear which is supported by the actual threats or even irrational reactions towards the low odd of negative events. Fatalities and 6 Note that we shall also consider local events occurring in the six countries under study, yet this subset of results will be less precise (such events are rare in these countries) and robust (due to the risk of endogeneity just given). 3

5 incidences that are physically and culturally closer (measured using genetic distance to common ancestors among people living in the terror country) generate larger impact consistent the idea that people perceive and react to actual threat. Yet the reaction is mostly exaggerated. We observe a similar overreaction among people who are living with a larger share of immigrants, particularly from Muslim countries. The natural disasters or other shocks does not explain the result. We evaluate the speed of adaptation and explore possible channels including fear, compassion, stock markets and other types of investments. We also investigate heterogeneous e ects, depending notably on physical and genetic distance to the victims and media use. The paper extensively discusses the behavioral and political implications of this seemingly irrational fear from terror. First, the rise in global terror intensity over the past ten years has not only a ect individual welfare but, through this, way have also triggered change in behavior (like extreme party and nationalist voting). Further work could match date and results of actual votes with global terror events, or match voters intentions (for instance using World Value Surveys) and global terror intensity. To check whether groups of people most a ected by global terror in terms of SWB are also those who consequently respond in terms of vote radicalisation, one could create cells (by age, rural/urban, etc.) to check the correlates in both SWB and voting behavior responses to the treatment. Second, the extent to which this fear adds up to the social and political consequence of the 2008 economic crisis is also a question for further research. Finally, an attempt should be made to distinguish the increased terror intensity of the past ten years from the increased exposure through the media, notably with the accelerated use of social media worldwide. 7 In the next section, we present the data construction, a global portrait of terror and the econometric speci cations. We discuss the results in section 4, overall or by country. We report the global cost of terror and discuss the role of accumulation and adaptation. Section 5 present results from a large set of channels mediating the relationship between terror and well-being. Section 5 concludes on the economic and political implications of the results. 2 Empirical Approach 2.1 An International Panel of Individual Well-Being Data Selection. The rst challenge is to construct a panel of individuals covering several countries and with comparable well-being measures and determinants. We focus on a relatively homogeneous group of countries, namely six rich countries from four continents (Eu- 7 Several studies have analyzed the fact that media di use fear, hatred, sympathy, etc., and hence change behavior such as radical voting (Della-Vigna et al., 2014), ethnic violence (Yanagizawa-Drott, 2014) or social capital and trust (Olken, 2009). 4

6 rope/eurasia, America, Oceania): Germany, the UK, Switzerland and the Russia Federation, the US and Australia. This choice is also guided by data availability: we combine the six household panels available in the world that contain SWB information at the individual level and over a relatively long period. These datasets and the years when SWB information is available are as follows: the German Socio-Economic Panel (GSOEP, ), the British Household Panel Survey (BHPS, except 2001), the Swiss Household Panel (SHP, ), the Russian Labor Monitoring Survey (RMLS, ), the Australian Household Income Dynamics (HILDA, ) and the US Panel Study of Income Dynamics (PSID, 2009, 2011, 2013). 8 These datasets also contain a wealth of individual characteristics that can be used as controls in SWB equations, some of which are easily made comparable across countries and over time (e.g., age, gender, marital status, family size). More di cult cases are treated below. Datasets are collected at di erent periods over (the longest panel is the GSOEP, starting in 1984). Our choice to focus on is determined by data constraints and a relatively homogeneous treatment of all countries under study notably the fact that key variables are missing in panels for some years and that the terror data, presented below, is not available for the year The nal sample is unbalanced. Some countries o er a better coverage than others: for instance, all the relevant information is available in the GSOEP over the period studied but SWB information is present in the PSID only in waves 2009, 2011, and Sensitivity analyses will show that this is not detrimental for our results in particular, results will not re ect the response to terror of a particular country. Sample selection is applied uniformly to all countries. We focus on individuals aged between 18 and 75. Our baseline excludes rst-generation migrants: introducing them would bring in confounding factors regarding terror events possibly happening in their home country. For panel estimations, we keep individuals that are interviewed more than once in the panel. In the six datasets, each person is interview only once in the year: the precise date of interview is recorded and used to match terror information, as described below. Measures of Well-Being. Self-reported measures of well-being, be it life satisfaction or mental health, provide new insights on phenomena that are di cult to apprehend with the traditional revealed preference approach (see Clark et al., 2008, and Senik, 2004, for enlightening surveys). A rapidly growing amount of evidence collected by economists and psychologists over the recent years has shown that SWB is not a pure statistical noise and can be validated in 8 These datasets have been used extensively in SWB studies, for instance in Ferrer-i-Carbonell and Frijters (2004, GSOEP), Clark and Oswald (1994, BHPS), Stutzer (2004, SHP), Senik (2004, RMLS), Feddersen et al. (2016, HILDA) or Brown et al. (2017, PSID). See Powdthavee (2015) for a clear overview of existing SWB datasets and their use. 5

7 numerous ways (Kahneman and Krueger, 2006), notably against behavior and more objective measures of well-being (Krueger and Schkade, 2008, Oswald and Wu, 2010). 9 Nonetheless, we keep in mind the possible lack of interpersonal comparability due to individual heterogeneity in self-perception about one s situation (Decanq et al., 2015). We treat this issue as a measurement error, namely by using large samples and accounting for individual xed e ects as a control for time-invariant heterogeneity in perception and personality. The SWB measure used in this study is based on the life satisfaction question, which is usually highly correlated with other subjective measures of well-being like self-reported happiness or composite indices of mental health (Clark and Oswald, 1994). The life-satisfaction question has the advantage to be present and relatively comparable across the di erent datasets at use. 10 The answer is reported on di erent scales: 5 points in the RMLS and PSID, 7 in the BHPS and 10 in the other datasets. 11 Hence, it is necessary to harmonize the scales across datasets. Since a majority of countries have an 11-point scale, our baseline approach consists in expanding the life-satisfaction answers in the PSID, the RLMS and the BHPS to 11 points. 12 An alternative approach, presented in the our sensitivity checks below, consists in collapsing answers to the least common denominator, i.e. the 5-point scale used in the RMLS and the PSID. 13 The mean level of SWB for each country is reported in Table A.1, in their original scale ( rst row), in the 0-10 harmonized scale (second row) and in the 1-5 harmonized scale (third row). The country ordering based on mean values is consistent across scales, with the highest score in Switzerland and the lowest in Russia. 9 Various aspects are investigated using SWB, such as the in ation-unemployment trade-o (Di Tella et al., 2001), the e ect of climate change (Frijters and van Praag, 1998), of air quality (Levinson,2012) or natural events (Kimball et al., 2006), or the procedural utiliy from decentralization (Fleche, 2017). 10 In the GSOEP, SHP and HILDA, the question is framed as: "How satis ed are you with your life as a whole, all things considered?". In the BHPS: "How dissatis ed or satis ed are you with your life overall?". In the RMLS: "To what extend are you satis ed with your life in general at the present time ". In the PSID: "Please think about your life-as-a-whole. How satis ed are you with it?". 11 In ascending order of satisfaction, answers are scaled as follows: from 0 ("not at all satis ed") to 4 ("fully satis ed") in the RMLS, from 5 ("not at all satis ed") to 1 ("completely satis ed") in the PSID, from 1 ("completely dissatis ed") to 7 ("completely satis ed") in the BHPS, from 0 ("completely dissatis ed") to 10 ("completely satis ed") in the GSOEP, SHP and HILDA. 12 For each individual in the PSID and RLMS, we draw a random discrete number in the intervals [0-2], [3-4], [5-6], [7-8], [9-10] for ordinal values 1 to 5 respectively (and similarly in 7 intervals in the BHPS). The best approach consists in bootstrapping estimations over a large number of such draws yet this gave indi erentiable results compare to a single draw, which we use for most of our analysis. We have also tried di erent reasonable assignments rules for the de nition of segments, again without much di erence with baseline results. Results with bootstrapped estimations and alternative de nitions are available from the authors. 13 That is, we assign values 1 to 5 to the intervals [0-2], [3-4], [5-6], [7-8] and [9-10] respectively, in the GSOEP, SHP and HILDA, and to intervals [0-1], [2-3], [4-4], [5-5], [6-7] respectively, in the BHPS. 6

8 Common Determinants of Well-Being. Another challenge pertains to the comparability of control variables across samples. Our discussion focuses on the main determinants of wellbeing, such as income, employment and health (Clark et al., 2008, Senik, 2005). Income is de ned in all surveys as the sum of all sources of income in the household after governmental transfers. We simply convert household income into 2011 USD using the CPIs and exchange rates taken from the World Bank Indicators. The employment status is a dummy variable which indicates people who are currently employed. The most common de nition of health is the self-assessed heath variable, rescaled in an ascending order (from 1, very poor, to 5, very good). Self-assessed heath is widely used in health economics, and despite being subjective, it has been shown to predict disability, chronic diseases and health care utilisation (Jusot et al., 2013). We control for the number of years of schooling, as provided for in the GSOEP, SHP, HILDA and PSID; we had to reconstruct this variable for the UK and Russia using the highest education attended and information about education systems. Finally, we account for region dummies, which are country-speci c by de nition, but not necessarily captured by individual xed e ects due to the fact that some people are geographically mobile over the course of the panel. 14 Table A.1 reports statistics for the key variables the main determinants of SWB, income and health, are consistent with the SWB ranking across countries, with Switzerland and Russia at both ends of the distribution. 2.2 Global Terror Data We use the Global Terror Database (GTD), a unique open-source dataset collected by the National Consortium for the Study of Terrorism and Responses to Terrorism. It provides comprehensive information on terror events in the world since 1970 until today (see LaFree and Dugan, 2007, for more information). There are some exceptions, including year 1993 that is missing. 15 The GTD applies a consistent de nition to the acts of terror: it de nes a terrorist attack 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" We use the 16 Federal States (Länder) of Germany, the 13 regions in the UK (9 English regions, Wales, Scotland, Northern Ireland), the 26 Cantons in Switzerland, a detailed classi cation of 13 Australian regions (constructed using information on state or territory of residence and population density), 40 political regions in Russia, and the 50 Federal States in the US. 15 Note that for some countries, we can conduct a much longer analysis. For Germany, notably, we can match terror data to the GSOEP for (excepted 1993), using health satisfaction rather than self-assessed health, which is missing before The results are very similar to the baseline estimations using the German selection. 16 To consider an incident for inclusion in the GTD, the incident must be intentional, must entail some level of violence or threat of violence (including property violence, as well as violence against people), and the perpetrators of the incidents must be sub-national actors (the database does not include acts of state terrorism). Moreover, terror events are de ned using three non-mutually exclusive criteria. First, terror events should have 7

9 It provides a complete set of information including the nature of the event, its location, the number of persons wounded or killed, the type of attack (e.g., hijacking, bombing), the type of the target (e.g., embassy, civilians), the purpose of the activity (e.g., political, religious). Figure 2.2 displays the number of terror events and casualties per day. Figure 2.2 shows a distribution summary of both events and casualties (both gures are capped at 50 for clarity). These graphs indicate that there are several terror events occurring almost every day around the globe. No event happened in only in 3:7% of the 7300 days beween January and December , and the average number of events per day was 9:6. There were no killings in only 10:2% of these days and the average casuality number was 22:5. These gures are striking: there is truly a high frequency of global terror, which may be perceived by world citizens through various media sources or other channels. This is what we shall expore and we will focus on the intensity of terror (number of events and casualties) more than on the occurrence of terror somewhere in the world on a particular day. Figure 2.2 point to a signi cant increase in terror intensity since the mid-2000s. The portrait of terror is completed by Figure A.1 and Table A.2 in the Appendix. The gure shows the geographical distribution of terror intensity: terror incidences and fatalities are widespread around the world for the period under study. The table reports the total number of event, the daily average fatalities/incidences and the breakdown by main characteristics of attack. There were 84; 367 events in total over the period, with an average of 18:1 daily events (and 34:2 casualities) per country year. The main medium of attack is explosion-bombing (49% of all events), responsible for 34% of all casualties. 17 Extended events (lasting more than 24 hours) represent a minority (4%). The table also provides these statistics by years and broad regions of the world. Figure A2 depicts the time trend, showing a relative decline in terror incidences and fatalities in the early 2000s (with exceptions like 9/11) and sharply increased after The table completes the picture by showing a rising use of explosion-bombing (representing almost half of all attacks in the last years) and an increased contribution of suicide attacks to the number of fatalities. Most terror events are located in East/Central Asia, the Middle East and North Africa (MENA) countries (58% of all events). Daily numbers of events and fatalities are also higher in these regions. political, economic, religious or social motives but exclude acts motivated by individual pro t or unrelated to broader societal change. Secondly, events are recorded as terror activity if they have the intention to coerce, intimidate or publicize to larger audience and, thirdly, if they stand outside international humanitarian law as re ected in the Additional Protocol to the Geneva Conventions of August The other attack types, not reported, are armed/unarmed assaults (26%), assassinations (11%), facility/infrastructure attacks (6%), hostage taking/kidnapping (6%), and others including highjacking (< 2%). 8

10 Daily Events and Casualties over

11 Distribution of Events and Casualties over the 7305 Days of Empirical Strategy and Identi cation For each (person date of interview) observation in the international panel dataset, we merge individual data with information on the terror events that occurred the previous day in the world, given the time zone where the person is located. Identi cation is based on short-term uctuations in well-being and terror intensity. It is similar to the valuation of time-varying local public good such as air quality, as in Levinson (2012) who match self-reported happiness with air pollution and weather information on the date and place of survey. In our case, we value a global bad, terror, assuming that the SWB recorded on day d in our panel is a ected with a one-day lag by terror events. We will investigate the sensitivity of our results to alternative timing assumptions. The speci cation of the empirical goes as follows. We estimate the life satisfaction of individual i recorded on year t and date d (day and month) as: W itd = X itd + T td 1 + t + d + ' i + C td 1 + " itd : (1) The latent well-being Witd is considered as a proxy for the unobserved welfare of a person, for which we observe an ordinal metric W itd = j on an ordered scale of life satisfaction categories j = 1; :::; J. The model combines both individual characteristics of individual i at the time of observation (year t, day d), X itd, and the intensity of terror the day before, T itd 1. As discussed, we principally focus on two de nitions of T itd 1, namely the global numbers of terror incidences and fatalities on day d 1. The treatment e ect is our key variable of interest, which we estimate in separate regressions for fatalities and incidences. 18 include a vector of characteristics describing the main event C td In sensitivity analyses, we also 1, as explained later. Individual time-varying variables in X itd include usual determinants of SWB, as discussed above, namely marital status, family characteristics (household size and number of children), log household income, work status, self-assessed health, education and region (age and gender are picked by xed e ects). We also control for year dummies t (to pick up the e ect of global shocks on well-being that are common to all countries in each year) and month dummies d (to capture the fact that certain months are under-represented in our panel due to seasonality in interview periods). In alternative regression, we interact these year and seasonal trends with the person s country dummy. Finally, we include unobserved individual e ects ' i, modelled as individual xed e ects in our panel estimations. Our baseline estimations hence consist of 18 These two variables are highly correlated, which prevents the identi cation of their di erentiated impact. Note also that other variables, like the number of serious injuries, will be used as alternative de nitions of terror intensity. 10

12 linear estimations treating j as a continuous variable in order to ease the inclusion of xed e ects. 19 Individual e ects are not only important to account for the heterogeneity that a ects answers to the well-being question, as argued before, but they may also be correlated with timeinvariant individual characteristics related to terror (risk aversion or propensions to experience fear or compassion, for instance). Cross-section variation in terror intensity allows identifying the e ect, for instance in the following example with two persons i and i 0 observed two days apart in year t: W itd W i 0 td 2 = (X itd X i 0 td 2) + (T error td 1 T error td 3 ) + (' i ' i 0) + (" itd " i 0 td 2): We avail of panel information so that time variation allows purging estimations from individual xed e ects: 20 W itd W it 0 d 0 = (X itd X it 0 d 0) + (T error td 1 T error t 0 d 0 1) + ( d d 0) + (" itd " it 0 d 0); and both "between" and "within" variations can actually be combined for di erence-in-di erence identi cation. In our example with i and i 0, subject to di erent terror intensities T error td 1 and T error td 3, taking the time di erence between years t and t 1 and assuming no terror events happened around the interview dates in t 1: W itd W i 0 td 2 = (X itd X i 0 td 2) + (T error td 1 T error td 3 ) + (" itd " i 0 td 2): Contrary to Metcalfe et al. (2011) who focus on two years (2000 and 2001) and days around 9=11, we exploit repeated double di erences by using daily variation in the intensity of terror driven by 84; 367 events over over 20 years. This identi cation hinges on the assumption that the timing of interviews is uncorrelated with the timing of terror. We do not require that terror occurs randomy over time, 21 even though the data shows a relatively uniform distribution of terror event throughout the year (see the rst column of Table A.3 in the Appendix). What we must assume is that individuals are not 19 Previous studies show no appreciable di erence between estimating SWB models with linear or latent dependent variable speci cation (Ferrer-i-Carbonell and Frijters, 2004). Yet, we will provide checks where we acknowledge the ordinal nature of the dependent variable, allowing for individual e ects in the nonlinear context of ordered discrete models model by using quasi- xed e ects à la Mundlak in ordered probits or the "Blow-Up and Cluster" xed-e ects ordered logit estimator of Baetschmann et al. (2015). 20 An example of similar cross-sectional identi cation with exogenous events is given by Levinson (2012). Additionally using panel information to control for individual e ects is done by Metcalfe et al. (2011) for one event (9/11) and one impacted country (the UK). 21 This may not be the case for some events. Some violent acts may occur on days that actually coincide with other newsworthy (and anticipated) events in order to minimize news coverage (Durante and Zhuravskaya 2018). Some others are, on the contrary, planned in order to maximize the coverage of terror in the media of target countries (Weimann, 2005). On the non-random timing of terror, see also Pape (2003). 11

13 selected into particular interview days in relation with terror events at least not with the intensity of terror as measure it. This threat to identi cation may apply when investigating terror events that happened in the interviewees country we turn back to this when looking at this speci c case. Yet, our main estimations concern the e ect of global terror. It is very unlikely that the decisions to conduct an interview (by the statistical institutes or their agents) or the decision to participate (by the interviewees) are in uenced by events taking place beyond borders. This might be the case for some very speci c world events like 9/11, but not for the rest of the 84; 367 events recorded in our data. Hence, robustness checks will focus on alternative estimations excluding events such as 9/11 or events happening within the country of interviewed individual. Note also that xed e ects should capture much of the time-invariant speci cities of these persons. Two other criterion should a ect the precision and magnitude of the e ect. Regarding the magnitude, individuals should be aware of terror events through the media or other sources of information. We will investigate treatment heterogeneity with respect to media use and to di erent types of distance that may alter awareness and the acuty of the e ect. Yet, our baseline estimates will necessarily capture the extent to which the six countries under study are on average a ected by global daily terror. This average e ect may not be zero if these countries leave connected with the rest of the world, which is likely and worth the conjecture. This average e ect, combined with the mean e ect of income on SWB, should readily provide an equivalent income valuation of terror as a global bad. Regarding the obtention of precise estimates, a su cient number of observations is required before and after the event, as in the case of 9/11 in Metcalfe et al. (2011). In our setting, however, we bene t from an extremely large number of events and of much variation in the daily intensity of terror. Nonetheless, we would miss the e ect if interviews are not spread enough over the year, especially if terror events also show some seasonality that mismatch interview periods. As point out above, Table A.3 shows a relatively uniform distribution of terror events over the year ( rst column). Besides, there is a fairly large overlap with the months during which interviews took place (second column). This is all that matter for us. Arguably, there is a seasonality in the distribution of interviews, yet it is not an issue per se. 22 Indeed, it is not detrimental that people report larger well-being scores in the summer, for instance. Indeed, we control for the month during which the interview took place and, most importantly, this is the di erential SWB between days surrounding the event that identi es the e ect (or them netted from the same days di erence in the previous year, in the double 22 This table shows the detailed distribution of interviews over the year for each country speci cally. The GSOEP is collected mainly during the rst half of the year, the PSID in spring and summer, the BHPS, SHP, RLMS and HILDA mosty in the end of the summer and in the autumn. For this last groups of surveys, there are between 3 and 5 months without interview in the year, either in spring or summy. 12

14 di erence approach). Finally, we may overstate the e ect of terror intensity especially for fatalities if what it reveals is the welfare impact of the main event of the previous day. To control for this seemingly omitted variable, we can actually include a set of characteristics about that event, which would capture its visibility and more qualitative forms of intensity (like the shock from suicide bombing or from particularly long attacks). In sensitivity analyses, we thus include C td 1, a vector of characteristics describing the main event, de ned as the event with the largest number of killed and wounded 23. These characteristics include dummies for attack speci cities (multiple attacks, extended attack (more than 24 hours), suicide attack), for attack types (armed assault, assassination, explosion/bombing, facility/infrastructure attacks, hijacking, hostage taking, unarmed assault), for criteria de ning a terror attack (intentionality, intention to coerce/intimidate/publicize to larger audience, standing outside international humanitarian law) and for 13 broad world regions where terror took place (summarized in Table A.2). 3 Results We rst estimate baseline models to elicit how daily global terror relates to the well-being of individuals in our overall sample and in speci c locations. We explore the sensitivity of these results to alternative measures of terror intensity, estimation methods and model speci cations. We then study the cumulated e ect of past events and how individuals adapt to these events over time. We explore the potential mechanisms through which terror a ect well-being in the very short-run (fear, stock markets). We document the heterogeneity of the e ect along di erent dimensions (like the distance to the victims). Finally, we suggest welfare calculations to proxy a global cost of terror. 3.1 Baseline Results Global Terror E ect. We start with the estimation of model (1) on our 6-country panel data. The tables of estimates that follow report the coe cients on terror intensity and, for the purpose of equivalent income calculations, the coe cient ln y on log household income. 24 Baseline estimates are presented in Table 1. The rst column for both incidences and fatalities 23 In case several events have the number of killed and wounded, we select the earlier event or one of them randomly results hardly change given there are very few such cases. 24 The complete set of life satisfaction estimates is available from the authors. It shows very standard results (as surveyed in Clark et al., 2008), summarized as follows. Essentially, income, good health and being married are positively related to SWB while being unemployed is negatively correlated. The impact of these variables is very comparable and stable across countries, a regularity indicating that SWB data contain reliable and potentially interesting information for welfare measurement (see also Di Tella et al., 2003, Akay et al., 2016). 13

15 shows estimates using the number of incidences/fatalities with individual FE and no control for attack characteristics C td 1. We rst estimate a series of SWB functions using the baseline model speci cation (??-??) in which we control for the full set of observed and unobserved individual regional, and terror events related characteristics to isolate the relationship between fatalities and incidences and well-being. As indicated, a simple dummy for the occurrance of one (or several) terror events worldwide would not provide much variation. Therefore, we must rather focus on the intensity of terror, which will be de ned by two main variables: the total number of terror events (incidences) and the total number of people killed and wounded? (fatalities) every day over Our baseline results are obtained with FE linear estimations and treating SWB as a continuous variable. We check the sensitivity of our results with respect to alternative estimators. Table A.XX in the Appendix reports a series of estimates, starting with the FE model, then moving to the Alternatively, we shall experiment with the Mundlak "quasi-... xed exoects" (QFE) model, which combines both between and within variation. This model allows for the inclusion of variables which drop from FE estimations, notably country exoects. Hence, the overall individual exoect is based on a slightly more structural speci... cation where i = h + Zi + Ageit + Y SMit + ui, with home country exoects h (for unchanging cultural inzuences of origin country on reported well-being), time-invariant characteristics Zi (gender and cohort exoects), two time variables (age and YSM, which are not identi... ed when using FE timedemeaning panel estimation with year exoects), and the Mundlak QFE ui.13. Acknowledging the ordinal nature of SWB data, we also show estimates of the "Blow-up and Cluster" FE ordered logit. The coec/ cient is still negative and signi... cant. We could not calculate marginal exoects but we can check the equivalent income measure, :972, which turns out to be only slightly larger than the linear FE estimation without country time exoects. Then we move to QFE estimates showing very similar results compared to the baseline ( :281 and :224 for QFE models without and with country time exoects, respectively). Yet, we also provide checks where we acknowledge the ordinal nature of the dependent variable, allowing for unobserved individual exoects in this nonlinear context by 14

16 using the QFE ordered probit and the "Blow-up and Cluster" FE ordered logit estimators (see Baetschmann et al., 2015). Table 1: Welfare Impact of Global Terror: Baseline Results Incidence of Terror Events Fatalities due to Terror Events #Incidences Incidences?0 #Fatalities Fatalit Log(Terror effects) *** *** *** *** (0.0025) (0.0026) (0.0097) (0.0013) (0.0014) (0.0051) Log(Household Income) *** *** *** *** *** (0.0029) (0.0029) (0.0028) (0.0029) (0.0029) (0.0028) Equivalent Income R Squared #Observations 747, , , , , ,383 Individual Effect FE FE FE FE FE FE Terror characteristics NO YES NO NO YES NO Linear estimations of life satisfaction on incidences (# or dummy) or fatalities (# or dummy). Models control for age, educ marital status, health indicator, household size, # of children, log household income, year, month, country specific regi residence (the 16 Federal States of Germany, the 13 regions in the UK, the 26 Cantons in Switzerland, a detailed classificat 13 Australian regions, 40 political regions in Russia, and the 50 Federal States in the US) and individual effects (fixed e FE). ``Terror characteristics": features of the main attack that took place the day prior to interview, including dummies for a specificities (multiple attacks, extended attack, suicide attack), for attack types (armed assault, assassination, explosion/bom etc.) and for 13 broad world regions where this event took place. Robust standard errors are presented in the parentheses. *** indicates significance level at 10%, 5% and 1% levels respectively. We begin with presenting results by merging six sample countries together into one global panel dataset and estimate our baseline well-being model speci cations given in equation (??-??) with individual xed-e ects. To be brief we present only the key parameters of interest, i.e., terror variables and household income to compare the magnitudes. The baseline estimates based on our xed-e ects model speci cation (?? including full error speci cation,??), which include the log of daily number of fatalities and the log of daily number of incidences, are presented in Column I and II of Table 4. The estimated parameters are by countries are given in the rows of the table. To be compact we do not present the estimation results for the socio-demographic and -economic characteristics of individuals. 25. The model speci cations 25 The results from the baseline estimates are presented in Appendix A.1. Our main interest in this paper is the relationship between terror measures and well-being. The other parameters, in particular for the socialdemographic variables, are estimated with the same sign and similar magnitute compared to that of SWB literature (Dolan et al., 2008). 15

17 allows for the full set of terror related characteristics. 26. The baseline results are presented in rst two columns of Table 4, split by terror fatalities, indecencies, and also for Europe/Eurasia (Germany, UK, Switzerland, and Russia Federation) and all countries together (Europe/Eurasia and, Australia, and USA). Terror e ects with respect to the log of fatalities and incidences a ect utilities negatively and the estimates are highly statistically signi cant (Column I and II). The parameter estimates of the log of terror fatalities is 0:0071 (ste:0:0015; t value = 5:45) and log of incidences is 0:0134 (ste:0:0028; t value = 6:45) for Europe/Eurasia. The parameter estimates are highly similar. The log fatalities is estimated as 0:0074 (ste:0:0015; t value = 4:45) and log of incidences is 0:0141 (ste:0:0025; t value = 6:45). One important nding is that the parameter estimates are highly similar for Europe and all countries together. We have conducted initial sensitivity checks. Most results are highly similar to baseline and they are left outside of the paper. First, in a stepwise methodology, we rst investigate the e ect of control variables. Terror variables are mostly orthogonal to the individual characteristics and there is no sizeable e ect on the estimates. Second, the terror related characteristics are excluded. The terror de nitions do not have particular e ect on the parameter estimates of fatalities and incidences. The terror dataset include information on the number of people who are wounded in each event. We now calculate the total number of wounded people and include it in our baseline xed-e ects speci cation. Conditional on the fatalities and incidences the log number of wounded people is positive yet it is not statistically signi cant. These results can be found in Table Appendix 1, where we present the full estimation results from the baseline. Terror in Own Country or Beyond. The global terror signi cantly a ects SWB. We now calculate the terror events which are happening our six sample countries and in any country of the world. We sum all fatalities and events occurring in sample countries during each day and estimate our baseline model for both the local and global fatalities and terror incidences. In line with the expectations terror a ects well-being more when it occurs with the own country. The e ect of local terror is about three times larger compared to global terror. The distance to the event might play an important role explaining the e ect. We investigate the distance (physical, cultural etc.) in detail below where we investigate the channels explaining the results. We present the p-value of the di erence between local and global terror e ects. The di erences are highly statistically signi cant in most cases. The di erence of local vs. global terror e ect is larger for European countries. The important result to highlight is that there is a negative and highly statistically signi cant global terror e ect on well-being. 26 These characteristic are mostly statistically signi cant and hightly correlated with the fatalities and incidences. We apply alternative strategies: rst we estimate the models by using total number of daily terror characteristics, i.e., total number of suidice bombing. Second strategy is to control for the characteristics of the top largest events with respect to fatalities. We prefer the later as it is expected to create the largest impact to be precevied by the respondents. Yet the results di er marginally by alternative formulations of terror characteristics. 16

18 In particular, terrorists may choose dates that coincide with religious or natioanl days, during which there are no interviews if this is a bank holiday. Nonetheless, recall that we study the impact of terror on the following day. Moreover, these cases represent a very small fractions of all terror events occurring in the country where the interview took place. Do the Global Terror E ect Di er across Countries? We now present the terror e ect for our six sample countries. The parameter estimates of fatalities and incidences are both negative and statistically signi cant for most of the countries. One important remark to note is that the parameter estimates are highly consistent across the countries. 27 The parameter estimate of the fatalities (Column I) ranges from 0:0055 and 0:0118 (with a std. err from 0:0023 and 0:0031). The fatalities are highly statistically signi cant with p-value ranging from p value = 0:001 to p value = 0:027. The largest e ect is observed for USA and Switzerland and these countries are followed by United Kingdom, Australia, and Germany. On the other hand, the parameter estimate of daily #incidences (Column II) ranges from 0:0055 and 0:0078 (with a std. err from 0:0023 and 0:0031). The #incidences are also highly statistically signi cant similar to that of resulting #fatalities (p-values ranging from p value = 0:001 to p value = 0:027). In an additional model speci cation we merge three European countries (Germany, United Kingdom, and Switzerland) into one dataset which represent Europe continent. The parameter estimate for the fatalities is 0:0061 (with std. err. of 0:0031 and p value = 0:001) and incidences is 0:0061 (with std. err. of 0:0031 and p value = 0:001). A comparison based on the estimated parameters reveals that the e ect is highly similar across continents with statistically insigni cant binary comparisons. Our results are highly in-line with the previous literature (e.g., Stutzer et al., 2009). The e ect of terror is global and a ecting individuals beyond the borders. Magnitudes and Comparison with the Literature. We now calculate the standardized coe cients to obtain relative size of the terror e ects. The results are presented in V and VI columns of Table 4 for the full sample, local vs. global terror and in Table 5 we present the magnitudes for each sample country. While a standard deviation increase in fatalities leads to about standard deviation reduction in well-being, a standard deviation increase in the incidences lowers well-being standard deviation. The magnitude of terror is almost the same for Europe/Eurasia and all countries together. In comparison with the income, the magnitude of terror on SWB is about 7 to 10% of that of income. The magnitude of terror e ects shows a small variation across countries. The largest e ect of fatalities is experienced among Australians and it is followed by Switzerland and Russia. The largest e ect of incidences, 27 SWB measure for USA, BHPS, and Russia are equalized using 100 bootstrap replications as described before. We note that the equalisation decrease the precision of the estimates. 17

19 Table 2: Terror E ect by Countries: Baseline Results Fixed Effects Quasi Fixed Effects Sandardized Coefficients I II III IV V VI #Fatalities #Incidences #Fatalities #Incidence #Fatalities #Incidences Europe: Germany (GSOEP) Log(Terror effects) ** *** ** *** ** *** (0.0024) (0.0046) (0.0024) (0.0045) (0.0018) (0.0020) R Squared #Observations 298, , , ,602 Europe: United Kingdom (BHPS) Log(Terror effects) ** *** ** *** ** *** (0.0023) (0.0048) (0.0023) (0.0047) (0.0024) (0.0024) R Squared #Observations Europe: Switzerland (SHP) Log(Terror effects) ** * * (0.0057) (0.0102) (0.0056) (0.0100) (0.0051) (0.0061) R Squared #Observations Eurasia: Russia (RMLS) Log(Terror effects) * * * (0.0031) (0.0056) (0.0029) (0.0053) (0.0034) (0.0042) R Squared #Observations Australia: Australia (HILDA) Log(Terror effects) ** * ** ** ** * (0.0037) (0.0068) (0.0037) (0.0067) (0.0033) (0.0039) R Squared #Observations North America: USA(PSID) Log(Terror effects) ** * ** (0.0077) (0.0131) (0.0068) (0.0115) (0.0079) (0.0086) R Squared #Observations Notes: The model specifications include full set of control variables (see table Appendix A). All models include country specific region of residence (16 federal states for Germany, 12 regions for Australia, 52 federal states for USA, 19 regions for United Kingdom, 26 regions for Switzerland), year dummies, region of terror dummies, type of terror dummies, terror classifications with three criteria. The models allow for individual fixed fixed effects, Quasifixed effects or Mundlak's formulation (with income, education, household size, health and age). The final columns presents the standardized coefficents based on fixed effects model. Robust standard errors are presented in the paranthesis. *, **, *** indicates significance level at 10%, 5% and 1% levels of significance respectively. 18

20 however, is experienced in USA and it is followed by Switzerland and Australia. 3.2 Sensitivity Analysis Estimators. Our baseline model speci cation is estimated using a linear individual xede ects to capture the correlation between unobserved and observed individual characteristics. We estimate alternative model speci cations. Table 4 and Table 5 report results from a correlated e ects model speci cation, i.e.., quasi- xed-e ects (QFE), in Column III and IV. The auxiliary speci cation of unobserved individual e ects is speci ed using within-means of timevariant characteristics including household income, individual working hours, health status, years of education, household size and age. QFE produces highly comparable results with that of the baseline for the full sample and for each country separately. Experimenting with the functions of unobserved individual e ects suggest that the parameter estimates and the statistical signi cance levels are not sensitive to the speci cation of auxiliary function. Among unreported results, we estimate models using both OLS and "Blow and Cluster" ordered logit xed-e ects model speci cation. OLS, which ignores the unobserved individual-e ects produce somehow larger parameter estimates of compared to that of baseline. The latter estimator aims to account the ordinal nature of SWB dataset. The parameter estimates of terror measures are both negative, larger and highly statistically signi cant for the full sample and for all countries. The Terror Day and the Day After We are going to present some initial sensitivity analysis. To be able to ensure that we merge the terror information to those interviews held between 24 and 48 hours of the terror events. We now investigate the terror e ect on the day of the terror as some people might have heard the events even in the same day especially for those events occurring during the early hours of a day. Next, in order to allow some further exibility we estimate a model in which we allow interaction between fatalities and incidences. The fatalities tends to generate lesser SWB e ect as the number of incidences are higher (Table 6, Column IV). Last but not the least, we use equalized scale for SWB measure. The results from the equalized SWB scale (see Footnote 16) suggests a slightly larger terror e ects. 28 Alternative Model Speci cations. Considering the high correlation between log of fatalities and incidences (0.83) the baseline model allows terror them in sperate speci cations. We now modify baseline and allow fatalities and incidences in the same model speci cation (Table 6, A and B). The results are highly in line with the baseline. The e ects are both negative and highly statistically signi cant. Next, we allow interactions between fatalities and incidences (Table 6, Column B). 28 Full set of results can be reported by authors upon request. 19

21 Equalization of Scales. We now present results from the common ve-points scale. The results are presented in columns VII and VIII. They are lower in size as expected. The signs and the signi cance levels are the same. Does Intensity Matter? We now generate dummy variables which characterize intense terror events. The median number of people died and wounded are 13 and 19 during last 20 years. The experience of terror events in a day in which more than median people died and/or wounded are presented in the second column of Table 3. The parameter estimate is negative and highly statistically signi cant. A similar result is obtained in all countries and for the full sample. The terror events such as 9/11 and London metro attack generates global level e ect. We test if the terror e ect reported above is an artifact of such large events. We now exclude large events with more fatalities than 3 rd quartile of the distribution of terror fatalities (45 people). The results are presented in Table 7 (Column I). The terror e ects are lower yet they are still statistically signi cant. We then focus on the terror events involving bombing/explosions. The results are highly similar to the baseline (Table 7, Column II). 3.3 Do We Learn Living with Terror? Adaptation and Accumulation Adaptations to Terror Related Shocks. Terror events generate negative shocks that people might adapt as time passes over the event. 29 The adaptation e ect might partly explain the convergent accumulative e ect shown above. We now focus on whether there is a global adaptation process to events, whether the adaptation is complete, and how many days does it take for recovery. Our approach to investigate adaptation is to use lags of terror measures on SWB (Di Tella et al., 2010). The idea is that the initial shock of terror - the negative e ect that we have reported - on SWB decreases and bounces back to the level before the event. The model that we specify is as follows: SW B it = X 0 it + +0 iet + 0 iet + +0 iet iet + it; (2) 29 Recent literature investigates adaptation for several positive or negative life-events such as income shocks, unemployment, marriage or dead of a belowed person. Studies suggest that people fully adapts to these events implying that after an initial shock SWB bounces back to the its average level (REFS). 20

22 Table 3: Sensitivity Analysis Europe/Eurasia The Terror Day Sum of the Terror day and the Day After Model Spec. 1 (a) Model Spec. 2 (a) Equalized Scale (5 points scale) Intensive Events (c) Fat. Inc. Fat. Inc. Fat. Inc. Fat. Inc. Fat. Inc. I II III IV V VI IX X XI XII Fatalities ** *** *** *** *** *** (0.0015) (0.0018) (0.0019) (0.0066) (0.0011) (0.0039) Incidences ** *** *** *** *** *** (0.0029) (0.0032) (0.0478) (0.0598) (0.0029) (0.0045) Fat.*Inc ** (0.0811) R Squared #Obs. 638, , , , , , , , , ,527 All countries Fatalities *** *** *** ** *** *** (0.0014) (0.0017) (0.0018) (0.0055) (0.0010) (0.0031) Incidences *** *** *** *** *** *** (0.0026) (0.0030) (0.0415) (0.0530) (0.0023) (0.0035) Fat.*Inc (0.0665) R Squared #Obs. 774, , , , , , , , , ,384 Note: Robust standard errors are presented in the paranthesis. *, **, *** indicates significance level at 10%, 5% and 1% levels of significance respectively. See also Table 4 6. (a) Fatalities and incidences are not log transformed and devided by 100. (b) Log of total number of wounded people each day and log of the total number of events with a property demage in each day. (c) Dummies are created using median number of fatalities and incidences. 21

23 where the terms are de ned as: iet = X K a;i;e k;t + X K c;i;e k;t + X K r;i;e k;t; (3) k=0 k=0 k=0 + iet = X K a;i;e+k;t + X K c;i;e+k;t + X K r;i;e+k;t; ; (4) k=0 k=0 k=0 iet = X K T error i;e k;t; ft error : F; Ig (5) k=0 + iet = X K T error i;e+k;t; ft error : F; Ig (6) k=0 it = t + m + r + ' i + " it (7) This large model includes K (k = 1; :::; K) days long lags iet and leads of terror + iet measured using fatalities F and incidences I measured daily and matched to the day of the interview e for every individual i measured in di erent periods t. The model speci cation includes also lags iet and leads of + iet terror related characteristics as terror criteria a dummies (3 set of dummies for each k), total number of multiple, extended, suicide attacks, and attack types dummies r (3 variables plus 10 dummies for each k), and region of terror dummies (13 dummies for each k). That is, the model includes K day lags and leads of all terror related variables included in the baseline model (1-3). To this end we calculate a week long daily lags of terror related measures and characteristics and three days of leads to determine the set level. The number of lags and leads are chosen arbitrarily. An important remark is that the results for the initial days after the shock are not sensitive to a higher number of leads and lags. The baseline model (2) is modi ed by including full set of lagged fatalities and incidences as well as the full set of terror related characteristics including region of terror and the attack type. We deliver our results in four panels of Figure 3. We focus on Europe/Eurasia and the globe in here (please see Appendix B for the adaptation pro les of each country). Indeed, gures suggest that there is a full and smooth adaptation to terror related shocks on SWB. The adaptation takes 3 to 5 days which supports the convergence of accumulating e ect previously presented. The speed of adaptation is almost the same for fatalities and incidences. In gures T indicates the day of terror (marked with vertical line). The initial shock is observed during the next day of the event, which is the baseline results reported above. Conditional on the fatalities, incidences and terror characteristics of the past and future events, the net e ect of terror events decreases as the number of days since interview days increases. 22

24 Coefficient of Terror on SWB Adaptation to Terror: Europe Terror Measure: Fatalities Daily Terror Effect on SWB 95% Confidence Intervals T 3 T 2 T 1 T T+1 T+2 T+3 T+4 T+5 T+6 T+7 Days Before and After The Terror Event Coefficient of Terror on SWB Adaptation to Terror: Europe Terror Measure: Incidences Daily Terror Effect on SWB 95% Confidence Intervals T 3 T 2 T 1 T T+1 T+2 T+3 T+4 T+5 T+6 T+7 Days Before and After The Terror Event Adaptation to Terror: All Countries Terror Measure: Fatalities Adaptation to Terror: All Countries Terror Measure: Incidences Coefficient of Terror on SWB Daily Terror Effect on SWB 95% Confidence Intervals T 3 T 2 T 1 T T+1 T+2 T+3 T+4 T+5 T+6 T+7 Days Before and After The Terror Event Coefficient of Terror on SWB Daily Terror Effect on SWB 95% Confidence Intervals T 3 T 2 T 1 T T+1 T+2 T+3 T+4 T+5 T+6 T+7 Days Before and After The Terror Event Accumulation of Recent Past. [Not sure it this is necessary] Terror events during the recent past might also have an accumulated e ect on the day in which the SWB is observed. To calculate the accumulated e ect recent terror history we focus on the one week period before the interview in which SWB is observed. We sequentially control sum the parameter estimates of lags during the last week prior the interview day as P 1 k=0 1F 1 k ; P 2 k=0 2F 1 k ; :::; P 7 k=0 2F 1 k. We then calculate the total e ect of terror of past k days as b 0 ; b 0 + b 1 ; b 0 + b 1 + b 2 ; ::. The model include full set of terror characteristics for each day (i.e., extended, suicide attack, multiple events, region of terror (13 dummies), and type of attack (10 dummies) for each day). We present the results in Figure 3. The terror e ect accumulates for the last 3-4 days The total e ect are statistically signi cant. The accumulated e ect of the fatalities and incidences are getting stable following the adaptation results. 23

25 Accumulated Effect: Europe Terror Measure: Fatalities Accumulated Effect: All Countries Terror Measure: Fatalities Coefficient of Terror on SWB Daily Terror Effect on SWB 95% Confidence Intervals T 1 T 2 T 3 T 4 T 5 T 6 T 7 Accumulated terror since T k Coefficient of Terror on SWB Daily Terror Effect on SWB 95% Confidence Intervals T 1 T 2 T 3 T 4 T 5 T 6 T 7 Accumulated terror from T k to T Coefficient of Terror on SWB Accumulated Effect: Europe Terror Measure: Incidences Daily Terror Effect on SWB 95% Confidence Intervals T 1 T 2 T 3 T 4 T 5 T 6 T 7 Accumulated terror from T k to T Coefficient of Terror on SWB Accumulated Effect: All Countries Terror Measure: Incidences Daily Terror Effect on SWB 95% Confidence Intervals T 1 T 2 T 3 T 4 T 5 T 6 T 7 Accumulated terror from T k to T 4 Why Do People are A ected by Global Terror? Why do people fear more from terrorism even though the likelihood of they could get hurt is much higher by another event such as a gun re or tra c accident (REF). One answer to this question is that people might have irrational fear towards violent events. People feel the threat is real and they experience a terror attack with a probability higher than actual calculations. Events occurring in other countries might be perceived as they would eventually a ect their lives in higher likelihood, for instance, through immigration and tourism. That is, the actual and perceived threat, measured with distance or exposure to terror, might be two of the mediating factor. People might generate fear if there are higher proportion of immigrants from the terror regions in the immediate living area. There can be indirect e ects of terror such as damaged the international economic relationships or daily stock market which are found to be largely a ected by terror to show that there is an independent global utility e ect of terror. In the following section, we investigate these potential channels using alternative set of proxies. The methodology follows two steps. First we test alternative hypothesis which might explain the result and investigate how the terror measures are in uences. To do this we allow for the 24

26 temporal and more permanent characteristics and investigate how terror parameters change. Second, we simply we transform the proxy measures relating to alternative explanations into mostly binary proxies and investigate the heterogeneity terror with respect to these measures to picture how alternative explanations mediates the terror and SWB relationship. Consider a binary proxy P it which shows the di erent categories of a particular channel, e.g., short and long distance to terror events. The terror measures are are then interacted with the proxy as follows: SW Bi;e+1;t = X it + iet + iet + it (8) iet = P it + 1 P it T error iet + 2 (1 P it )T error iet (9) iet = a + c + r ; (10) it = t + m + r + ' i + " it : To investigate how binary proxy P it mediates the terror e ect we systematically test H 0 : 1 = 2. We note that some proxies we identify is available only for some countries and some proxies are measured using di erent scales in di erent countries. The de nition of all other symbols are the same as in baseline. 4.1 Distance Terror in Own, Border Country or Beyond Previous results suggest that there is a sizeable and highly signi cant negative e ect of global terror. At this point we investigate the terror e ect for those events in the own country, border country, or anywhere else. We generate new set of variables as follows: we count fatalities and incidences from six sample countries and any other country. We repeat the same exercise for the terror within the border countries of our sample countries. The results from the baseline model speci cation with the new set of terror variables presented in Table 9A1. In line with the expectations, terror within the own country in uence the utilities almost three times larger than that it occurs in any country of the world. The di erences between parameter estimates are statistically di erent from zero in conventual signi cance levels. We then increase the orbit of border countries. Strikingly, the results are highly consistent with the previous. Terror in border countries leads to a e ect on SWB. Yet the di erences in parameter estimates are smaller compared to the terror occur in own country vs. in any other country. Country of Terror and Physical Distance A lower distance to a terror event or a higher degree of accessibility (e.g., barriers like oceans as in USA and North Africa) of terror groups/terrorists to the location of the respondents might generate higher utility loss. Obviously. a higher distance is expected to generate a lesser utility lost at least in terms of the 25

27 potential life threatening e ect of terror. In the following sections we investigate the di erent aspects of distance. We note that up to this point our model speci cations controls for the continents/region of the terror events. 13 terror-region dummies capture the distance of terror groups among other time-invariant unobserved characteristics of these regions. The reason that we prefer this regional classi cation over continental split is that it captures important regions of terror, e.g., MENA countries better. One remark at this point is that controlling for region dummies have a moderate e ect on the parameter estimates of terror measures. We now re ne the regional controls to capture the unobserved regional heterogeneity in a deeper level. To re ne regional dummies to capture distance as well as the unobserved bilateral economic and political relationships between our sample countries and the country of terror, we allow for country of terror xed-e ects and present results in Table 9A1 split for fatalities, incidences, and also for Europe/Eurasia and all countries together. The estimates are negative and highly statistically signi cant. To show the e ect of country xed-e ects on the baseline, we present the percentage change of the parameter estimated compared to the baseline (Table 4). The negative sign indicates that controlling for terror country- xed e ects 22:9% (11:4%) reduces the baseline estimates of fatalities (incidences) for Europe/Eurasia and 15:6% (8:2%) for all countries together. We interpret this result as the terror country speci c time-invariant heterogeneity - including physical, cultural, genetic, and religious distances etc. - only partially explains the terror e ects. Below, we are going to investigate alternative types of distances between sample countries and country of terror. To investigate the e ect of distance to event deeper, we use the exact latitudes and longitudes of terror incidences. Two important issues emerges. First, we shall select a representative location for each respondent in the sample countries. To this purpose we use the centroids, i.e., central latitude and longitude of an geographical area, of each sample country The centroid is more representative of each sample country - especially for Australia, Russia, and USA - when we consider geographical distribution of respondents within the country. 30 Second, we shall de ne a measure of distance which can represent distance to several incidences during a day. We prefer the mean distance 31 (in kilometers) to terror events in a day. 32 The baseline model speci cation (equation (??-??)) is controlled for the log of mean distance to terror events, conditional on the full set of characteristics including the region of terror xed-e ects. The results are given 30 We have estimated the models using the latitude and longitude of capital cities (e.g., Berlin and Moskow). The results are practially the same. 31 We calculate median distance and distance to the largest event with respect to number of fatalities. The results are highly similar and can be presented upon request. 32 There can be alternative ways to calculate the distance on a sphare. We use stratightforward approach based on the trigonomatric birdview distance from the centoid of each sample country to the latitude and longitdue of each terror location. 26

28 in Table 9A2. The estimates are highly similar to that of baselines (Table 4). The parameter estimates of fatalities and incidences are only reduced 2 4%. We now focus on the fatalities and incidences with respect to low and high mean distance in a day. To de ne an indicator variable we use median of the daily distribution of mean distances (P = 1 if mean distance larger than median distance). The estimation results of our interaction model speci cation in equation (8-10) are shown in Table 9A3. The terror e ects are larger when the events occur in a closer orbit of the respondents location, especially for the fatalities and Europe. We present the p-value of the test comparing terror e ects for low and high mean distance. The di erences are statistically signi cant only for fatalities among European people. The insigni cant result for the all countries might be explained with the fact that there are large barriers, e.g., oceans, between main terror regions and Australia and USA, which in turn implies that distance does not correlate much with the terror e ect. Re ning Terror Measures with Physical Distance. The measures used above give equal weights to each terror incidence independent of the distance to the respondents location. We combine terror fatalities with the distance to terror incidence for each event to calculate a distance weighted sum of total number of fatalities. Consider fatalities in each terror incidence in a day f i;d as a random vector F d = ff 1;d ; :::; f Nd ;dg and the distance to each event as a random vector of distances (calculated using latitude and longitude information) d = f@ 1;d ; Nd ;dg, where N d is a the number of event during a day. The distance weighted sum of fatalities in a day is P N d i=1 i;d max( d ) )f i;d. The log of distance weighted fatalities is then used as the measure of terror in our model speci cation. The results are presented in Table 9.A4. The e ect is negative and highly statistically signi cant. The size of the e ect is 2:5% lower than baseline. Overall, the physical distance only partially explain the terror e ects. Cultural Distance. The negative e ect of terror on SWB might not only be explained by the physical distance but also by the cultural proximity between sample countries and the country of terror and also the victims/targets of terror attacks. To test how cultural proximity mediates terror e ects, we build measures for cultural distance based on the shared historical values between countries and victims. We rst test how terror a ects people in our sample countries when the victims/target share the same or another nationality. The GTD dataset collects information for the nationality of three major victims/targets. In most of the terror incidences the information about the national of victims/target is captured by the one major nationality. We simply use this yet add the information from the second and third victim/target groups when the information is missing. Using the information on the nationality of victims/target killed or attached in the country of terror, we identify the total number of fatalities and incidences in a day to generate to distinct variables for the victims from the same nationality and nationality of another country. The results are presented in Table 9B1. The e ect of same nationality fatalities 27

29 is larger than that of the other countries for both Europe and all countries together. Yet the di erences are statistically signi cant only for Europe ( 0:21 vs. 0:003, p value > 0:001; for Europe; 0:21 vs. 0:003, p value > 0:001; for all countries). The e ect of number of events which targets people with the same nationality of our sample countries is large and statistically di erent when terror groups targets victims from any other nationality ( 0:21 vs. 0:003, p value > 0:001; for Europe; 0:21 vs. 0:003, p value > 0:001; for all countries). Going a step further we investigate whether the e ect of fatalities and incidences di er when the victims are from a Western or another country. We note that all our sample countries shares Western values. As all our sample countries share Western values, the likelihood, for instance, that events occurring in Western countries take place in media more and people might be informed more about the events. 33 Using a similar methodology we identify events which targets victims from Western countries. We sum all fatalities from Western countries and incidences within a day. The results are given in Table 9B.2. The e ect of fatalities are statistically di erent for both Europe and all countries together ( 0:21 vs. 0:003, p value > 0:001; for Europe; 0:21 vs. 0:003, p value > 0:001; for all countries) implying that the e ect of terror with Western fatalities e ects the happiness of people more. A similar result is obtained when we consider the incidences which targets Western people ( 0:21 vs. 0:003, p value > 0:001; for Europe; 0:21 vs. 0:003, p value > 0:001; for all countries). To check whether the terror attacks are in Western countries in general, we identify fatalities and incidences in Western countries independent of the target of the attackers. The results are in Table 9B3. The results are highly similar to the previous. They are signi cantly di erent for both fatalities and incidences, and for Europe and for all countries together. The results implies that a cultural similarity plays an important role on mediating the terror e ects for both fatalities and incidences. Religious Distance. To investigate how cultural similarity mediates the terror-swb relationship, we now focus on the religious similarity between our sample countries, terror countries and religion of terror victims. To obtain religious distance measure we identify the share major religion in each country of terror. As our sample countries are all predominantly Christian we focus particularly on the Christians and also Muslims to obtain further results from a terror country/victims from a distant religion. The religion data is obtained from World Religion 33 The de nition of "Western" countries is unclear. We use the list of countries supplied in... We have also experimented with the alternative de nitions. The results hardly change. The results obtained by using other de nition of cultures are also available. One example of this one is Eastern countries. Basically there is no di erence in fatalities and incidences by Eastern values...[aa: Check this with alternative formulations]. Among the unreported results, we also calculate for the EU countires combined with or without Western Europe with alternative set of countries. The results are highly similar. 28

30 Table 4: Distance: Physical/Cultural and Religious/Genetic Distances to the Event Europe Eurasia All Countries Europe Eurasia All Countries #Fat. #Inc. #Fat. #Inc. #Fat. #Inc. #Fat. #Inc. A. Physical Distance 1. Controlling for the country of terror fixed effects B. Cultural and Religious Distance 1. Fatalities/incidences with victims/targets from the same or other countries Terror effect *** *** *** *** Same countries *** *** *** *** (0.0020) (0.0035) (0.0018) (0.0032) (0.0045) (0.0055) (0.0041) (0.0049) % Change Other countries *** *** *** *** #Obs. 606, , , ,629 (0.0019) (0.0034) (0.0018) (0.0031) 2. Controlling for the average physical distance to terror incidences p value Terror effect *** *** *** *** #Obs. 606, , , ,629 (0.0022) (0.0041) (0.0020) (0.0037) 2. Fatalities/incidences with victims/targets from the Western or other countries % Change Western counries *** *** *** *** #Obs. 567, , , ,669 (0.0031) (0.0038) (0.0029) (0.0034) 3. High average physical distance to the terror event Not Western countries *** *** Low *** *** *** *** (0.0020) (0.0034) (0.0018) (0.0031) (0.0024) (0.0043) (0.0022) (0.0039) p value High *** #Obs. 606, , , ,629 (0.0040) (0.0080) (0.0032) (0.0059) 3. Terror incidences in Western countries p value Western countries *** *** *** *** #Obs. 567, , , ,669 (0.0035) (0.0041) (0.0032) (0.0037) 4. Revising terror measures: distance weighted fatalities Non Western countries *** *** ** Terror effect *** *** (0.0020) (0.0034) (0.0018) (0.0031) (0.0021) (0.0019) p value % Change #Obs. 606, , , ,629 #Obs. 606, , Terror incidences in Christian countries C. Genetic Distance High #Christians ** *** *** *** 1. Controlling for the average genetic distance to the terror country (0.0031) (0.0059) (0.0028) (0.0051) *** *** *** *** Low #Christians ** ** ** * (0.0021) (0.0037) (0.0020) (0.0036) (0.0025) (0.0043) (0.0023) (0.0039) % Change p value #Obs. 553, , , ,826 #Obs. 561, , , , High average genetic distance to the country of terror 5. Victims/Targets from Christian countries Low High #Christians *** *** *** *** (0.0038) (0.0065) (0.0037) (0.0062) High *** *** *** *** Low #Christians (0.0024) (0.0043) (0.0023) (0.0040) p value p value #Obs. 553, , , ,826 #Obs. 562, , , , High average genetic distance to the victims/target 6. Terror in Muslim countries Low High #Muslims ** ** * (0.0037) (0.0064) (0.0036) (0.0062) (0.0031) (0.0050) (0.0028) (0.0044) High *** *** *** *** Low #Muslims *** *** *** *** (0.0024) (0.0043) (0.0023) (0.0041) (0.0025) (0.0047) (0.0023) (0.0042) p value p value #Obs. 554, , , ,877 #Obs. 561, , , ,328 Notes: The models specification include the full set of variables used in baseline model, including local region fixed effects, time dummies, terror related characteristics, and individual fixed effects. %Change shows the percentage difference between the terror parameter compared to the baseline model specification with the same sample size after delating the missing values. ***,**,* indicate statistical significance at the leve 0.01, 0.05, and 0.10 respectively. Robust standard errors are in parantheses. 7. Victims/Targets from Muslim countries High #Muslims * * (0.0032) (0.0050) (0.0029) (0.0045) Low #Muslims *** *** *** *** (0.0025) (0.0047) (0.0023) (0.0042) p value #Obs. 562, , , ,577 29

31 Database. 34 To identify predominantly Christian countries we use median share of Christians within each terror country. As in the previous analysis, we then create measures of fatalities and incidences based on whether the terror occur in a Christian country or victims/targets are from Christian countries. The terror in Christian countries generates larger impact with respect to both fatalities and incidences among European people and all over the world. This is not surprising as all sample countries are mainly Christian implying that a lower religious distance generates larger terror e ects. A similar, even stronger, result is obtained when we investigate the religion of the victims killed in the terror attack or the incidences in which terrorists attach to Christians. To check the e ect of religious distance on the utility e ect, we now consider terror occurring in Muslim countries or mainly targeting Muslim people. The result is striking. There is no statistically di erent in uence of terror among Christians. The terror measures based on the religion of the terror country or victims/target are correlated with the measures based on cultural distance, i.e., Western countries, about Genetic Distance To proxy biological and cultural distance between our respondents and the terror country or the victims we use a genetic distance measure (Spolaore and Wacziarg, 2009). The genetic distance measure that we adapt aims to proxy biological and cultural di erences between populations which are transmitted across generations. The measure is mainly used to proxy expected biological di erences between randomly selected individuals from two populations. The measure is also considered to be an alternative proxy for the cultural similarity as genetically similar populations share common culture over history. We test whether terror e ects di er when it occurs in genetically similar country or more speci cally genetically similar victims. To this end, we rst match the genetic distance between our sample countries with the terror country for each event using the country of terror and the country of origin of victims in each event. Later, we calculate the mean genetic distance using all events occurring during a day. Controlling for the mean genetic distance do not in uence the e ect of fatalities and incidences on individuals SWB (Table 9C1). However, we nd that terror e ects are highly heterogeneous with respect to genetic distance. We now de ne two variables indicating countries and victims country of origins with high mean genetic distance using median of the genetic distance distributions. These results are presented in Table 9C2 and 9C3. The evidence suggest that the terror e ects are large and statistically signi cant only when the terror country or the victims are not genetically close to the respondents in our sample countries. The di erences are highly statistically signi cant for both fatalities and incidences. We have also re ned the fatalities using the genetic distance information. We calculate the weighted sum of fatalities and estimate the fatalities for low and high genetic distance. The result is highly similar and 34 Please see further information. [AA: Check when you have internet] 30

32 therefore they are not reported here. There can be alternative interpretations of this result. First, people who are genetically closer shares similar culture and they might be living in close proximity. Therefore, the result simply re ects the cultural distance (Table 9B). Second, the results are also in line with the evolutionary arguments such as altruism and kin selection towards genetically close populations, or with cultural relationships such as racism. 4.2 Exposure Big Cities/Population Density. Terror groups aim to generate large impact and they most often target large metropolises or densely populated areas. Therefore, people living in large metropolises might fear more of a potential terror attack. That is, the terror e ect might be explained with the demographics of the immediate living area. To investigate this, we fetch regional population and population density information for each country. Two approaches are adapted: rst is to use data from population registers of each country. Yet we only have access for the regional population data of Germany (96 ROR), USA (50 Federal States), and Russia (40 Metropolitan regional units). Second approach is simply based on the sampling weights of each data source. Since the datasets at use are all nationally representative, we are able to infer at the minimum whether individual lives in a largely in a big city/metropolis. Local population density dataset is more scarce than local population information and the detailed panel information is available only for Germany (96 ROR) and USA (50 Federal States). We test rst whether people live in big cities/metropolises fear more from the terror using two alternative sources of information. To de ne whether individual lives in a highly populated area we use 3 rd quartile of the regional population distribution within each sample country. The result presented in Table 10A1 and 10A2 o ers clear evidence: individuals living in largely populated areas are a ected more from the terror fatalities ( 0:21vs: 0:003, p value > 0:001, for Europe; 0:21vs: 0:003, p value > 0:001, for all countries) and incidences ( 0:21vs: 0:003, p value > 0:001, for Europe; 0:21vs: 0:003, p value > 0:001, for all countries). The di erences are highly statistically signi cant and highly similar when two sources of information are compared, i.e. metropolitan area based on sampling weights (Table 10A1) and register information for three countries (Table 10A2). We turn our attention to population density using counts per square kilometer. Panel information is available only for Germany and USA. We de ne high population density regions using the 3 rd quartile of the regional population density distribution. The result presented in Table 10A3 (right part of the table) suggest highly similar results. The negative e ect of terror is larger among people living in densely populated areas. Tourism. We test whether the terror e ects relate to the number of tourists in each region. The detailed regional information is available only for Germany (96 RORs). The impact of 31

33 tourism on the terror e ects is a prior unclear. One on hand, tourism might positively correlate with fear. For instance, the terror groups might attack touristic regions to generate a larger international impact. On the other, tourism generate physical and social capital which correlate both in physical security and also further communication across cultures, which in turn might lead people in touristic regions experience lower terror e ect. We now test how tourism in the region relate to terror parameters. We use the number of overnight-stays in 96 ROR regions from 1998 to We generate a dummy which takes one for those regions with tourists larger than median number of overnight-stays". 35 The interaction model with the terror fatalities and incidences are presented in Table 10A4. The model also includes 96 ROR level dummies to capture further unobserved time-invariant regional characteristics and also timevariant population size in the region (the correlation between population size and overnightstays is -0.26). The result suggest that people in high touristic regions are a ected less from the terror fatalities and incidences. Yet, the di erences are only marginally statistically signi cant for the case of terror fatalities ( 0:091 vs: 0:003, p value > 0:0811). Exposure through Media and Internet We now investigate how media and internet mediates the terror e ect. Individuals who follows news from TV or internet (e.g., social media) might obtain more information about the international terror and potential threats. We now focus on the individual behavior and how well-being relates to terror fatalities and incidences. That is, we test whether terror a ect di ers by information that individuals acquire about the terror events. To this end we exploit information the hours of TV watching per day and hours of internet use per day. The information is available in all panel datasets except USA. Using sample country-speci c median hours of TV watching and internet usage we de ne two dummy variables for high TV watching individuals and high internet using individuals two proxy higher level of information about the terror events. The results from the interaction model suggest that a higher level of information is related to a higher terror e ect. The di erences are not statistically signi cant either for TV watching or internet usage. Social Life and Social Capital Next, we test whether terror damage the social life and social relationships. For instance, people who fear from terror might prefer not to socialize outside. We use a proxy based on how frequently individuals go out/eat outside/socialize with friends. Immigration. There is a hot debate today whether people living in areas with a larger share of migrants are in uenced from the terror event through indirect channels. In particular, 35 The de nition of overnight stays include the number of non-resident visitors who spend time in the region irrespective of the mode of accomodation (including spending the night in camping sites, cars and trains). Please see for information. 32

34 Table 5: Exposure: Regional characteristics/media and Internet/Social Life and Social Cpaital/Tourism/Immigration Europe Eurasia All Countries Europe Eurasia All Countries #Fat. #Inc. #Fat. #Inc. #Fat. #Inc. #Fat. #Inc. A. Regional Characteristics: Population and Metropoles A. Regional Characteristics: Density and Tourism 1. Individual is living in the big city/metropolitian area (#) 3. Population density in the residence region (###) Small city Low density (0.0028) (0.0049) (0.0028) (0.0048) (0.0037) (0.0067) (0.0036) (0.0064) Big city/metropole *** *** *** *** High density ** *** *** *** (0.0041) (0.0041) (0.0040) (0.0040) (0.0052) (0.0095) (0.0050) (0.0091) p value p value #Obs. 384, , , ,378 #Obs. 232, , , , Individual is living in the big city/metropolitian area (##) 4. Living in high touristic region (a) Small city * ** *** Low * * (0.0020) (0.0037) (0.0019) (0.0033) (0.0040) (0.0073) Big city/metropole *** *** *** *** High (0.0029) (0.0029) (0.0027) (0.0027) (0.0044) (0.0077) p value p value #Obs. 614, , , ,383 #Obs. 232, ,660 B. Individual Behavior: Media and Internet B. Individual Behavior: Social Life and Social Capital 1. Watching TV/News (b) 3. Individual is highly outgoing/social (eating out, going bars/bubs) (c) Less frequently Less frequently (0.0046) (0.0087) (0.0046) (0.0087) (0.0090) (0.0160) (0.0046) (0.0077) Daily ** ** ** ** Very often ** *** (0.0030) (0.0060) (0.0030) (0.0060) (0.0125) (0.0205) (0.0059) (0.0097) p value p value #Obs. 212, , , ,781 #Observations 78,937 78, , , Internet/Social media use (b) 4. Individual trusts other people in general (b) Less frequently Not Trusts others ** ** ** *** (0.0049) (0.0084) (0.0048) (0.0081) (0.0049) (0.0084) (0.0044) (0.0073) Daily * ** * Trust others (0.0034) (0.0059) (0.0033) (0.0054) (0.0053) (0.0090) (0.0048) (0.0081) p value p value #Obs. 225, , , ,318 #Observations 169, , , ,237 C. Immigration: Share by Country of Origin (+) C. Immigration: Shares by Religon and Culture (+) 1. Controlling for the immigrants' share in the residence region 4. Controlling for the Muslim immigrants' share in the residence region Terror Effects * *** ** *** Terror Effects * *** ** *** (0.0026) (0.0048) (0.0025) (0.0047) (0.0026) (0.0048) (0.0025) (0.0047) % Change % Change #Obs. 269, , , ,704 #Obs. 269, , , , High shares of immigrants in the residence region 5. High shares of immigrants from predominantly Muslim countries Low Low (0.0034) (0.0064) (0.0033) (0.0060) (0.0035) (0.0064) (0.0034) (0.0061) High *** *** *** *** High *** *** *** *** (0.0033) (0.0062) (0.0033) (0.0061) (0.0033) (0.0062) (0.0032) (0.0060) p value p value #Observations 269, , , ,704 #Observations 269, , , , High shares of immigrants from top 10 terror countries in the residence region 6. High shares of immigrants from Western countries Low Low (0.0034) (0.0064) (0.0033) (0.0061) (0.0034) (0.0064) (0.0033) (0.0060) High * *** *** High *** *** ** *** (0.0034) (0.0062) (0.0034) (0.0060) (0.0033) (0.0062) (0.0033) (0.0061) p value p value #Obs. 269, , , ,704 #Obs. 269, , , ,704 Notes: The dependent variable corresponds to answers to the question How satisfied are you at present with your life as a whole? (values range from 0 to 10). (#) Regional population register data available only for Germany (96 RORs), USA (50 Federal States), and Russia (40 Regions) (##) The results are obtained using the regional sampling weights of our datasets. (###) Register based regional population density information is available only Germany (96 RORs) and USA (50 Federal States) (a) Time variant regional data is available only for German 96 RORs. (b) All countries (only some years) without USA (PSID) (c) The data is available only for Germany, United Kingdom, and Australis (only some years) (+) The detailed register and census based data on regional immigration is available for Germany (96 RORs) and USA (50 Federal States) 33

35 recent terror incidences in Muslim countries and terror conducted by radical religious groups increased the tension between natives in Western countries and Muslim immigrants. We aim bring evidence on whether the terror e ect on SWB is driven by the large share of immigrants from the countries of terror. To address this issue we ideally need to have spatial information on the exact share of immigrants country of origins within a smaller region in our sample countries. The detailed information on the exact number of individuals split by their country of origin in relatively smaller regions over years is available only for Germany (in 96 ROR level between 1998 and 2013) 36 and USA (50 Federal States level between 1999 and 2013). 37 First, we control our regressions for the share of immigrants. The models allows for ROR and Federal State xed-e ects. The parameter estimates of terror in Table 10C1 are negative and statistically signi cant with an without immigrant share in the immediate living area. 38 Going a step further, we from a dummy indicating regions with a high share of immigrants (more immigrants than median). A higher share of immigrants in the region is associated with a larger terror e ects on SWB (Table 10C2). The di erences are highly statistically signi cant. Next, we investigate whether a high share of immigrants from top 10 (with respect to number of terror activities during last two decades) terror countries relates to terror e ects. 39 The parameters of terror measures are larger among regions with a high share of immigrants from the top 10 countries. The di erences are statistically signi cant in Europe/Eurasia (only Germans in this case) only when we consider the number of the terror events (Table 10C3). Do Immigrants Religion and Culture Matter? The e ect of Muslim immigrant among the native community in Western countries is a hot topic today. The debate raises when in particular terror occurs in Muslim countries or by radical Islamist terror groups conduct the terror. At the this point we bring evidence on whether the terror is associated with a larger e ect when the share of Muslim people is larger. In order to de ne Muslim countries, we sum 36 The immigration data is created as follows: For Fermany we use alternative sources of data. First, we use the INKAR (see Footnote 5 and the wabpage: dataset which includes administrative data for 96 ROR. We then link spatial statistics from the Central Register of Foreign Nationals (Ausländerzentralregister) which supplies registered data on the 174 immigrants country of origins (nationality) from 404 districts of Germany (Kreise) located within RORs. GSOEP dataset does not supply link for the individuals district (Kreise). Using the lookup les we rst aggregate the number of immigrants from eahc country of origin in ROR level and then match it with the information in GSOEP for years [AA: Ahemd could you please write this part in details]. Immigration data fro USA is obtained using ACS for years 1999, 2011, and The immigration de nition used for the USA is based on the rst generation immigrants country of birth. Using the identi er supplied by the data on the rst generation immigrants we count the number of individuals from each country of origin living in 50 Federal States. 38 Share of immigrants enters in SWB equation positively. We investigate how the share of Muslim immigrants relates to SWB of natives in Germany and USA. The result is highly inline with Akay et al., (2014), in both Germany and USA, share of Muslim immigrants relate to SWB of natives positively. 39 Top 10 terror countries are... 34

36 the share of immigrants from countries which are predominantly Muslim (more than 75%). 40 The parameter estimates of terror is large and statistically signi cant only in regions with large share of Muslim immigrants. The di erences are also highly statistically signi cant. We now check whether the share of culturally distant immigrants relates to the terror estimates. We calculate the total share of non-western immigrants in each region. The result is strikingly consistent. The terror e ect is larger when there is a large not only when there is a large share of Muslim immigrants but also by the share of immigrants from non-western countries. The di erences are highly statistically signi cant. media coverage of other events Eisensee, T. and Strömberg, D. (2007). News Droughts, News Floods, and U.S. Disaster Relief. Quarterly Journal of Economics, 122: Shocks, Markets, and Bilateral Relationships Global Events: Natural Disasters. We now investigate other daily negative events which might confound with the terror e ect when they coincide. To this end, we exploit daily natural disasters information occurring in any location in the world. The natural disaster information (The Emergency Events Database, EM-DAT) 41 at use is unique and include daily number of natural disaster related fatalities, total number of people a ected, duration (in total number of days), type (8 types including ood or volcanic activity) and country of origin of disasters. Having cleaned the events with missing information, we identify 4730 natural disasters between 1994 and The daily natural disasters information is merged with our daily global terror dataset. The days in which there is no natural disaster the missing information is replaced with zero. One important point is that natural disasters typically take longer time than terror events. Therefore, the initial starting date might not be the exact peak point with the largest damage. To deal with this issue we are going to allow for the duration of natural disasters in our regressions. We modify our baseline model speci cation by adding daily natural disaster related characteristics which includes country and type of natural disaster, duration, log of fatalities, and log of the number of people a ected. Controlling for natural disaster related characteristics expands the terror e ects, in particular, terror fatalities, by %4:2 for Europe/Eurasia and %6:1 for all countries (Table 11A1). The results imply that terror events are perceived more negative when there is another simultaneously occurring negative event. Indeed, the e ect of terror occurring 40 The countries are...note that the most countries are also in the list of top 10 terror countries. 41 The dataset is obtained from The Emergency Events Database (EM-DAT) - Universite catholique de Louvain (UCL). See for further information. 35

37 in day with at least one natural disasters is larger (Table 11A2). The di erences are significantly di erent for the terror incidences especially when all countries together ( 0:0101 vs. 0:0165, p value = 0:117, for Europe/Eurasia; 0:0099 vs. 0:0188, p value = 0:014, for all countries). However, it is unlikely that an individual can obtain information about the small scale natural disasters. In our next analysis, we identify large natural disasters coinciding terror events (Table 11A3). The large events are identi ed using the 3 rd quartile of number of the people a ected due to natural disasters. The interaction of terror fatalities and incidences suggests a clear evidence that terror events e ect SWB higher when there is a large natural disaster. We also note that the result is practically the same when we use natural disaster related fatalities to identify the large events. Bilateral Economic Relationships with Terror Countries. The terror events might generate temporal economic shocks as they impair economic interactions between countries. The literature already reports that terrorism induce an indirect negative e ect on the international trade (Mierrieks and Gries, 2013). Obviously, the e ect of terror on the economic relationships would take longer time than how terror might relate with the utilities. Yet the terror might in uence the expectations of the people for the potential future loses. We now test whether the bilateral economic relationships between our sample and terror countries relate to the terror e ect on SWB. To identify the volume of economic relationships between countries we use two sources of information. First is the total and raw material exports and imports between countries during years from 1994 to The data are obtained from The World Input-Output Tables (WIOT). 42 Our strategy is as follows: we rst identify the volume of total and raw material exports and imports in USD (PPP international 2010 prices) between our sample countries and the country of terror, and then we merge the data using the country of terror for each day. The second measure for the bilateral economic activity is the Foreign Direct Investment (FDI) between countries. The dataset is obtained from OECD and calibrated using 2010 PPP international USD. 43 FDI dataset is not available for Russia during the sampling period used in our analysis. Using a similar strategy, we merge the FDI information for each country of terror. Conditional on the full set of characteristics, the terror country and individual xed-e ects, the log of export and import between countries is found to reduce the parameter estimates of terror about %8:4 for Europe/Eurasia and %5:4 for all countries (Table 11B1). A similar result is obtained (Table 11B2) when we control for the raw material (e.g., fuels) exports and imports. 42 The world input-output tables for our sample countries over each year of observation is obtained from 43 The FDI statistics are obtained from OECD statistics website, The con dential FDI investment information is considered to be missing and excluded from the analysis. FDI is considered to be zero when it is explicity mentioned as absolute zero in the registers. Our sample also excludes Russia due to insu ciency of data. 36

38 In Table 11B3 we present results when we control for the log of bilateral FDI from and to the terror countries. The terror e ects are robust and a ected about %1 2 with the inclusion of bilateral FDI in our benchmark speci cations. Markets The last but not the least, the channel which can explain the e ect of terror is the temporal uctuations in global and local nancial markets. Terror might re ect on the daily uctuations on the stock market. Indeed, several studies shows that terror events correlate with the uctuations in the stock prices. One argument is that the well-being e ect of terror might be due to changes in the markets, a negative shock in the stock markets might negatively correlate with the SWB of individuals as it implies higher risk in future income possibilities. To test this we collect daily local stock market closing prices for each sample country, the daily Dow Jones global index, international gold and oil prices, and also the exchange rates. We then merge daily market indices with our SWB datasets. Indeed, there is a correlation between our terror measures and stock market uctuations. A number of fatalities and incidences are negatively and modestly correlated with the stock market changes. We now control for the stock market changes in the same regression. Indeed, the parameter estimates of terror measures are slightly smaller. Yet the parameter estimates are highly statistically signi cant implying that there is a distinct global terror e ect on well-being. Controlling for the global stock prices and also returns from the global stock market only slightly reduces the size of terror estimates What is the Global Cost of Terror? In SWB literature reported well-being is considered as an adequate measure for the utility, which allows us to calculate the monetary value of terror as a public bad using marginal rate of substitution between our terror measures and income (Kahneman and Sugden, 2006; Stutzer et al., 2009; Levison, 2012). To capture the cost or terror we use our estimated well-being functions as the overall average utility function and then the cost is simply the income level which should be added on the income level of an individual to compensate the utility lost which is due to terror, i.e., marginal rate of substitution between income and terror measures 45. The 44 Indexes involve seasonal time trends. Our regressions allow for year and month of observation xed-e ects. To deal with the potential e ect of time trends, we use Hodrick-Prescott lter to eleminate seasonal uctuations. The terror e ects are not a ected while the relationship between global prices and happiness is highly a ected. These results are not reported here as it is beyond the main focus on the paper. They all can be provided upon request. 45 This approach has recently been applied to value several intangible goods including airport noises (van Praag and Baarsma, 2005), air pollution and climate (Rehdanz and Maddison, 2005; Welch, 2006) and cost of terror in a country (Stutzer et al., 2009). The method has important advantages over the methods based on state preferences... 37

39 Table 6: Shocks, Billiteral Economic Relatioships and Markets Europe Eurasia All Countries Europe Eurasia All Countries #Fat. #Inc. #Fat. #Inc. #Fat. #Inc. #Fat. #Inc. A. Shocks: Natural Disasters at the Day of Terror 1. Controlling for total number of killed and affected people C. Reactions of Markets: Controlling for the chance in the stock market prices: Dow Jones Global Industrial Avarage (W1100)(b) Terror effect *** *** *** *** Terror Effects *** *** *** *** (0.0015) (0.0029) (0.0014) (0.0027) (0.0016) (0.0030) (0.0015) (0.0027) % Change % Change #Obs. 614, , , ,384 #Obs. 531, , , , There is a natural disaster at the day of terror 2. Returns of Dow Jones Glabal Industrial Average at the end of terror day No Disaster *** *** *** *** Positive return ** ** *** (0.0020) (0.0035) (0.0018) (0.0032) (0.0026) (0.0040) (0.0023) (0.0036) Disaster *** *** *** *** Negative Return *** *** *** *** (0.0020) (0.0036) (0.0018) (0.0032) (0.0023) (0.0034) (0.0020) (0.0031) p value p value #Obs. 614, , , ,384 #Obs. 396, , , , Terror effect when there are high intensity natural disasters 3....the stock market prices: local markets (#) Not Large *** *** *** *** Terror effect *** *** *** *** (0.0017) (0.0032) (0.0016) (0.0029) (0.0015) (0.0029) (0.0014) (0.0026) Large *** *** *** *** % Change (0.0026) (0.0044) (0.0023) (0.0040) #Obs. 590, , , ,834 p value Returns of local stock markets at the end of terror day (c) #Obs. 614, , , ,384 Positive return ** ** B. Bilateral Economic Relationships with Terror Countries (0.0031) (0.0047) (0.0027) (0.0041) 1. Controlling for imports and exports with terror country Negative Return *** *** *** *** Terror effect *** *** *** *** (0.0019) (0.0031) (0.0017) (0.0028) (0.0016) (0.0029) (0.0014) (0.0026) p value % Change #Obs. 456, , , ,834 #Obs. 614, , , , the international gold prices (d) 2. Controlling for imports and exports with terror country: raw material Terror Effects *** *** *** *** Terror effect *** *** *** *** (0.0015) (0.0028) (0.0014) (0.0026) (0.0016) (0.0029) (0.0014) (0.0026) % Change % Change #Obs. 605, , , ,101 #Obs. 614, , , , the international crude oil prices (d) 3. Controlling for biliteral FDI with terror country (a) Terror Effects *** *** *** *** Terror effect *** *** *** *** (0.0015) (0.0029) (0.0014) (0.0026) (0.0018) (0.0033) (0.0016) (0.0030) % Change % Change #Obs. 605, , , ,101 #Obs. 479, , , , the exchange rates (#) Terror Effects *** *** *** *** (0.0015) (0.0028) (0.0014) (0.0026) % Change #Obs. 605, , , ,101 Note: The dependent variable corresponds to answers to the question How satisfied are you at present with your life as a whole? (values range from 0 to 10). The models include full set of characteristics used in the baseline specification. %Change refers to the percentage chance of the terror estimates on well being relative to the baseline model with the same sample size. All models are estimated using individual fixed effects model. (a) The prices are per barrel. (a) No data available for Russia for the sampling period (b) W1 100 is obtained from Bloomberd ( as all other price indices. (c) The local stock markets are the sample country specific Dow Jones indices obtained from Bloomberg. (d) The prices are per ounce. Oil price is measured per barrel. (#) The the exchange rates are calculated relative to Japanease YEN at the day of terror. Robust standard errors in parentheses. */**/***Indicate significance at the 0.1/0.05/0.01 level 38

40 cost of terror is simply. Cost t ln(t ln(y B = b b Y e Yt : (11) T^ error t Identi cation of parameters based on the daily variation of terror events allows us to identify the e ect for each day. That is, given the log-linear model speci cation we can either estimate the daily cost of terror using Y i or we can simply use the yearly averages Y t. The estimates of mean daily cost and the standard errors (based on the delta method) for each country, Europe/Eurasia, and all countries together are presented in Table 11. The cost di ers depending on the terror measure used. For Europe, the de nition based on the fatalities suggests that the yearly cost is yearly 57:6$ per capita (p value = 0:000) and for all countries are included in the analysis the cost is 74:4$ per capita. The de nition based on the incidences produce cost 305:7$ (p value = 0:000) for Europe 590:21$ (p value = 0:000) per capita/year. 46 The percapita total cost (fatalities and incidences are together) is 524:1$ for Europe and 664:6$ for all countries together. We now conduct some comparisons. We have obtained GDP per capita (with PPP 2011 international prices) for sample countries during the period of observation. 47 The mean GDP per capita levels are given in Table 11. We then calculate the percentage of the per capital yearly cost within the GDP per capital from seperate estimates for each country, Europ/Eurasia, and all countries together. The highest percentage is obtained for USA with 7:8% of the GDP per calita/year and it is followed by UK with 5:4% and Australia with 5:1%. The lowest cost relative to the GDP per capita is experienced in Germany with 0:71% and it is followed by Russia with 0:83%. When we pool the countries together the estimates somhow di er. Overall, the per capita yearly cost of terror is 1:63% of GPD per capita in Europe/Eurasia and 1:87% of GDP per capita for all countries together. Finally we obtain information on the yearly military expenditures. The mean levels of military expenditures during the sampline period are presented in Table Strikingly, yearly cost per capita is almost 100% of the overall military cost in Europe and about 83% for all countries together. Comparison with the Literature. Stutzer et al. (2009) estimate of the fatalities and incidences is 0:0074 and 0:0066 and the estimate of the log of household income is 0:168 for the terror events happening in Wales. The average household income is 20; 000 US Dollars. The 46 We have also calculated the equalivalent income using standard and modi ed OECD scales. The scale is based on the number of adults and kids in the household. The standard (modi ed) scale gives weights to individual 1, other adults 0.7 (0.5) and 0.5 (0.3) to each kids. The results are highly comparable with when using equal weights for each family member. 47 The GDP information is abtained form the World Bank Indicators. Please see 48 The military expenditures data used is obtained from 39

41 0:0074 per capita (mean household size is 2:75) marginal rate of substitution is :20; 000: 1 0:168 0: :2$ for fatalities and :20; 000: 1 = 291$ for incidences. The t-test for the di erence between the estimates is statistically di erent for the for the fatalities yet it 0:168 2:75 statistically 2:75 = indi erent for the incidents. This result is consistent with the idea that the terror in uence the well-being of people almost in the same degree independent of it occurs in the same country or any place in the world. 5 Concluding Discussion Assessing a more general impact of terror is important because terror has become global, as argued above, and because social and psychological cost at the global level may exceeds local implications. It also matters to assess whether the exposure of world citizens to repeated terror events through social and class media is large because it would not be without consequences on their economic, health and political behavior. In particular, terror is know to exacerbate fear, risk aversion and feelings of uncertainty (Becker and Rubinstein, 2011). In this respect, it may widely change economic behavior (e.g. savings versus consumption) and health behavior (stress related intake of drugs). If the main channel of the well-being impact of global terror is the di usion of negative feelings and fear, consequences may also be political, with a radicalisation of voters towards political movements that promise more security, national defense and repression even among populations living in relatively safe parts of the world. Terror is a global phenomena today. It threats the lives of people mostly in Western countries. Recent events such as marathon attack in US, Brussels airport attach in Belgium, and devastating multiple attacks in Paris, France increased the fear and security measures all over the European and other Western countries. These events are rare yet they generate disproportionately signi cant global fear. The probability that they can occur again is actually much lower than what people perceive especially compared to any other potentially negative events such as natural disasters, tra c accident, or gun re homicides. 49 To our knowledge rst time in the literature this paper brings several panels of subjective well-being datasets from countries in di erent regions of the world to investigate the global in uences of terror fatalities and events on the well-being of individuals who are living in regions where the odd of terror attack is extremely low. One other novelty of this paper is the identi cation strategy: we rst identify each terror activity and immediate consequences by the date and country of origin between 1994 and Figures suggest that 3,024 (among them 2,996 are killed in 9/11 attack) individuals died in the borders of USA due to foregn-terror who entered into country through immigration or tourism, while more than 320,000 Americans died in gun res from 1996 to 2002 (Nowrasteh, 2016). 40

42 Per Capita Cost of the Global Terror (USD) Per Capita GDP Total Cost / GDP Per Capita Military Expenditure Sample Countries #Fatalities #Incidences Total (a) USD (b) % USD (c) % Total Cost / Military Ependitures Germany ** *** *** 32, (9.29) (70.83) (71.44) United Kingdom * ** ** 30, (54.22) (685.29) (687.43) Switzerland * * 46, (81.77) (542.83) (548.96) Russia Federation * , (68.30) (429.54) (434.93) Australia ** * * 37, (127.43) (968.38) (976.73) USA , (303.34) ( ) ( ) Europe/Eurasia *** *** *** 31, (12.80) (99.55) (100.37) All Countries *** *** *** 35, (14.39) (110.87) (111.80) Notes: The fixed effects model specifications include full set of control variables (see table Appendix A). Robust standard errors are presented in the parentheses. *, **, *** indicates significance level at 10%, 5% and 1% levels of significance respectively. The results in this table can be calculated using the mean values in Table 2 and Table 4 except standard errors (obtained using delta method). (a) The standard errors are calculated by simply pooling the standard errors from two estimates. (b) GDP per capita is based on 2011 international prices. The fig (c) The military expenditure is based current LCU 2011 prices. It is the average per capita military expenditure during the respective sampling period. 41

43 We then merge the terror events with the microeconomic datasets using the next day of the terror activity. Our investigation suggest a large, negative, and highly statistically signi cant e ect of terror fatalities and incidences on the well-being of people next day. Importantly the models allows for the unobserved individual e ects in a xed-e ects speci cation which generates consistent estimators under potential non-random selection of individuals into the day of interview. The terror accumulates. We nd that the recent terror has an important additive e ect. Yet people adapt to negative e ects of terror as the days since terror events increases. The adaptation of terror is quick and takes 3-5 days. The paper calculates the cost of terror using compensating variations. The global cost of terror is calculated as $57.6 Billion using the aggregated population of the countries in our data. This gure is %5-7 of the overall GDP. Our estimates are not statistically di erent from the other estimates of the cost of global terror, $52.3 Billion...(see XXX) p-value of the t-test is One other novelty of the paper is that we calculate the accumulated and out of adaptation cost of terror. This cost is much larger than the cost reported in the literature once the accumulation and adaptation is accounted for. Finally we have investigated alternative channels why people develop such a strong fear towards events with odds of happening much lower than other negative potentially life threatening events. Individuals attitudes and beliefs plays important role. Supporting this ndings, we also report a larger e ect for those married individuals with dependent kids. The e ect tends to be higher for less educated, unemployed, poor individuals. We then investigate the distance to the event. The result suggest people lose higher well-being for the closer events and in culturally similar locations. Originally, we use genetic distance data to approximate cultural distance between terror location and the location in which the well-being is observed. One striking result is that people fear only when the event occur among the people who are culturally close. Several other proxies for distance and also distance weighted measures of terror supports the same conclusion. We have used immigration to investigate if the fear from terror generates larger e ect when there are immigrants from the terror countries in the immediate living environment. The result strongly support the previous ones. The result is generated by both the irrational fear of people and the degree of actual threat simultaneously. As a competing explanation we have investigated other sorts of damages of terror such as on bilateral economic relationships, on stock markets and also other parallel natural disasters. The negative global terror e ect on well-being is distinct and cannot be explained economic damages and other negative global events. There are important limitations of this paper. While we bring datasets from six Western countries, where the SWB data is consistently available and it is possible to form a large panel dataset, we are not able to test the validity of our results for he other parts of the globe.. Thus this paper is only capable to interpret the results for the Western countries, in average. Second, the paper uses investigates the role of immigration partially. The e ect of immediate 42

44 environment and the potential links to the terror is also left for the future research. The previous analyses take as their reference the attacks which are low frequency and high impact. Other studies look at the impact of high-frequency, low-impact attacks. The latter are incorporated into the decisions of economic agents as a permanent cost and their consequences may be highly important, especially when they impinge upon a sector which is a key one for national development. In fact, this importance turns it into one of the main terrorist targets. These conclusions are reached by studies such as those of Abadie and Gardeazabal (2003) and Buesa (2004, 2006) for the Spanish Basque Country References [1] Abadie, A. and Gardeazabal, J. (2003): "The economic cost of con ict: A case study of the Basque Country. American Economic Review, 93, [2] Abadie, A. and Gardeazabal, J. (2008): "Terrorism and the World Economy", European Economic Review, 52 (1),1-27 [3] Akay, A. and Martinsson, P. (2011). Does Relative Income Matter for the Very Poor? Evidence from Rural Ethiopia, Economics Letters, 110: [4] Akay, A., Bargain, O. and K.F. Zimmermann (2012). Relative Concerns of Rural-to-Urban Migrants in China, Journal of Economic Behavior and Organization, 81: [5] Akay, A., Giulietti, C., Robalino, J. D., & Zimmermann, K. F. (2014). Remittances and well-being among rural-to-urban migrants in China. Review of Economics of the Household 12 (3), [6] Akay, A., Karabulut G. and P Martinsson (2013) The E ect of Religiosity and Religious Festivals on Positional Concerns: An Experimental Investigation of Ramadan. Applied Economics, 45 (27), [7] Akay, A., Martinsson, P., and Medhin., H. (2011). Does Positional Concern Matter in Poor Societies? Evidence from a Survey Experiment in Rural Ethiopia. World Development, 40: [8] Alpizar, F., F. Carlsson, and O. Johansson-Stenman How Much Do We Care about Absolute versus Relative Income and Consumption? Journal of Economic Behavior and Organization 56:

45 [9] Becker, Gary S. and Yona Rubinstein (2011), Fear and the Response to Terrorism: an Economic Analysis, CEP Discussion Paper No 1079 [10] Berrebi, C. and J. Ostwald (2015): "Terrorism and fertility: evidence for a causal in uence of terrorism on fertility," Oxford Economic Papers, 67(1), [11] Berrebi, C. and H. Yonah (2016): "Terrorism and philanthropy: the e ect of terror attacks on the scope of giving by individuals and households," Public Choice, 169(3), [12] Berrebi C. and E.F. Klor (2010): "The Impact of Terrorism on the Defence Industry," Economica, 77(307), [13] Berrebi, C. and E. F. Klor (2006): "On Terrorism and Electoral Outcomes," Journal of Con ict Resolution, 50(6), [14] Blanch ower, D.G. and Oswald, A.J. (2008). Is Well-being U-Shaped over the Life Cycle?, Social Science & Medicine, 66: [15] Blomberg, SB, Hess G, and Orphanides, A (2004): "The Macroeconomic consequences of terrorism", Journal of Monetary Economics, 51(5), [16] Blomberg, S.B. and G.D. Hess (2006): "How Much Does Violence Tax Trade?", Review of Economics and Statistics, 88 (4), [17] Blunk, S.S., Clark, D.E. and McGuibany, J.M. (2006): "Evaluating the Long-Run Impacts of the 9/11 Terrorists Attacks on US Domestic Airline Travel", Applied Economics, 38(4), [18] Brown, J.R., J.D. Cummins, C.M. Lewis and R.Wei (2004). "An Empirical Analysis of the Economic Impact of Federal Terrorism Reinsurance." Journal of Monetary Economics 51:1, [19] Brown, G., J. Gathergood and J. Weber (2017): "Relative Rank and Life Satisfaction: Evidence from US Households", SSRN electronic journal. [20] Chen, A.H. and T.F. Siems (2004): "The E ects of Terrorism on Global Capital Markets", European Journal of Political Economy, 20(2), ; Chen and Siems nd that the e ect of major terrorist attacks on capital markets, on average, lasts from one to three days [21] Chernick, H. and Haughwout, A.F. (2006): "Tax Policy and the Fiscal Cost of Disasters: NY and 9/11", National Tax Journal, 59: [22] Clark, A., and Oswald, A. (1994): "Unhappiness and unemployment", Economic Journal, 104, [23] Clark, A.E., P. Frijters, and M.A. Shields (2008): "Relative Income, Happiness, and Utility: An Explanation for the Easterlin Paradox and Other Puzzles", Journal of Economic 44

46 Literature, 46: [24] Clark, A., O. Doyle and E. Stancanelli (2017): "The Impact of Terrorism on Well-being: Evidence from the Boston Marathon Terrorist Bombing", Geary WP2017/08 [25] Cummins, J. David and Christopher M. Lewis (2003): "Catastrophic Events, Parameter Uncertainty and the Breakdown of Implicit Long-Term Contracting: The Case of Terrorism Insurance." Journal of Risk and Uncertainty 26:2-3, [26] Decancq, K., M. Fleurbaey & E. Schokkaert (2015): "Inequality, income, and well-being". in: Anthony B. Atkinson & François Bourguignon (eds.) Handbook on Income Distribution, Volume 2A. Elsevier [27] Della-Vigna, S., Enikolopov, R., Mironova, V., Petrova, M and Zhuravskaya. E. (2014): "Cross-Border Media and Nationalism: Evidence from Serbian Radio in Croatia", American Economic Journal: Applied Economics, 6(3), [28] Dolan, P., T. Peasgood and M. White (2008). Do We Really Know What Makes Us Happy?A Review of the Economic Literature on the Factors Associated with SWB, Journal of Economic Psychology, 29: [29] Draca, M., S. Machin and R. Witt (2011),: "Panic on the Streets of London: Police, Crime and the July 2005 Terror Attacks", American Economic Review, 101(5), [30] Drakos, K. (2004): "Terrorism-Induced Structural Shifts in Financial Risk: Airline Stocks in the Aftermath of the September 11th Terror Attacks." European Journal of Political Economy 20:2, [31] Durante, R. and Zhuravskaya, E. (2018): "Attack When the World Is Not Watching? International Media and the Israeli-Palestinian Con ict", Journal of Political Economy, forthcoming. [32] Easterlin, R. (1974). Does economic growth improve the human lot? Some empirical evidence. In David, R. and Reder, R. (Eds.), Nations and Households in Economic Growth: Essays in Honor of Moses Abramovitz. New York: Academic Press. [33] Easterlin, R.A. (1995). Will raising the incomes of all increase the happiness of all? Journal of Economic Behavior and Organization, 27: [34] Egger, P. and M. Gassebner (2005): "International terrorism as a trade impediment?", Oxford Economic Papers, 67(1), [35] Enders, W. and Sandler, T. (1996): "Terrorism and Foreign Direct Investment in Spain and Greece", Kyklos, 49(3), [36] Enders, W., Sandler, T. and Parise, G.F. (1992): "An Econometric Analysis of the Impact of Terrorism on Tourism", Kyklos, 45, 1992,

47 [37] Farzanegan, M.R., T. Krieger and D. Meierrieks (2016): "Does terrorism reduce life satisfaction?", Applied Economics Letters [38] Feddersen, J., R. Metcalfe and M. Wooden (2016): "Subjective wellbeing: why weather matters", Journal of the Royal Statistical Society A, 179(1), [39] Ferrer-i-Carbonell, A. and Frijters, P. (2004): "How important is methodology for the estimates of the determinants of happiness?", Economic Journal, 114: [40] Fleche, S. (2017): "The Welfare Consequences of Centralization: Evidence from a Quasi- Natural Experiment in Switzerland", r&r for the Review of Economics and Statistics [41] Fleischer, A. and Buccola, S. (2002): "War, Terror and the Tourism Market in Israel", Applied Economics, 34(11), [42] Frey, B. and A. Stutzer (2002). What can economists learn from happiness research? Journal of Economic literature 40(2), [43] Frey, B.S., Luechinger, S. and Stutzer, A. (2007): Calculating Tragedy: Assessing the Costs of Terrorism. Journal of Economic Surveys, 21(1), [44] Frey, B.S., Luechinger, S. and Stutzer, A. (2009): The life satisfaction approach to valuing public goods: The case of terrorism", Public Choice, 138, [45] Friedland, Nehemia and Merari, Ariel. The Psychological Impact of Terrorism: A Double- Edged Sword. Political Psychology, December 1985, 6(4), pp [46] Frijters, P. and B.M.S. van Praag (1998): "The e ects of climate on welfare and well-being in Russia", Climatic Change, 39, [47] Galea, S., Ahern, J., Resnick, H., Kilpatrick, D., Bucuvalas, M., Gold, J. and Vlahov, D. (2002): "Psychological sequelae of the September 11 terrorist attacks in New York City", New England Journal of Medicine, 346: [48] Glaeser, E.L. and Shapiro, J.M. (2002): "Cities and Warfare: The Impact of Terrorism on Urban Form", Journal of Urban Economics, 51(2), [49] Gould, E. D. and Klor, E. F. (2010): "Does Terrorism Work?, Quarterly Journal of Economics 125(4), [50] Guidolin, M. and E. La Ferrara (2010): "The economic e ects of violent con ict: Evidence from asset market reactions", Journal of Peace Research, 47 (6), [51] Gupta, S, Clements, B, Bhattacharya, R, and Chakravarti, S (2004): "Fiscal consequences of armed con ict and terrorism in low and middle income countries, European Journal of Political Economy, 20(2),

48 [52] Henning F. and O. Listhaug (2013): "It can happen here: the impact of the Mumbai terror attacks on public opinion in Western Europe", Public Choice 156, [53] Jones, B. F. and Olken, B. A. (2009): "Hit or Miss? The E ect of Assassinations on Institutions and War, American Economic Journal: Macroeconomics 1(2), [54] Jusot, F., S. Tubeuf and A. Trannoy (2013): "E ort or circumstances: how important is their correlation for inequality of opportunity in health?", Health Economics, 22(12): [55] Kahneman, D. and Krueger, A.B. (2006): "Developments in the Measurement of Subjective Well-Being", Journal of Economic Perspectives, 22, 3-24 [56] Kamstra, Mark J.; Kramer, Lisa A. and Levi, Maurice D. Winter Blues: A SAD Stock Market Cycle. Amerian Economic Review, March 2003, 93(1), pp // In additional analysis (not reported) we nd that, controlling for the weather variables, life satisfaction is lowest in the fall (September-November). This is likely a re ection of the well-known seasonal a ective disorder (SAD). See Kamstra et al. (2003) for discussion of SAD [57] Karolyi, A. and Martell, R. (2010): "Terrorism and the stock market", International Review of Applied Finance Issues and Economics, 2(2), [58] Kibris, A. (2011): "Funerals and elections: the e ects of terrorism on voting behavior in Turkey", Journal of Con ict Resolution, 55(2), [59] Kimball, M., H. Levy, F. Ohtake, and Y.Tsutsui (2006): "Unhappiness after Hurricane Katrina", NBER Working Paper No [60] Krueger A.B. and Schkade, D.A. (2008): "The reliability of subjective well-being measures", Journal of Public Economics, 92: [61] LaFree, G. and L. Dugan (2007): "Introducing the Global Terrorism Database", Terrorism and Political Violence, 19, [62] Levinson, A. (2012): "Valuing public goods using happiness data: The case of air quality", Journal of Public Economics, 96(9), [63] Metcalfe, R., N. Powdthavee, and P. Dolan (2011): "Destruction and Distress: Using a Quasi-Experiment to Show the E ects of the September 11 Attacks on Mental Well- Being in the United Kingdom", Economic Journal 121 (550), F81 F103 [64] Montalvo, J. G. (2011): "Voting After the Bombings: A Natural Experiment on the E ect of Terrorist Attacks on Democratic Elections", Review of Economics and Statistics 93(4), [65] Olken, B.A. (2009): "Do Television and Radio Destroy Social Capital? Evidence from Indonesian Villages", American Economic Journal: Applied Economics, 1(4),

49 [66] Oswald, A., & Wu, S. (2010): "Objective con rmation of subjective measures of human well-being: Evidence from the USA", Science, 327, [67] Pape, R. (2003): "The Strategic Logic of Suicide Terrorism", American Political Science Review 97(03), [68] Powdthavee, N. (2015): "Would you like to know what makes people happy? An overview of the data sets on subjective well-being", Australian Economic Review, 48, 3, [69] Rohner, D. and B.S. Frey (2007): "Blood and ink! The common-interest-game between terrorists and the media", Public Choice, 133 (1 2), [70] Romanov, D., A. Zussman, and N. Zussman (2012): "Does Terrorism Demoralize? Evidence from Israel", Economica 79, 313, [71] Rubin, G.J., Brewin, C.R., Greenberg, N., Simpson, J. and Wessely, S. (2005): "Psychological and behavioural reactions to the bombings in London on 7 July 2005: cross sectional survey of a representative sample of Londoners", British Medical Journal, 331: 606 [72] Salguero, J.M., A. Cano-Vindel, I. Iruarrizaga, P. Fernández-Berrocal, S. Galea (2011): "Trajectory and Predictors of Depression in a 12-Month Prospective Study after the Madrid March 11 Terrorist Attacks", Journal of Psychiatric Research 45(10), [73] Schlenger, W.E., Caddell, J.M., Ebert, L., Jordan, B.K., Rourke, K., Wilson, D., Thalji, L., Dennis, M., Fairbank, J.A. and Kulka, R.A. (2002): "Psychological Reactions to Terrorist Attacks: Findings From the National Study of Americans Reactions to September 11", Journal of the American Medical Association, 288: [74] Senik, C. (2004): "When Information Dominates Comparison: Learning From Russian Subjective Panel Data", Journal of Public Economics, 88, [75] Senik, C. (2005): "Income distribution and well-being: what can we learn from subjective data?", Journal of Economic Survey, 19(1), [76] Silver, R.C., Holman, A., McIntosh, D.N., Poulin, M., Gil-Rivas, V. (2002): " Nationwide Longitudinal Study of Psychological Responses to September 11", Journal of the American Medical Association, 288: [77] Straetmans, S., Verschoor, W.F.C. and Wol, C. (2008): "Extreme US Stock Market Fluctuation in the Wake of 9/11", Journal of Applied Econometrics, 23: [78] Stutzer, A. (2004): "The role of income aspirations in individual happiness", Journal of Economic Behavior & Organization, 54(1), [79] Valiño, A., M. Buesa, and T. Baumert (2010): "The economics of terrorism: an overview of theory and applied studies", in M. Buesa and T. Baumert (Eds.), The Economic Repercussions of Terrorism. Oxford, UK: Oxford University Press,

50 [80] Vorsina, M., M. Manning, C.M. Fleming, C.L. Ambrey and C. Smith (2015): "The welfare cost of terrorism, Terrorism and Political Violence [81] Weimann, G. (2005): "The Theater of Terror", Journal of Aggression, Maltreatment & Trauma, 9:3-4, [82] Winkelmann, L., and Winkelmann, R. (1998). Why are the unemployed so unhappy? Evidence from panel data, Economica, 65: [83] Yanagizawa-Drott, D. (2014): "Propaganda and con ict: Evidence from the Rwandan genocide", Quarterly Journal of Economics, 129(4),

51 A Appendix A.1 Descriptive Statistics Table A.1: Descriptive Statistics Germany United Kingdom Switzerland Russia Federation Australia USA All Countries GSOEP ( ) BHPS ( ) SHP ( ) RLMS ( ) HILDA ( ) PSID ( ) Life satisfaction: country specific scale (a) (1.754) (1.260) (1.347) (1.124) (1.440) (0.867) (2.285) Life satisfaction: 0 10 harmonized scale (b) (1.754) (1.563) (1.347) (2.343) (1.440) (1.758) (1.992) Life satisfaction: 1 5 harmonized scale (c) (1.060) (1.075) (0.890) (1.124) (0.950) (0.867) (1.113) Female (=1) (0.500) (0.498) (0.496) (0.495) (0.499) (0.466) (0.499) Age (15.612) (15.650) (15.236) (15.894) (15.518) (14.074) (15.639) Married (=1) (0.490) (0.463) (0.487) (0.471) (0.473) (0.499) (0.482) Household size (1.231) (1.328) (1.392) (1.486) (1.442) (1.555) (1.361) Household income (2011 USD) (d) 45,960 55, ,460 12,930 67,630 69,770 48,800 (37,310) (42,610) (96,420) (38,020) (62,620) (106,220) (55,440) Currently employed (=1) (0.485) (0.474) (0.444) (0.482) (0.456) (0.450) (0.477) Health Status (5 point scale) (0.928) (0.935) (0.635) (0.693) (0.949) (1.036) (0.915) #Observations (individual x time) 298, ,600 40, , ,164 24, ,384 Source: Own calculations from GSOEP, BHPS, SHP, RLMS, HILDA and PSID. The samples retains individuals aged years old and excludes firstgeneration migrants. Standard deviations are in parentheses. (a) Measured on a 11 points scale in GSOEP, SHP and HILDA, a 7 point scale in the BHPS, a 5 point scale in PSID and RLMS. (b) The scales are harmonized into a 11 point scale (0 10) (c) The scales are harmonized into a 5 point scale (1 5) (d) Household income is converted into USD using yearly average exchange rates. 50

52 Table A.2: Portrait of Terror: Breakdown of Fatalities and Incidences ( ) Total # events Mean # Mean # fatalities incidences per country per country & year & year Explosion bombing Characteristics of the attack Suicide bombing Longer than 24 hours Multiple attacks % fatalities % incidences % fatalities % incidences % fatalities % incidences % fatalities % incidences All Terror Events 84, (40.60) (13.02) (0.32) (0.22) (0.21) (0.07) (0.11) (0.08) (0.18) (0.19) Daily Mean Terror by Years , (28.33) (10.06) (0.25) (0.24) (0.01) (0.01) (0.08) (0.05) (0.05) (0.23) , (26.43) (13.44) (0.29) (0.22) (0.12) (0.03) (0.10) (0.06) (0.08) (0.21) , (19.76) (6.53) (0.35) (0.26) (0.16) (0.08) (0.20) (0.13) (0.22) (0.28) , (23.63) (4.17) (0.36) (0.23) (0.32) (0.16) (0.12) (0.12) (0.19) (0.20) , (22.99) (9.11) (0.33) (0.17) (0.27) (0.06) (0.12) (0.08) (0.20) (0.16) , (37.52) (12.15) (0.22) (0.11) (0.19) (0.04) (0.10) (0.04) (0.21) (0.14) Daily Mean Terror by World Regions North and Central America 2, (30.51) (9.61) (0.29) (0.23) (0.11) (0.04) (0.08) (0.07) (0.12) (0.22) South America 7, (38.95) (11.85) (0.29) (0.24) (0.13) (0.04) (0.10) (0.10) (0.14) (0.22) East and Southeast Asia 7, (38.64) (13.55) (0.32) (0.21) (0.21) (0.07) (0.12) (0.08) (0.18) (0.20) South and Central Asia 25, (36.03) (12.96) (0.31) (0.20) (0.22) (0.07) (0.11) (0.07) (0.18) (0.17) Western and Eastern Europe 6, (43.38) (11.90) (0.31) (0.25) (0.15) (0.05) (0.10) (0.07) (0.14) (0.25) Russia and New Ind. Countries 2, (46.69) (10.22) (0.33) (0.22) (0.22) (0.07) (0.14) (0.10) (0.18) (0.17) Middle East and North Africa 23, (42.85) (13.83) (0.32) (0.21) (0.23) (0.09) (0.10) (0.07) (0.19) (0.17) Sub Saharan Africa 7, (47.21) (12.06) (0.30) (0.21) (0.19) (0.06) (0.12) (0.08) (0.19) (0.19) Notes: Authors' own calculations using the Global Terror Dataset Fatalities represent the number of persons killed. Incidences are the counts of all terror incidences Standard deviations are in parentheses. 51

53 Figure A.1: Global Portrait of Terror 52