Tourism and Troubles: Effects of Security Threats on the Global Travel and Tourism Industry Performance

Literature on the effects of security threats such as terrorism, political instability, and geopolitical power-plays on travel and tourism has produced mixed results with scant attention paid to the spillover effects on the tourism economy (e.g., employment, leisure expenditure, travel, and tourism services’ contribution to gross domestic product). This study provides a conceptual framework for the transmission of direct, indirect, and induced spillover effects of security threats on travel and tourism service industries. It uses rigorous methodological design and non-spatial and spatial panel-data analyses to examine the effects of security threats on tourism demand and economy. The conceptual framework and results of spatial panel data provide novel insights into security threats’ spillover effects on spatial inter-connectivity in the tourism service industry. The results show that security threat indices have significant negative impacts on tourist receipts, but they also contribute positively to employment, leisure expenditure, and tourist arrivals. Our conceptual model and substantial findings will inform both policymakers and future research.


Introduction
Tourism is one of the largest contributors to gross domestic product (GDP), economic development, and job creation (UNWTO, 2018).Jus and Misrahi (2021) report that prior to the COVID-19 pandemic, the travel & tourism sector's direct, indirect, and induced impacts contributed US$9.2 trillion to the global economy and supported 334 million jobs in 2019, while it directly contributed 10.4% of the world's GDP.Tourism also influences the growth of tourism-led satellite service activities and the global economy (Frechtling, 2010;Sinclair, 1998;Smeral, 2006).Moreover, the development of tourism demand is one of the key drivers of service growth and trade development (Kim et al., 2006;C.-C. Lee & Chang, 2008).
Nonetheless, the global travel and tourism (T&T) service sector has been afflicted with persistent and episodic security threats over the past two decades (Aran˜a & Leo´n, 2008;Goldman & Neubauer-Shani, 2017;Pizam, 1999;Saha & Yap, 2014), including the recent COVID-19 pandemic (Farzanegan et al., 2021).For example, the global service economy, especially the T&T service industry, has suffered gigantic financial losses and capacity dormancy due to travel restrictions and facility closures, among others, during the COVID-19 pandemic (Bausch et al., 2021;Wolf, 2020).While this study is not based on COVID-19 due to the paucity of panel data, its framework is extendable to the current pandemic and is amenable to COVID-19 policies and research paradigms.
There is evidence that direct and spillover effects of different security threats (e.g., terrorism, political instability, war, etc.) tend to slow growth in T&T service sectors (Walters et al., 2019).For instance, UNWTO (2022) reports that a prolonged conflict between Russia and Ukraine could translate into a loss of US$14 billion in tourism receipts globally in 2022.Furthermore, Koch (2022) argues that the Russian war on Ukraine will have global ripple effects across many industries, especially on travelers, travel agencies, airlines, and cruise operators.However, the wider effects of global and destinationspecific security factors, such as defense capability (e.g., military expenditure, nuclear, and heavy weapons, weapons imports, exports, and armed services personnel) and geopolitical power-plays (e.g., deaths from external conflict, displaced people, UN peacekeeping funding), on the T&T service sector (e.g., leisure tourism spending, contribution to employment, contribution to GDP) have been overlooked in prior studies.While the link between country risk factors and economic activities is becoming increasingly evident, C.-C. Lee and Chen (2021) point out that there is a surprising lack of empirical evidence on country risk factors and tourism development.
Furthermore, some questions regarding the lack of conceptual framework to assess the direct, indirect, and long-term spillover effects on tourism, as well as the extent to which security threats impact the tourism industry, remain unanswered.Is there a conceptual framework to show the direct, indirect, and induced spillover effects of security threats on the T&T service industry?To what extent do security threats affect the tourism economy (e.g., T&T's direct contribution to employment, leisure tourism spending, and GDP)?Our systematic review of classical studies in the fields of global security, tourism economy, and tourism demand presents a concise taxonomy of the literature and theoretical gaps in the existing literature (see Appendix A).The review table in Appendix A also juxtaposes our study with germane research focusing on typology of security threats influencing T&T industry performance.Consequently, the review enables us to identify new variables and postulates a holistic integration of new indicators to model the relationships between security threats, tourism demand, and the tourism economy.
We argue that relationships between T&T and security threats exhibit complex geographical and locational spillover externalities that ripple over regions, countries, and continents (Neumayer, 2004;Neumayer & Plu¨mper, 2016).Tobler's (1970) first law of geography sets forth the paradigm of spatial inter-connectivity, which is inherent in global tourism and security incidents.Marrocu and Paci (2013, p. 72) show that spatial inter-connectivity ''fully accounts for spatial dependence, generally featured by tourism flows, which exhibits a quite complex pattern since tourists' movements are affected not only by geographical distance and by origin and destination-specific features, but also by the characteristics of neighboring locations at both origin and destination.''Similarly, terrorism incidents, which are location specific, adversely affect the tourism-sector performance of neighboring locations and faraway destinations (Groizard et al., 2022).
Arguably, security threats have hampered interconnected service supply chains, such as restaurants, transport, hotel, leisure and sport recreation activities, employment, travel services, manufacture of beverages and foods, logistic service, tax revenues, capital flows for investment in new hotel construction, retailing businesses, the creative industry, etc.However, existing literature overlooks the empirical modeling of spatial interconnectivity related to security threats and the tourism economy.Although international tourism has received some attention at country and international levels, intercountry spatial units have largely been ignored (Krajnˇa´k, 2021).Further, Duan et al. (2022) emphasized the importance of multi-country or multi-regional studies to highlight the spillover, ripple, or contagion effects based on geographical locations.Therefore, this study addresses the above oversights.
This study makes four major contributions.First, our study synthesizes and highlights the integration of security related theoretical constructs (i.e., defense capability, geopolitical power-plays, police, and security services) which have received scant attention.Scholars, including Aran˜a and Leo´n (2008), Corbet et al. (2019), and A. Liu and Pratt (2017), call for a further examination of the relationship between security and tourism issues.Specially, C.-C. Lee andChen (2021, p. 1446) stated that ''knowledge is limited as to whether all types of country risks exhibit similar impacts on tourism development.''Furthermore, Fourie et al. (2020, p. 209) claimed that ''little is known about how safety and security differences between countries may affect the choices of tourists to travel.'' In response to the abovementioned calls, this study draws on tourism security theory proposed by Pizam and Mansfeld (2006) and provides initial international evidence regarding the effects of different country risk variables on tourism development variables.This study includes 24 security threats-related explanatory variables classified into six typological constructs (i.e., perception of crime, political environment, terrorism, the impact of police and security services, defense capability, and geopolitical power-plays), and five tourism outcome variables into two dependent constructs: tourism demand (i.e., tourism arrivals and tourism receipts) and tourism economy (i.e., employment, leisure expenditure, and T&T services' contribution to GDP).Hence, our study differs from prior studies (e.g., Aran˜a & Leo´n, 2008;Corbet et al., 2019;C.-C. Lee & Chen, 2021;A. Liu & Pratt, 2017;Saha & Yap, 2014) as we test the universality of the proposed tourism security theory by analyzing wider tourism economy-related variables.
Second, this study advances the literature on the tourism economy by using new outcome variables (namely leisure tourism spending, T&T's direct contributions to employment, and GDP), which remains unexplained by security threats issues.Most existing studies focus solely on the theory of international tourism demand variables (e.g., tourist arrivals and tourist receipts) (Crouch, 1994;Dogru et al., 2017;A. Liu & Pratt, 2017;Pizam, 1999;Saha & Yap, 2014).While tourism is regarded as one of the main contributors to destination economies GDP (C.-C. Lee & Chang, 2008;C.-C. Lee & Chen, 2021;Oh, 2005), scholars have not paid appropriate attention to the effects of security threats on the three indicators of the tourism economy (see Appendix A).In fact, Fourie et al. (2020, p. 212) ascertain that ''the previous literature on the effect of security threats on tourism mainly explores the impact of terrorism on inbound tourism.''Moreover, selection bias seems to exist across the tourism and terrorism literature as most of the studies predominantly emphasize on United States, Europe, Central Asia, or MENA countries (Duan et al., 2022;Krajnˇa´k, 2021).As a result, we extend and deepen the extant literature by analyzing 161 country data on how diverse security threats influence the tourism economy with significant novel findings in the global tourism arena.Subsequently, our study augments extant knowledge by not only enlightening what is known, but also by putting forward novel arguments related to global insecurity literature in the T&T field.
Third, this study proposes a conceptual framework offering schematic channels for the transmission of global security threats and their influence on the spatial interconnectivity of the T&T industry (see Figure 1).The framework provides comprehensive insights into the dimensions of security threats and related direct, indirect, and long-run spillover effects on T&T service industries.It also hypothesizes that global security threat indices have repercussions on service industry employment, service consumers' leisure-based expenditure, tourist receipts, travel services, and growth.
Moreover, this framework envisions that security threats have a spillover effect on tourism-related services.Tourism and allied service sectors tend to bear the brunt of the economic effects of insecurity and instability (Dekimpe et al., 2016); thus, the spillover effects of global security threats can be significant, as their impact is not limited only to the aforementioned business sectors, but also affect global economic growth (Karl et al., 2015;Khan, 1997).Numerous studies have explored the relationship between insecurity and tourism demand (e.g., Coshall, 2003;Goldman & Neubauer-Shani, 2017;Saha & Yap, 2014), but scholars have neither proposed a conceptual framework nor empirically examined the impact of security threats on the tourism economy using a global sample of panel time-series data and robust econometric methodology.Subsequently, this study attempts to remedy the aforementioned gaps in tourism outcomes and spillover literatures by using a spatial panel model to reveal new insights into long-run spillover effects on the T&T industry.However, due to limitation on availability of data for mediating industry/country-level variables in the framework, this study focuses on direct and long-term spillover effects of outcome variables due to security threat covariates.In addition, the proposed framework can be used to determine cybercrime and COVID-19's impact on global service industries.We believe that the framework developed in this study is the first attempt to advance and extend the COVID-19 research agenda put forward by Bausch et al. (2021).
Finally, from a methodological perspective, we used a unique three-pronged combination of the existing literature (see Appendix A), our knowledge of possible relationships between the variables, and automatic (machine learning) variable selection methods that enable us to select covariates, specify, and test our empirical models (Efron et al., 2004;Tibshirani, 2011) to produce reliable estimates.Thus, we advance the tourism security theory with robust new empirical evidence.We also make significant contributions to the theory of global security threats and the T&T economy.
The remainder of this paper proceeds as follows.The review of global security threats and the T&T economy section discusses the theoretical underpinnings and related literature on terrorism-led tourism and its spillover effects.The section on methodology, data, and preliminary tests presents the models, describes the data and sample, and offers a preliminary analysis and tests.The results and discussion section presents the empirical results of the models and discusses the spillover effects of security threats.Finally, we present the conclusions and policy implications.
Table 1 and Figure 2 summarize some of the major terrorist acts and trends in attacks and their fatalities.Over the past two decades, real and perceived public security incidents and their aftermaths have drawn widespread attention in traditional and social media; in turn, this has led to an exponential growth in visibility and awareness of security threats to tourists and travelers (Birkland, Akamavi et al. 2004;Jetter, 2017;Walters et al., 2019).Several studies have shown the tourism industry's sensitivity to securityrelated news and hasty changes in security arrangements, which are the most important determinants of destination choice for potential tourists/travelers (Boakye, 2012;So¨nmez & Graefe, 1998b;Sullivan-Taylor & Wilson, 2009).Fourie et al. (2020) have shown that the multiplier effect of terrorism and insecurity on tourism is regressive due to travelers and tourists' risk perceptions of destinations.Other scholars have argued that terrorist attacks and insecurity impede growth in T&T service industries.For instance, Aran˜a and Leo´n (2008) and Walters et al. (2019) reported that hotel occupancy levels, restaurant takings, airline passenger numbers, and retail revenues all decline when there are terrorism and other security concerns.In addition, security threats have negative effects on prospective tourists' perceptions of comfort, safety, and leisure choices of a destination country (Li, Yang et al., 2021).The negative effects are not only limited to the time of the crisis but also have prolonged effects long after the incidents (Cavlek, 2002).In addition, visitors' perceptions of security threats have spillover and halo effects on neighboring countries that are not directly impacted by the conflict or crisis (Lepp & Gibson, 2003).
Contrarily, a few scholars have suggested that terrorism does not always hurt tourism (Morakabati & Beavis, 2017;Yaya, 2009).For instance, Saha and Yap (2014) revealed that because people are inquisitive by nature, tourism demand tends to increase up to a threshold following terrorism incidents in nations with low to moderate political risk.Furthermore, global terrorism has generated a new and unique dimension to tourism on the so-called ''dark side'' of the tourism spectrum (P.R. Stone, 2012, p. 1), referred to as dark tourism (Lennon & Foley, 2000;P. R. Stone, 2006;Strange & Kempa, 2003), morbid tourism (P.R. Stone, 2012), atrocity heritage tourism (Kang et al., 2012), thanatological framework and thanatourism (Light, 2017;P. Stone & Sharpley, 2008), grief tourism (Lewis, 2008), sacred memorial sites (Podoshen & Hunt, 2011), popular shrine/altar and ritual space (Iliev, 2020), from lieux de me´moire to noeuds de me´moire (Fuggle, 2020), victimhoodscape or thanatoptic/ dark heritage (Hooper & Lennon, 2016), and anamnesis  tourism (Seaton, 2002).This paradigm has a common denominator of the aforementioned themes, which integrates them to assign the basis for a common thread for dark tourism.For instance, Jacobs (2004, p. 311) noted that ''Ground Zero became a religious shrine for a dark pilgrimage with the placement of other sacred objects at the site-rosary beads, religious medals, and memorial candles.''Geopolitical risks such as war, military-related tension, and nuclear threats contribute to a decrease in tourist arrivals and demand (Demir et al., 2019;Tiwari et al., 2019).The Turkish invasion of Cyprus in 1974 and the Syrian and Iraq war demonstrate how badly geopolitical tensions can affect tourism demand and the regional T&T economy (Farmaki, 2017;Mehmood et al., 2016;Sharpley, 2003).In addition, recent tensions on the Sino-Indian border have had a negative impact on tourism in Ladakh, Manali, and Lahaul-Spiti-major tourist destinations in India-where tourists could not enter the region (Gettleman et al., 2021).Recently, Parkin and Ratnaweera (2022) argue that the current Russia-Ukraine war has a severe fallout with an unwelcomed twist, which causes a huge economic disruption and an austere effect on T&T services.In fact, Bu¨lbu¨log˘lu (2022) reports that Turkish tourism expects a 30% loss due to the Russia-Ukraine war.Koch (2022) also conveys that this war affects the travel industry (e.g., airlines, cruises) with longer routes and distances, and greater fuel costs.Additionally, evolving sensitive geopolitical pressures facing China and Taiwan may deeply threaten T&T industry performance (Gillen & Mostafanezhad, 2019;Lim, 2012).Moreover, Balli et al. (2019) have reported that, while geopolitical risk factors adversely affected tourism demand in some countries, others remain unaffected by the risk of a geopolitical power-play.Furthermore, substantial geopolitical tensions, political turmoil, recent coup d'e´tats, and rising and on-going terrorism attacks in the Sahel region present evolving real security threats to T&T industry performance in West Africa, East Africa, and surrounding regions (Benedikter & Ouedraogo, 2019;Dowd & Raleigh, 2013;Gaibulloev & Sandler, 2011).However, the impact of geopolitical risks on the tourism economy remains understudied (Akadiri et al., 2020;Demiralay & Kilincarslan, 2019;Gozgor et al., 2022) and needs further investigation, especially from a global perspective.
In addition, there is an ongoing debate in the field of defense economics literature regarding the impact of public expenditure on defense capability, economic growth, and tourism inflows.While some studies provide evidence that military expenditure positively impacts economic growth, other studies suggest that it hinders economic growth through various channels.Scholars (e.g., J. P. Dunne et al., 2005;Mylonidis, 2008;Pieroni, 2009) have argued that increased military expenditure can impede economic growth by constraining other government expenditures or crowding out investment.Moreover, Bottom and left axis: explaining time and amount of terrorist incidents, injuries and deaths in thousands; top and right axis: explaining years, amount of tourist arrivals, and economic cost of terrorism in billions (log scale).Khalid et al. (2020) have shown that increased military expenditure can hurt investment in the tourism industry and international tourism inflows.However, military spending can also enhance economic growth and employment (e.g., Wijeweera & Webb, 2009;Yildirim et al., 2005).Nassani et al. (2017) noted that an increase in military spending and arms exports has a positive impact on net tourism receipts, and thus significantly influences tourism growth.In addition, a similar notion was put forth by Yildirim et al. (2005, p. 294), who report ''that military expenditure enhances economic growth in the Middle Eastern countries.''However, tourism scholars have yet to examine the effect of defense capabilities on tourism demand and the economy.
Furthermore, law enforcement officials/forces can play a pivotal role in preventing future terrorist attacks and restoring communal faith and a destination's image (Paraskevas & Arendell, 2007;So¨nmez et al., 1999).Studies have ascertained that an increase in the number of security forces and police services in a tourist destination can help elevate tourists' perception of a destination's attractiveness (de Albuquerque & McElroy, 1999;Tyagi et al., 2016) and safety (Barker & Page, 2002;Tarlow & Santana, 2002).An increase in the presence of security forces can also enhance tourists' perception of police effectiveness (George, 2003;Tyagi et al., 2016), crime prevention (Mawby, 2014), and security (Cruz-Mila´n et al., 2016).However, a few studies have argued that an overt display of security has an opposite effect on tourists' perception of safety.For instance, Boakye's (2012) study in Ghana found that the prominence of law enforcement agencies made tourists feel insecure and served as a constant reminder of the need to remain vigilant.
Terrorism and conflicts are major security threats that have spillover effects on service industries, leading to economic losses (Abadie & Gardeazabal, 2008;O ¨cal & Yildirim, 2010).Greenbaum et al. (2007) noted that terrorist incidents decrease the number of firms and employment in the year following an attack.For instance, Causevic and Lynch (2013) showed that political conflict in Croatia, Serbia, Bosnia, and Herzegovina had negative effects on economic and employment indicators.However, Bagchi and Paul (2018) argued that the increase in terrorist activities increased military expenditure, which contributed to a decline in youth unemployment in MENAP countries (Middle East, North Africa, Afghanistan, and Pakistan).In general, security threats such as terrorism, crime, political uncertainty, and war can have negative direct, indirect, and long-run spillover effects on employment and tourist receipts.Hence, the next section focuses on the empirical investigation of security threats' effects on spatial inter-connectivity in T&T service industries.

Methodology
We posit that the outcomes of a country's T&T industry performance (i.e., tourism demand and tourism economy) are influenced by prevailing security factors and the perceived atmosphere of peace, criminality, and terrorism incidents.Accordingly, key T&T demand variables (tourist arrivals and tourist receipts) are determined by many country-specific public safety, security, peace, and terrorism-related factors (e.g., the perception of criminality in the country, homicide rates, internal conflicts, political instability, terrorist incidents, etc.).While the existing literature has used a variety of empirical approaches to analyze the relationship between tourism demand and security/safety-related covariates, the economic variables for the T&T sector (e.g., tourist spending, tourism contribution to employment, and economic growth) have largely been overlooked.
We argue that the inconclusiveness of findings in the literature emanate from three commonly unresolved econometric and design issues which we address in this study.First, we posit that the chosen econometric and empirical analyses in earlier papers often tend to ignore omnipresent problems of endogeneity and unobserved heterogeneity in the models which are principally due to omitted variables, simultaneity bias, and measure errors in terrorism and security-related covariates.The problem is also evident in recent strands of the literature.For example, a recent study by Seabra et al. (2020) used vector autoregressive models (VAR) to establish the significant and strong effect of terrorist incidents in European countries on tourist arrivals in Portugal.
The VAR framework provides flexibility in examining the relationships between variables; however, their rather complex atheoretical approach to modeling multivariate relationships is a major drawback.VAR systems often require researchers to determine the long-run relationships before ensuing short-run changes that policymakers may want to know.Feridun (2011), Wang (2009), andZhang et al. (2021) used the lag-error correction type autoregressive distributed model (ARDL) to predict tourism recovery from a crisis environment.ARDL models possess the advantages of VARs, render a more general determination of lag structures, and are amenable to panel data.
However, the ARDL framework tends to perform poorly in the presence of stochastic trends, as it tends to model random trends at the expense of the underlying relationship.Saha et al. (2017) and Antonakakis et al. ( 2019) applied panel VAR and country fixed-effect models to large panel-data models, respectively, to study the determinants of tourist arrivals and tourism effects on economic growth.While the panel VAR and standard fixed-effect models are well executed in both studies, the two approaches are weak in dealing with panel structures afflicted by problems of heteroscedasticity in the panel and cross-sectional dependence (Greene, 2018;Pesaran, 2007Pesaran, , 2015)).
Second, most empirical studies tend to display evident inertia in sticking with a narrow set of established and suspected explanatory variables in their modeling frameworks.To address this challenge, we started with large models containing a large set of relevant explanatory variables, most of which have not been used in previous studies.We proceeded to reduce the models using the least absolute shrinkage and selection operator (LASSO) to reach the smallest possible subset for the best explanatory content (Efron et al., 2004;Ghysels & Marcellino, 2018;Tibshirani, 2011) (see FRS and LASSO on p.15).Consequently, in a data-rich environment, we developed models that are large enough to capture pertinent and new explanatory variables beyond those used in previous literature (e.g., Antonakakis et al., 2019;Saha et al., 2017;Seabra et al., 2020) but also parsimonious and internally valid (Ha¨nnika¨inen, 2017;Wedel & Kannan, 2016).
Finally, we used one of the largest panel-data sets, consisting of a global sample of 161 UN countries, and a complete dataset panel spanning over 10 years (2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018)(2019) on country-level T&T industry performance and security threats.The coverage and comprehensiveness of the panel data proved that our data can meet this study's objectives.We have used a triumvirate of research designs based on the extant literature, our understanding of priori statistical and economic relationships between the putative variables, and ''machine learning'' model reducing computer-intensive methods (LASSO) to develop estimable, robust, and parsimonious models.Morever, we used the extended Bayesian information criterion to research the lambda shrinkage parameters for the models (Chen & Chen, 2008).

Data Collection
Table 2 presents a list of variables and their definitions, measurements, credible data sources, and some related studies.For example, our data sources are in line with prior studies such as World Bank Data (Goldman & Neubauer-Shani, 2017;Nassani et al., 2017), World Travel andTourism Council (Ca´rdenas-Garcı´a et al., 2015;Peeters & Eijgelaar, 2014), Economist Intelligence Unit (Demir et al., 2020;Gaventa & Barrett, 2012;Kilian & Hicks, 2013), and Global Terrorism Index (A.Liu & Pratt, 2017).Moreover, Table 2 also shows that this study encapsulates several variables (e.g., access to weapons, incarceration, external conflicts fought, displaced people, nuclear and heavy weapons, weapons imports, and reliability of police services) that have not been considered in previous studies (Corbet et al., 2019;A. Liu & Pratt, 2017).The data analysis was conducted using Stata v.16.

Modeling the Framework
Consider a panel-data structure with k distinct explained and explanatory variables fY it , X it = (x 1, it , x 2, it , . . ., x k, it )g, i = 1, . . .N, t = 1, . . ., T, where i denotes entities (countries) and t denotes time (in years).The baseline panel-data regression model used to establish the effects of the explanatory variables on the explained variables is as follows: where Y it denotes a panel T&T dependent (outcome) variable of the countries in our sample.Specifically, Y it is a N31 vector of international tourist arrivals, leisure tourism spending, international tourism receipts, direct contribution to employment, and direct contribution to GDP.
X denotes the N3K matrix of explanatory variables (our models contain the following independent variables: perceptions of criminality, homicide, access to weapons, incarceration, political instability, political terror, intensity of internal conflict, impact of terrorism, terrorism incidents, external conflicts fought, displaced people, UN funding for peacekeeping, relations between neighboring countries, military expenditure, nuclear and heavy weapons, weapons imports, and the reliability of police services, security officers and police).C FE denotes N À 1 country fixed-effect dummies (equal to 1 for the ith country and 0 otherwise); T FE denotes T À 1 time-fixed dummies (equal to 1 for the ith country and 0 otherwise); a i denotes country-specific fixed effects; and g t denotes or the action must be outside the context of legitimate warfare activities.
It captures the direct effects of terrorist-related violence, in terms of its physical effect, but also attempts to reflect the residual effects of terrorism in terms of emotional wounds and fear by attributing a weighted average to the damage inflicted in a year to a country.
Proposition: The standard fixed-effect estimator cannot be unbiased in the presence of unobserved endogeneity (see Appendix B).The presence of unobserved endogeneity in equation ( 1) leads to biased estimates of the coefficients.Consequently, we use a battery of econometric tests to identify the correct estimator underpinned by a battery of model specifications and diagnostic tests to enable us to correct the estimation method.In addition, econometric challenges such as heteroscedasticity, serial correlation, and cross-sectional dependence are inherent in country-level panel data (Driscoll & Kraay, 1998;Pesaran, 2007); the data used in this study are no exception.

FRS and LASSO
The LASSO model-reduction approaches use ''soft thresholding'' rules to order and select independent variables in a manner that reduces collinearity and increases parsimony by minimizing the sum of the square of the error term.Both the forward selection regression (FRS) and LASSO methods achieve model reduction by selecting a subset of variables that minimizes the residual sum of squares of regression.However, we use the LASSO because the approach holds substantial statistical advantages over FSR for model reduction and selection (see Burnham & Anderson, 2002;Harrell, 2001;Tibshirani et al., 2005).
From equation ( 1), the soft-thresholding model shrinkage approach, solves the following minimization problem: where l denotes a Lagrange multiplier that controls the magnitude of the penalty imposed on the model.Therefore, the larger the value of l the greater the penalty imposed on the model, including additional explanatory variables, and vice versa.F and C are functions of the residual sum of squares and b in equation, respectively.Here, For a given dependent variable in equation ( 1), the FSR procedure regresses x 1, it on y it and stores the residual ĵ1, it and then proceeds to search for an explanatory variable in X which has the highest correlation with ĵ1, it , say x 2, it .In the second step, it proceeds to regress ĵ1, it on, x 2, it and a new residual ĵ2, it .Thereafter, the iteration process continues until all explanatory variables in X are ranked.The FSR tends to retain fewer orthogonal variables in contrast to iterative hard-thresholding methods that tend to select a set of highly collinear variables (Bai & Ng, 2008;Bulligan et al., 2015;Ghysels & Marcellino, 2018).However, the LASSO regression starts with least angle regression (LARS) to delineate independent variables that are highly correlated with the dependent variables (Efron et al., 2004).However, while the LARS is indifferent to the sign of the correlation between y it and the candidate variable in X, the LASSO restricts the sign of the correlation, which prevents it from switching.For a model with independent M variables, LASSO operationalizes the model-reduction process by solving the problem in equation ( 2) as follows: For a large two-dimensional panel-data model, in equation (1), of the relationship between T&T demand/economy and terrorism/security entails several often collinear covariates, so we LASSO to shrink the models before conducting pre-estimation and specification tests and model estimation.Consequently, this study addresses the empirical challenges and possible anomalies associated with heteroscedasticity, serial correlation, and crosssectional and temporal dependence in panel data (Ammermann & Patterson, 2003;Chudik & Pesaran, 2015;Petersen, 2009).

Preliminary Analysis
Table 3 summarizes the statistics of dependent and independent variables.The standard deviations and coefficient variations confirm large degrees of variation within and between the variables.These properties lend support to further statistical investigations of the multivariate relationships among the variables in the modeling framework.Following from the model selection procedures, we conducted five pre-estimation diagnostic and specification tests: (i) fixed versus random effect delineation using the (Hausman, 1978) test, (ii) the Breusch-Pagan Lagrange multiplier test for significant panel effect versus no panel effect (Breusch & Pagan, 1980), (iii) a test for serial correlation in panel (Drukker, 2003;Wooldridge, 2010), (iv) the modified Wald test for heteroscedasticity in panel (Greene, 2018, p. 598), and (v) the Pesaran's crosssectional dependence test (Pesaran, 2007(Pesaran, , 2015)).These tests are vital for understanding the properties of the residuals of every study involving time-series cross-section data.Ignoring them often leads to the imposition of theoretically driven assumptions about residual properties and covariance matrices without empirical justification.Table 4 contains results of pre-estimation and model specification tests for the panel data.The results show a preference for the fixed-effect panel-data model (Hausman tests) over the random effect model, and further tests ruled out the ordinary least squares estimator for all five outcome variables (Breusch-Pagan Lagrange multiplier tests).The results further confirm autocorrelation and heteroskedasticity for all models (Wooldridge and modified Wald tests).Moreover, further tests revealed evidence of cross-sectional dependence in models for all the dependent variables (Driscoll & Kraay, 1998;Hoechle, 2007).Consequently, we apply the panelcorrected standard error (PCSE) estimator which uses the feasible generalized least squares estimator to robustly address the above shortcomings in our data (Doran & Kmenta, 1986;Parks, 1967).The PCSE produces robust inferences and is consistent in the presence of non- spherical errors that originate from serial correlation problems, heteroskedasticity, cross-sectional dependence, and a combination of the three problems that are typical in social, political, and economic variables (Bailey & Katz, 2011;Beck & Katz, 1995;Greene, 2018;Hoechle, 2007).
Assuming that t = 1, . . ., T i (where T t = T) and the error term, v it , in equation ( 1) is heteroskedastic and cross-sectionally dependent, the model can be written as Consequently, in the covariance matrix of a model with heteroskedastic and serially correlated error terms, the disturbance terms are where O is a block diagonal matrix (NT3NT) of the matrix N3N of contemporaneous covariance, S, along the diagonal line.The elements of S, Ŝ, are derived from OLS residuals equation ( 1): Therefore, the estimator of Ô can be obtained from the block diagonal matrix comprising the Ŝ matrices on the diagonal line.Assuming that the error terms are spherical, O = s 2 I where I is the identity matrix and our panel is balanced where N is T3N a matrix of residuals and, therefore, O is obtained from where denotes the Kronecker (direct) product of matrix.However, in the case of an unbalanced and subsequently non-spherical residual, as in this study, the covariance matrix is Therefore, the PCSE estimator is obtained from computation of the square root of the diagonal elements as follows: Furthermore, Appendix C reports pairwise correlation coefficients to understand the nature of any relationships between the IV's.

Spatial Panel Model
We proceeded to address possible effects of spatial interdependence and spatial heterogeneity due to the geographical inter-connectivity of security threats' effect on tourism by augmenting our results with the spatial Durbin panel models (Anselin, 2003;Anselin & Rey, 2010;Elhorst, 2014;Tobler, 1970).This analysis enables us to understand externalities due to inherent spillover effects (Figure 1) (Anselin, 2003;Chhetri et al., 2017).Consequently, we used vectors of tourism sector dependent and independent variables outlined in equation ( 1) to specify the spatial Durbin model (SDM) as follows: where Y it , X it , b are as denoted in equation ( 1), z it denotes a vector of spatial and time effects; W is an N3N rownormalized spatial weights matrix depicting sample units, r it and u it denote the spatial parameters relating to dependent and independent variables, respectively, m i is the vector of spatial fixed effects, and a t is the vector of timefixed effects, e it are independently and identically distributed error terms for all country units with zero mean and variance s 2 .Notice that the m i and a t parameters can also be spatial random effects.The weighting matrix was derived in two stems.First, longitude and latitude coordinates were used to develop the matrix of distances D ij in kilometers of capital cities of all the countries in our sample.Second, we used D ij spatial weighting matrices, W, of countries in our sample (Drukker et al., 2013).The advantage of the spatial panel econometric approach lies in the ability to capture the extent to which neighboring countries' explanatory variables affect tourism outcome variables in our model (Partridge et al., 2012).In addition, the approach can address empirical challenges relating to non-randomly distributed error terms (Elhorst, 2014) and render spillover effects which are inherent in our framework.We brought the spatial dimension to our analyses to complement the PCSE results with spatial spillover effects.The estimations of spatial panel data used a combination of Stata in-built and user-written spatial panel-data commands (Belotti et al., 2017;Drukker et al., 2013).Our approach to spatial panel-data model estimation for this paper followed six steps.First, we augmented variables in our original model (used to obtain PCSE results) with geographical location variables in the form of longitude and latitude coordinates of capital cities of all the countries in our dataset.Second, we declared the data a spatial panel-data dataset.Third, we used longitude and latitude coordinates to generate distances in kilometers between the countries (Baum & Hurn, 2021).Fourth, we generated normalized inverse-distance weighting matrices using the Stata spmatrx create command combined with normalize (spectral) (Drukker et al., 2013).Fifth, we used the spatial weight matrices to conduct spatial panel-data diagnostics and specification tests.The results of the tests in this step, reported in Table 5, confirm statistically significant spatial dependence and fixed effect Spatial Durbin model (FE-SDM) and random effect Spatial Durbin model (RE-SDM).Finally, we proceeded to estimate FE-SDMs and RE-SDMs reported in this paper.
The results of Moran tests established significant spatial dependence for all the outcome variables used in this study.These results largely corroborate the tests for Pesaran tests for cross-sectional dependence in Table 5.However, the spatial Hausman test results favored the spatial fixed-effect Durbin models (FESDM) tourism arrivals, tourism receipts, and contribution to employment outcome variables, while the RE-SDM is favored for leisure spending and contribution to GDP.Further examination of model information criteria strongly favored the FE-SDM specification without L.-F. Lee and Yu (2010)'s transformation to the data.

Direct Causal Effects of Security Threats on T&T Demand and Economy
Following the battery of preliminary model specifications and diagnostic tests outlined above, we used the PCSE for our models.Table 6 shows the empirical results for the five tourism-related outcome variables used in this Akamavi et al.
study.Interestingly, the empirical results of the model show that the perception of crime and violent crime (i.e., robberies, assaults, kidnappings, and extortion) has an insignificant effect on tourism demand or economy variables.The reason behind such an outcome could be that the attractiveness of a destination subdues the effect of perceived crime.Altindag (2014, p. 8) notes that ''the attractiveness of a country may partly compensate for the probability of victimization.''Moreover, Alleyne and Boxill (2003) show that the impact of crime on tourist arrivals is mitigated by increased advertising and promotion of a destination.Alternatively, travelers can take precautions to reduce their perceived risk of crime without limiting their travel behavior (Barker et al., 2003).
Our findings also reveal that homicide rate and level of access to weapons in a country have direct significant negative influences on international tourist demand and economy indicators.Specifically, these findings show that destinations, which are notorious for high homicide rates and have easy access to weapons, face challenges to attract visitors, get lower tourism receipts, leisure spending, and benefit from lower tourism contribution to GDP.This implies that tourism demand and tourism economy tend to dwindle in countries, which experience high levels of homicide and preponderance of weapons used to commit homicide and violent crimes.The evidence from tourism-crime literature suggests that homicide/murder incidents have a greater impact on tourist arrivals than property crime (e.g., Alleyne & Boxill, 2003;Fourie et al., 2020;George, 2010;Pizam, 1999).However, the results show that incarceration has a significant positive effect on tourism arrivals, tourism receipt and leisure spending.We argue that incarceration has deterrent effects on perpetrators of crime rendering wider physical, psychological, and public safety benefits, which positively influence the intention to visit and spend time in destination and boost overall tourism demand and economy (Akerlof & Yellen, 1994;Wilhite & Allen, 2008).
Our analysis further shows that the total number of terrorist incidents has a negative influence on the growth rate of international tourist arrivals and the tourism receipts of destination countries.Moreover, highly insecure destinations affected by terrorism incidents fail to benefit from tourism.Tourism declines when perceived safety risks from terrorism incidents are high, or related information is communicated through the media (Kapus´cin´ski & Richards, 2016;B. Rittichainuwat & Chakraborty, 2009;Seabra et al., 2013).Terrorism incidents have adverse effects on expected cash flows and can depress travel industry stocks (Demiralay & Kilincarslan, 2019).However, we did not find any significant relationships between terrorism incidents and tourism leisure spending, T&T contribution to employment and GDP, which challenge the existing line of thought (e.g., Abadie & Gardeazabal, 2008;Blomberg et al., 2004) that a negative divergence of overall country's GDP results from terrorist shocks.
Meanwhile, one of the compelling results from our analysis shows that the overall impact of terrorism (terrorist-related violence attributing to physical and/or emotional damage to a country) has a significant positive influence on all T&T demand and economy variables.This finding can be explained from different perspectives of dark tourism, destination crisis management, and/or destination substitution.First, our findings are in line with the literature of dark tourism (Biran et al., 2014;N. Rittichainuwat, 2008;Strange & Kempa, 2003).Beyond mere leisure, recreation, and safety concerns, tourists may also be motivated to travel to terrorism-related destinations with motives ranging from a desire to honor victims to interest in seeing the grim magnitude of terrorism.
Second, Saha and Yap (2014) have noted that terrorism increases tourism demand up to a threshold, and then substantially lowers the value of tourist arrivals after that threshold.However, destination crisis management can rebuild a country's image and minimize terrorism's impact on tourism demand (Avraham, 2015;A. Liu & Pratt, 2017).Therefore, a country can attract tourists even though it has a history of instability and/or terrorist incidents.Third, the positive relationship between security issues and tourism demand and economy variables could be highly substitutional in nature for the international destination market.Previous studies (e.g., Aran˜a & Leo´n, 2008;Yechiam et al., 2005) have shown that if the degree of substitution among products is low, security issues have a less significant impact on tourist behavior.Therefore, travelers may opt for alternative destinations with similar characteristics, which they view as close substitutes (Neumayer, 2004;Yaya, 2009).This also contributes to confirming that tourists do not stop traveling when faced with insecurity; rather, they choose a safer destination (Bonham et al., 2006;Seabra et al., 2020).For instance, Afonso-Rodrı´guez and Santana-Gallego (2018) found that terrorist attacks in MENA countries have positive substitution effect on tourist arrivals in Spain.Moreover, Buigut et al. (2022) found that increased terrorism activity in Thailand increase tourist arrivals in Malaysia from four continents (Europe, North America, Oceania, and Asia) as well as overall.Hence, our result shows that despite terroristrelated violence attributing to physical and/or emotional damage to a country, overall impact of terrorism not necessarily will have a negative impact on tourism sector development, rather it can boost tourist arrivals, receipts, leisure spending, T&T contribution to employment, and GDP.
Our findings show that tourism demand and the tourism economy can be negatively affected by the political instability (e.g., transfer of power, coup d'e´tat, likeliness to opposition party coming to power and causing disruption, accountability and level of discretion, risk of international tensions affecting the economy).The results also demonstrate that intensity and duration of internal conflicts fought (e.g., economic sanctions, level of tense situation across country, group violence in sporadic incidents or systematic violence throughout the country, or civil war) have direct negative impact on tourist arrivals, tourism receipt, and leisure spending.However, intensity of internal conflicts does not show any impact on T&T contribution to GDP, while duration of internal conflicts fought does not have any impact on tourism economy variables.
Meanwhile, our interesting finding shows that the political terror (e.g., rule of law implication, political imprisonment, executions, disappearances, torture/brutality, etc.) and intensity of internal conflict have significant positive influence on the T&T contribution to employment.Despite the prevalent notion of political instability's negative impact on a country's economic output, some studies (e.g., Abu Murad & Alshyab, 2019;Campos et al., 2012;Jong-A-Pin, 2009) have shown that political instability and/or violence can have a positive or inconclusive impact on economic growth.One of the reasons for such a counter-intuitive relationship is that economic growth may be more responsive to a country's economic policies than its instability (Aizenman & Marion, 1993).A country can have an unstable internal political environment, but it would not necessarily alter overall rate of employment in a country while the country pursue consistent economic policies (Ali, 2001).Besides, we did not find any relationship between deaths from internal conflict and T&T service demand and economy variables.Overall the results show that tourism is sensitive to the state and nature of a destinations' political uncertainity.
Contrarily to de Albuquerque and McElroy (1999), Cruz-Mila´n et al. (2016), and Feickert et al. (2006), our results indicate that the number of security officers and police has a negative influence on tourist arrivals, T&T contribution to employment, and GDP.This result challenges the predominant notion that an increased number of uniformed security officers can increase tourists' perception of safety.Rather, our findings suggest that it might make tourists apprehensive, have a direct negative impact on tourist demand, and can indirectly affect the tourism economy.However, our results confirm that the reliability of police services can significantly enhance tourist demand (i.e., tourist arrivals and receipts) and have a positive impact on T&T contribution to GDP.Our results indicate that a high level of confidence in police may improve the sense of destination security, reassuring visitors and increasing tourist arrivals.This finding is in line with Cruz-Mila´n et al. (2016), who observed that the perceived effectiveness of security forces offers tourists a sense of protection, which eventually has a positive impact on the destination's economy.Our results show that tourists' perception of security is more reliant on the quality of police services than the quantity of security personnel present in the destination country.
While the defense spending and economic growth nexus is still elusive (Yakovlev, 2007), our study presents some stimulating findings.We show that an increase in military expenditure (i.e., national armed forces, paramilitary forces, customs protection officers, and border guards) can have negative impact on tourism receipts and contributions to employment but insignificant impact of tourism arrivals.We assume that increased spending on military forces can portray higher perceived risks of security threats, or the likelihood of armed conflict break out, which can negatively impact tourist receipts.Khalid et al. (2020) show that high levels of military expenditure tend to depress the tourism demand compared to countries that devote different levels of military spending.In addition, different studies (i.e., P. Dunne & Watson, 2000;Huang & Kao, 2005;Malizard, 2014;Tang et al., 2009;Yildirim & Sezgin, 2003) indicate that military expenditure can negatively affect a country's employment rate both in the short and long run.However, we also found that a country's weapon imports, exports, armed services personnel, and stock of nuclear and heavy weapons have significant positive impact on at least one of the tourist demand and economy indicators.The possession of nuclear and heavy weapons, higher active armed services personnel in military forces, and/or trading weapons might signal the strength of countries' national security and capability to protect itself and tourist destinations from external threats (Nassani et al., 2017).To our knowledge, this is the first study to offer empirical evidence regarding the effect of defense capability indicators on tourism demand and economic outcomes.
Finally, our results confirm that overall geopolitical power-plays have a negative influence on tourism demand and the tourism economy.We found that the number and duration of extraterritorial conflicts a country is involved in has a negative impact on its tourist arrivals, leisure spending, T&T contribution to employment, and GDP, which is similar to the defense economic literature (Demir et al., 2019;Farmaki, 2017;Tiwari et al., 2019).While the results show that deaths from fatality related to external conflict only has negative relationship with tourism leisure spending, we did not find significant relationships between neighboring countries relations (e.g., aggressiveness in politicians' speeches or in protectionist measures, serious tensions, economic and diplomatic restrictions, etc.) and tourist demand and economy variables.
We also reveal that a refugee population and the number of internally displaced people have negative influences on the tourism industry.Our results contradict prior studies' notion of ''immigration-led tourism'' (e.g., Balli et al., 2016;Etzo et al., 2014;Mehmood et al., 2016), but our finding is in line with the argument that refugee crises hurt tourism economy (e.g., Ivanov & Stavrinoudis, 2018;Pappas & Papatheodorou, 2017).Host communities' intimidating behavior toward refugees also affects the tourist experience and contributes to an unfriendly destination image, resulting in lower tourism demand (Ivanov & Stavrinoudis, 2018;Moufakkir, 2015).In addition, UN funding for peacekeeping has a negative impact on tourism demand and the tourism economy.Generally, UN peacekeepers tends to prevent conflicts, minimize violence, strengthen national security, and restore peace in a region/nation.Hence, one of the reasons for such negative relationships could be that increased funding on peacekeeping missions generate a sense of insecurity of related destinations and higher perceived risk to the T&T industry, leading to a negative impact on the industry performance.While such results are unique, it opens an intriguing avenue for future investigation into UN funding for peacekeeping and the tourism economy.

Long-run Spillover Effects
Table 7 and Appendix D present the marginal impacts (long-run spillover effects) and partial derivatives (coefficients) of our SDM estimators respectively.It should be noted that, unlike the coefficients of PCSE estimators in Table 6, which assume spatial independence, the partial derivative of the dependent variables with respect to covariates for the SDM estimators are not simply equivalent to the marginal effects of the covariates on the outcome variables (Golgher & Voss, 2016;LeSage & Pace, 2009).Therefore, the coefficients in Appendix D are mainly used for hypotheses testing in spatial econometrics practice because they are considered inaccurate for interpreting spatial effects in the models (Elhorst, 2012).Consequently, we will limit our discussion of spatial effect in the models to spillover effects from the impact measures presented in Table 7.However, the results in Appendix D show statistically significant spatial lag coefficients and/or ancillary variances for all the models.These results indicate significant overall spatial dimensions to the relationship between the outcome variables and the covariates in our models.
Table 7 reports estimate of the long-run spillover effects (direct/own-country effects), long-run indirect spillover effect (indirect/cross-country effects), and total effects of the outcome variables attributed to each of the covariates the models.The direct effects capture the average impact of the outcome variable attributable to the covariates in each country, while the indirect effects are the average effects across neighboring countries.Consequently, the long-run effects indicate spatial feedback effects of the covariates on the outcome variables due to spillover effects (LeSage & Pace, 2009).This implies that a change in the outcome variable for a particular country, connected with each covariate, will affect that country directly and possibly influence other (neighboring) countries indirectly (Elhorst, 2012).
Specifically, the results show that a change in the perception of criminality relates to large negative indirect (cross-country) and total spatial spillover effects on leisure tourism spending and tourism contribution to GDP but has insignificant spillover effects on tourism contribution to employment and no discernable effect on tourism arrivals and receipts.This implies that a change in the perception of criminality of a particular country tends to have negative cross-country impact on leisure tourism spending and tourism contribution to employment in neighboring countries, but negligible direct effect on other tourism economy variables.While previous studies have shown negative own-country effect of criminality on tourism arrivals (see, de Albuquerque & McElroy, 1999; Levantis & Gani, 2000; Michalko, 2004) these findings on negative cross-country spillover effects of perception of criminality on leisure tourism spending and tourism contribution to GDP, but non-discernable own-country spillover effects, is unique.While the results of PCSE indicate that perception of criminality has negligible causal impacts on tourism demand and tourism economy variables, our results from spatial panel data analysis show that the perception of crime has significant negative spillover effects on destinations in neighboring countries.To the best of our knowledge, this is the first attempt to show that perception of criminality tends to have negligible own-spillover effects on leisure tourism spending and tourism contribution to GDP but it leads to significantly negative cross-country spillover effects on neighboring countries.
Moreover, the results indicate significant negative own-country spillover effect of tourism receipt, leisure tourism spending, contributions of tourism to employment, and GDP to change in homicide rates, but no cross-country effects on tourism in neighboring countries except for employment.The results further show that higher incarceration rates generate strong positive ownand cross-country spillover effects on leisure tourism spending.This result confirms that incapacitation and deterrent benefits of incarceration directly benefits tourism spending of a country, and the benefits indirectly spillover to neighboring countries.
In addition, we show that incarceration rates significant positive own-country spillover effects on tourism receipts as well as positive cross-country spillover effects on tourism contribution to GDP of neighboring countries.Consistent with the security threats theory and literature, the results show that political instability exert negative effect on own-country and cross-country spillover effects on tourism receipts and contribution to GDP.This indicates that unstable political environment in a country directly hurts tourism outcomes of a country, and also generates negative externalities, which immiserate neighboring countries' tourism performance.Interestingly, the results further show that internal conflicts and deaths resulting from internal conflicts render a significant negative own-country spillover effect on tourism arrivals but insignificant cross-country and total spillover effects.However, the results show that deaths from internal conflicts exert significant negative influence on own-and cross-country (and total) spillover effects on tourism contribution to GDP.
The results further show that UN peacekeeping funding exert significant negative influence on own-spillover effects of tourism receipts and leisure tourism spending, as well as significant cross-country effects on contribution to GDP.Additionally, signs and magnitude of statistical of significance of the spillover effects are in line with results from the PCSE models.While results from the PCSE model in Table 6 did not show discernible effect of neighboring countries relations on tourism arrivals, the FE-SDM results attribute significant negative crosscountry and total spillover effects of tourism arrivals to neighboring countries relations.However, results from both PCSE and FE-SDM models attribute significant negative causal and own-country spillover influence tourism receipts, respectively, to military spending relationship.Furthermore, the results of PCSE and RE-SDM model results are, albeit with positive signs, for causal effects and own-country spillover effects of leisure tourism spending and contribution to GDP attributable to nuclear and heavy weapons.
While PCSE estimator results show that nuclear and heavy weapons render significant positive causal effect on tourism arrivals, possession of nuclear and heavy weapons generate significant negative cross-country and total spillover effects on tourism arrivals in neighboring countries.Moreover, weapons imports demonstrate significant positive long run spillover effects on tourism receipts and leisure spending, while significant negative cross-country spillover effects of tourism arrivals emanate from weapons exports.Finally, armed services personal of a country render significant positive own-and cross-country and total spillover effects on tourism arrivals and tourism receipts.

Conclusions and Policy Implications
This study elucidates, synthesizes, and integrates the existing body of knowledge to offer a clear conceptualization of security threats.Thus, it advances our understanding of the direct, indirect, and long-run spillover effects of global security threats on service industries.The empirical results clearly show that security issues strongly affect the performance of a country's tourism sector and have substantial impacts on the spatial inter-connectivity in tourism service satellite accounts (Frechtling, 2010;Smeral, 2006).
This study has answered calls from scholars (e.g., Corbet et al., 2019;Duan et al., 2022;Krajnˇa´k, 2021;C.-C. Lee & Chen, 2021;A. Liu & Pratt, 2017;Pizam & Mansfeld, 2006;Seabra et al., 2020) for in-depth insights into tourism security theory and related tourism economy issues.Given the new empirical evidence, a large proportion of tourist expenditure related to service sectors in a host country can positively or negatively contribute to economic growth.Our results indicate that security threats have significant negative causal and spillover effects on international tourist arrivals, and the contribution of tourism to employment and GDP.These findings are in line with the report of Shah and Aneez (2019), which shows that the Easter Sunday bombing in Sri Lanka resulted in 70% decline in tourist arrivals, a shortfall of $800 to 900 million in tourism revenue, 3% deficit to GDP, and thousands lost their jobs.A supportive report of World Trade Organization (WTO, 2002) also substantiates that the Bali bombs cost US$2 billion in international and domestic tourism earnings and rendered 2.7 million people jobless.Our findings reveal the need for authorities and stakeholders to take decisive measures against security threats and manage risk perceptions of destinations to protect the tourism service sector.Besides, international cooperation is extremely important in reducing the risk of terrorist attacks in destination countries.While terrorist incidents are relatively unpredictable, appropriate intelligence sharing can provide information that can be leveraged to alert destinations and identify weaknesses in security systems.
Moreover, to minimize security crises in a particular destination, transformative crisis management plans need to be developed by all tourism stakeholders (e.g., governments, agents, media, and the local tourism industry).In the case of terrorist incidents, a rapid crisis management strategy should be implemented to accelerate the recovery process.Reinstating a secure image of destinations involves a multi-step, holistic approach that synthesizes pragmatic measures using marketing strategies (Avraham, 2015).Such an approach eases tourists' psychological barriers in selecting travel destinations and may help destinations contain spillover effects.Tourism managers in conflict-prone destinations should be prepared to modify their marketing strategies quickly (e.g., reduce hotel and associated supplementary service prices, enhance safety measures, booking alterations, flight prices, etc.) to repair the destinations' image after a crisis.
Additionally, assessing police service reliability can be a valuable approach for managing destination image.Ensuring the reliability of police services will rebuild confidence and reassure prospective tourists, which will subsequently boost tourism demand and the tourism economy.While increasing the number of security and police forces can generate anxiety or fear within the tourist community, our research sheds light on enhancing police reliability to ensure sustainable tourism sector development.Preventing crime through arrests and confinement and fewer homicides and internal conflicts can project an image of reliable police forces, which can help build the required confidence to reassure prospective travelers.Consequently, demonstrating reliability safety can help develop an image of a resilient destination, thus attracting a growing number of visitors to boost destinations' service economies.Hence, policymakers in destination countries should raise awareness on the reliability of police services to reassure travelers.
We also propose extending the role of destination management organizations to tourism security-related strategic development planning.Along with such implementation, we also suggest that increasing social capital (e.g., trust and collective community relationships) should be prioritized to enhance countries' resilience and recovery from security threats.Such social capital can project a sense of communal assurance for tourists, which can also lead to improved economic activities.
In addition, our study presents some thoughtprovoking findings worthy of further investigation.Although global terrorism has dual effects on the tourism industry's receipts, employment, travel services, leisure expenditure, and the service sector's contribution to GDP, scholars have overlooked the spillover effects of dark tourism.Additionally, there is a lack of panel data on global dark tourism.Therefore, scholars should undertake empirical investigations on dark tourism's impact on the T&T service industry.Moreover, other potential factors, such as corruption and socio-cultural issues, may also have significant negative effects on the five dependent variables.Economic policy's effect on T&T service sector investments also warrants further attention.In addition, the proposed framework needs refinement through more empirical studies to confirm this study's results, which can provide further evidence to corroborate tourism security theory and health outbreaks.
This study is also subject to certain limitations, which unfolds avenues for future research.One of the main limitations of this study is the unavailability of industry-level of mediating variables (e.g., hospitality services, transport services, business services, the retailing industry, supply chain, etc.) to explore the indirect and long-run effect of security threat variables.In addition, seasonality, visitors' socio-economic and cultural differences, or attractiveness of destination can affect tourist's travel patterns and preferences, which future research should consider deepening the understanding of different T&T demand patterns and economic contribution.Moreover, travel purposes (i.e., business/professional, visiting friends/health/religion, or leisure/recreation/holidays) can influence tourists response to security issues, which should also be considered by the future research.Furthermore, we augment Pizam and Mansfeld (2006)'s study with additional typologies such as natural disasters, health-related threats, industrial hazards, and cybercrime as crucial security threats to the T&T industry (see Table 8).Hence, we suggest to investigate the impact of these security threats on T&T performance.For example, there is insufficient data on disease outbreaks such as SARS, H1N1, MERS, Ebola, and COVID-19 to empirically estimate their spillover effects on the tourism economy and tourism demand.It is evident that the upsurge in the COVID-19 pandemic has substantially ruined the T&T service industry.This is in line Financial Times Reporters (2020) who conveyed that COVID-19 is the worst crisis since the Second World War.Moreover, UNWTO (2020) reports a 74% reduction in international tourist arrivals in 2020.Furthermore, UNWTO (2021), the COVID-19 pandemic has had a massive impact on the global economy and livelihoods, affecting 100 million direct tourism jobs and resulting in an estimated economic loss of US$1.3 trillion.Indeed, the dramatic variation and severity of COVID-19 have noticeably increased the perceived risks and threats associated with the T&T industry (C.-C. Lee & Chen, 2021;Zheng et al., 2021) The findings of this study indicate that travelers and tourists react to insecurity, which leads to suspension, cancelation, or substitution of their travel plans.This inherent reaction to insecurity is not dissimilar to the traveler's reaction to the COVID-19 pandemic.For instance, international tourism may give way to domestic tourism or staycations.Hence, we can argue that the pandemic has transformed the perception of health risks associated with tourism (Qiu et al., 2020).However, we lack COVID-19 panel data to include in this empirical modeling investigation/exercise of security threats' spillover effect on service industries.Zhang et al. (2021) made the observation that COVID-19 ''data limitations and the unprecedented context of this pandemic, traditional statistical forecasts could not incorporate the effects of the related factors.''Due to limitations of COVID-19 panel data, we put forward the use of the revised framework for further studies; this study on security threats could be extended by examining the spillover effects of the COVID-19 pandemic and other disease outbreaks (e.g., MERS, Ebola) on the tourism economy and tourism demand.Based on the above reasoning, our conceptual framework of security threats' spillover effects can be useful in determining wider empirical impact pathways of COVID-19 on global service industries.As modern-day policymakers and managers often draw on ideas in scholarly works, this study offers conceptual understanding and empirical evidence of the spillover effects of global security threats on the T&T service industry.
Appendix A. Review of Some Key Studies.

Figure 1 .
Figure 1.Impact of security threats on service industries economy: spillover effect model.*Other risk factors have not been included in the analysis due to lack of data.**T&T services have excluded from the analysis due to lack data.

Figure 2 .
Figure 2. Tourist arrivals, terrorist acts, and economic cost of terrorism in between 2010 and 2019.Source: Data collected from The Global Terrorism Database (2020), Institute for Economics & Peace (2020), The World Bank Databank (2020).Bottom and left axis: explaining time and amount of terrorist incidents, injuries and deaths in thousands; top and right axis: explaining years, amount of tourist arrivals, and economic cost of terrorism in billions (log scale).

Table 1 .
Precis of Some Terrorist Attacks and Effects.
21-Apr-19 Sri Lanka Easter Bombings: A series of explosions were reported at three churches and three hotels in several cities in Sri Lanka targeting Christians and foreigners.

Table 2 .
Explanation of Variables and Data Sources.
Intentional acts of violence or threat of violence by a non-state actor, which must be aimed at attaining a political, economic, religious and/or social goal, or intend to coerce, intimidate, or convey some other message to a larger audience(s) than the immediate victims,

Table 4 .
Greene's (2018)0)d Model Specification Tests for the Data.Note.Pesaran CD 1 and CD 2 Pesaran denote test statistics for the unrestricted and restricted models, respectively.Results show that we cannot reject (weak) cross-sectional dependence/contemporaneous correlation for all models.p-Valuesarereported in parentheses.Tests Wooldridge test refers toWooldridge (2010)for serial correlation/autocorrelation in panel data.Heteroscedasticity test refers toGreene's (2018)modified Wald test for groupwise heteroscedasticity.

Table 5 .
Spatial Panel Specification Tests.

Table 6 .
(continued)Note.Degrees of freedom for Wald Chi 2 test (Wald Chi 2 (df)) for International Tourism-Arrival Wald = 20, International tourism receipt = 12, Leisure Tourism Spending = 18, Contribution to Employment = 16, Contribution to GDP = 21.a This study recognizes that the tourism economy outcome variables (i.e., T&T services contribution to GDP, tourism employment, leisure expenditure) might not be independent for a given country.Therefore, the economy outcome variables in our models are not projected onto one another to avoid internal validity problems resulting from double counting.