key: cord-0690682-1g7hh3wp authors: Shao, Yuhong; Huang, Songshan (Sam); Wang, Yingying; Li, Zhiyong; Luo, Mingzhi title: Evolution of international tourist flows from 1995 to 2018: A network analysis perspective date: 2020-10-22 journal: Tour Manag Perspect DOI: 10.1016/j.tmp.2020.100752 sha: 1107aed1c29257e59e1781531c1e07731ff7e787 doc_id: 690682 cord_uid: 1g7hh3wp Tourist arrivals and tourism revenues have been extensively studied to evaluate international tourist flows, whereas the structure and evolution of these flows have received less attention. Based on international tourist arrival data from 221 countries/regions during the period 1995–2018, this study applies network analysis to explore the structure and evolution of international tourist flows, and the roles and functions of countries/regions in the international tourist flow network. The results of this study reveal that the network density of international tourist flows is increasing. Countries/regions in Europe, East Asia and North America generally occupy a significantly important position within the international tourist flow network, especially Germany and China. Those geographically close countries/regions demonstrate the same or similar roles and positions in international tourism. This study has significant implications for tourist destination management and marketing. International tourism has become a popular global leisure activity worldwide (Keum, 2010) . According to a report released by the World Tourism Organization (UNWTO), the magnitude of international tourist arrivals rose to 1.4 billion in 2018, ahead of the forecast by UNWTO (UNWTO, 2019) . Likewise, the revenues of international tourism increased from US $485.178 billion in 1995 to US $1.649 trillion in 2018 (UNWTO, 2020) . In this regard, international tourist flows have attracted the attention of both the global tourism industry and academic research (Zhang, Li, & Wu, 2017) . Previous research has mainly evaluated international tourist flows from the perspectives of tourist arrivals or tourism revenues (e.g., Balli, Balli, & Louis, 2016; Hall, 2010; Huang, Han, Gong, & Liu, 2019; Yang, Liu, & Li, 2018; Zhang et al., 2017) . Essentially, international tourism is a place-oriented activity with tourist flows across country borders (Deng & Hu, 2018; Keum, 2010) . However, few studies have focused on the structures of international tourist flows worldwide, especially the dynamic changes of these flows. According to Yang et al. (2018) , today's world order faces unprecedented backlash, as does the global tourism industry. Understanding the structure and evolution of the international tourist flows is conducive to the implications for the development of infrastructure, product, destination and others, as well as the management of tourism's impacts on society, environment and culture (Lew & McKercher, 2006) , which can be helpful for policymakers and tourism firms to improve market competitiveness and destination management. Network analysis is an approach with a set of methods and tools to map and measure the patterns, flow and strength of relationships between actors (Casanueva, Gallego, & García-Sánchez, 2014) , and has been applied in the study of tourist flows within a region or between selected countries (Zeng, 2018) . Following previous studies, from a global perspective, this study employs this approach to investigate the roles and functions of countries/regions acting as tourism origins or destinations during the period 1995-2018, further revealing the evolution of the international tourist flow structures. The rest of this paper is structured as follows: The next section provides a brief review on tourist flow. The following section reports the data sources and methodology. Then, the results, discussion and conclusions are presented. The last section provides implications as well as future research and limitations of this paper. destination regions. In this regard, tourist flow refers to the movement of tourists from an origin place, through transit regions, to a destination and the stay of tourists in these regions (Oppermann, 1995; Zeng, 2018) . According to Bowden (2003) , tourist movement encompasses three basic elements: intensity, direction and pattern. Generally, the intensity is analyzed under the fields of "tourist demand" or "tourism forecasting" since it is related to the volume and frequency of tourist flows. Direction and pattern, which reflect the static and the dynamic elements of tourist flows among regions, respectively, are usually discussed under the term "tourist flow" (Bowden, 2003) . The dynamic element mainly centres on the flows between origin and destination regions. In contrast, the static element is composed of several factors, such as tourism destinations, overnight stays, accommodation types, and the gateways between origin and destination regions (Oppermann, 1992) . A large amount of research has been conducted on tourist flow at different geographic scales (Amelung, Nicholls, & Viner, 2007) . The geographic scale reflects the hierarchy and functional arrangements of spatial issues, which is important for exploring tourist flow (Bowden, 2003) . According to Xia, Zeephongsekul, and Arrowsmith (2009) , the geographic scale of tourist flow can be attributed to the macro-and micro-levels based on distance. The macro-level refers to a relatively large distance of hundreds of kilometres (Xia et al., 2009) , which is often regarded as inter-destination movement pattern (Lau & McKercher, 2006) . In contrast, the micro-level is considered to be a relatively short distance, such as from an attraction to another attraction, which refers to intra-destination movement pattern (Lau & McKercher, 2006) . As far as the geographical scale of tourist flow is concerned, this study estimates the tourism flows between 221 countries/regions around the world from the macro-scale or inter-destination movement perspective. The pattern of tourist flow involves various items of information, which is conducive to designing tourist packages, providing attractive combinations of attractions, proposing tourism guidance policies and marketing management (Lew & McKercher, 2006; Xia et al., 2009) . A large amount of research has attempted to map tourist flow through various methods (Leung et al., 2011) . The traditional techniques for tracking tourist flows mainly depend on observations, interviews or questionnaires (Zeng, 2018) . Researchers are asked to track the tourists' movements to develop a map of tourists' distribution within a given destination (Dumont, Roovers, & Gulinck, 2005) . In addition, tourists are required to retrace their movements through self-administered questionnaires (Xia et al., 2009 ). However, limited by time and cost, these techniques usually obtain a limited amount of data and lack the needed accuracy. With the development of technology, new tracking techniques are applied to record the information of tourist flows, such as the Global Positioning System (GPS) and land-based tracking systems, which have proven to be effective tools for estimating the spatial flows of tourists over time (Shoval & Isaacson, 2007) . However, the above two kinds of techniques, namely the traditional techniques and new tracking techniques, are applied to tourist flows at the micro-or meso-level. Regarding the macro-level, the panel data published by organizations are regarded as important sources for researching tourist flows (Liu, Li, & Parkpian, 2018; Lozano & Gutiérrez, 2018; Su & Lin, 2014) . The large amounts of data available, the use of the same statistics definitions, easy accessibility and long-term data availability, are considered as the main advantages of panel data, which contributes to the wide use of panel data in exploring international tourist flows. For example, Li, Meng, and Uysal (2008) explored the tourist flows among the Asia-Pacific countries for the years of 1995 and 2004. Based on panel data from 1990 to 2002, Keum (2010) examined the patterns of international tourist flows between South Korea and its 28 major trading partner countries. Additionally, scholars have applied several methods, related to data mining methods and statistical methods, to identify the spatio-temporal patterns of tourist flows, including but not limited to the field of international tourist flows. These methods involve the Clustering Method (Asakura & Iryo, 2007) , Gross Travel Propensity Index (GTP) (Li et al., 2008) , Geographic Information System (GIS) Analysis (Connell & Page, 2008) , and Markov Chains (Xia et al., 2009) . For example, Asakura and Iryo (2007) applied the Clustering Method to reveal the topological characteristics of the tourist movement in Kobe, which contributes to finding the hidden behaviour of tourists. Connell and Page (2008) employed GIS analysis to map car-based tourist flows in Loch Lomond and Trossachs National Park. Xia et al. (2009) employed Markov chains to estimate the outcomes and trends of events related to the patterns of tourist flows across Phillip Island, Australia. Recently, scholars have introduced and applied network analysis to reveal a relatively comprehensive picture of international tourist flows (e.g., Leung et al., 2011; Zeng, 2018) . Compared with other statistical methods (e.g., Clustering Method), the tourist flow network based on network analysis can be visualized and is easy to understand (Leung et al., 2011) . Accordingly, network analysis reveals the roles, functions and cohesiveness groups of destinations, which provides more implications for destination managers (Kang, Lee, Kim, & Park, 2018; Scott, Cooper, & Baggio, 2008) . Mainly based on mathematics and graph theory, network analysis is an approach that uses a set of methods and tools to map and measure the patterns, flow and strength of relationships between actors (Casanueva et al., 2014) , which makes network analysis different from other analysis methods (Scott et al., 2008) . The relationships can be of various types, including but not limited to goods, services, information, and social support; the actors establishing relationships with each other can be individuals, organizations and other linked information/ knowledge entities (Haythornthwaite, 1996) . Although the network analysis technique was mainly developed in economic sociology, researchers have applied mathematical models to estimate the structures of various relationships, indicating that network analysis was not limited to the social field (Scott, 1991) . Moreover, researchers like Granovetter (1973) , Burt (1992) , Watts (1999) and Lin (2001) , have furthered the research on network analysis, making it widely used in various fields. Currently, scholars have employed network analysis to estimate the flow paths and patterns of international tourists within a destination. In these studies, a destination is regarded as an actor within the tourist flow network, while tourists from one destination to another is viewed as the relationship between destinations (Zeng, 2018) . For example, Leung et al. (2011) utilized network analysis to analyze the pattern of overseas tourist flows in the most visited tourist attractions throughout the Olympics in Beijing. Likewise, Zeng (2018) estimated the structure and characteristics of Chinese tourist flows in Japan through itineraries from travel services and trip diaries. Lozano and Gutiérrez (2018) explored the structure and interactions between source and destination markets in the global tourism network in 2013. Wu, Wang, and Pan (2019) combined numerical simulation and network analysis to construct an agent-based network of inbound tourism in China and numerically investigated the responses of the inbound tourist flows in some scenarios of practical significance. However, these above-mentioned studies, mainly centring on particular regions or selected countries or specific year, hardly contribute to the understanding of the competitiveness of destination countries/ regions worldwide. In this regard, the purpose of this study is to estimate the structures and evolution of international tourist flows during the period 1995 to 2018 from a global perspective, which will be further conducive to proposing general tourism development planning for most countries/regions worldwide. Y. Shao, et al. Tourism Management Perspectives 36 (2020) 100752 3. Data and methodology The annual data for bilateral tourist flows were collected from the UNWTO, which covers 221 countries/regions from 1995 to 2018. This data set is compiled by destination countries/regions based on the number of inbound tourists. The data for 1995 and 2018 are the earliest and latest data sets that can be obtained, respectively. Although this data set is widely used in the field of international tourism (Balli et al., 2016; Yang et al., 2018; Zhang et al., 2017) , three issues in the data set need to be emphasized. First, different destination countries/regions adopt different definitions in statistics. Among 8 statistics definitions currently adopted by destination countries/regions, the most commonly used statistics are arrivals of non-resident tourists at national borders (by country of residence), arrivals of non-resident tourists at national borders (by nationality), arrivals of non-resident visitors at national borders (by country of residence), and arrivals of non-resident visitors at national borders (by nationality). Following the study of Yang et al. (2018) , when cleaning the data, this study gave preference to the above definitions associated with border; 4 other definitions related to accommodation were considered when the above four border-based statistics were missing. Second, different countries/regions had different tourism statistics systems, and several countries/regions reported data for a subset of origin countries/regions. Third, this study unified the names of countries/regions to avoid the ambiguity caused by different statistical systems, such as unifying "State of Palestine" into "Palestine" and "Congo, Democratic Republic of the" into "Democratic Republic of the Congo". Network analysis aims to analyze the structure of relationships (displayed by links) between given entities (displayed by actors) in social or economic phenomena (Haythornthwaite, 1996) . It employs a set of techniques to explore the characteristics of a whole network, as well as the roles and positions of these entities within the network (Shih, 2006) . In this study, we applied network analysis to explore the structure of international tourist flows, where the countries/regions are treated as "actors", the tourist routes between origin and destination countries/regions are regarded as "links". Fig. 1 shows a simple case with five countries/regions (labelled A, B, C, D and E). Fig. 1A is a network graph, showing the relationship of international tourism among these five countries/regions. For example, tourists from country/region A visit C, D and E, and do not travel to B; additionally, country/region A only receives tourists from D. According to the graph, an asymmetric matrix can be built (see Fig. 1B ), in which a row represents the destination countries/regions and a column stands for the origin countries/regions. This type of matrix above merely describes the presence or absence of the given type of relationship. However, each route between two countries/regions carries a specific number of tourists, which is considered as "weightings", yielding a valued matrix. To be specific, the (i, j)th cell (row i, column j) carries a number that represents the number of outbound tourists from country/region i to country/region j. On this basis, the matrix of 221 countries/regions used in this study is constructed. The rest of this section introduces the indicators of network analysis which are appropriate for this study. To estimate the structure of international tourist flows, this study applied three indicators of network analysis, namely density, degree centrality and blockmodel. Among them, density is the main indicator for the structure of a whole network (Casanueva et al., 2014) , while degree centrality and blockmodel are important indicators to examine the structure of actors within a network (Borgatti, Everett, & Johnson, 2018) . To be specific, density is a measure of cohesion, which means the connectedness of a network. This indicator can be interpreted as the probability of a link between each pair of randomly selected actors (Borgatti et al., 2018) . Degree centrality, which is measured by the number and value of links that an actor has, is suitable for analyzing the structural roles and positions of each actor within this network. However, although degree centrality is the primary indicator for the structure of an actor, it cannot contribute to understanding the importance of links between actors (Asero, Gozzo, & Tomaselli, 2015) . Thus, we enriched the analysis by employing the blockmodel. The term "blockmodel" was first proposed by White, Boorman, and Breiger (1976) to explain the social structure in terms of interconnections among actors within a social network. Two actors that occupy the same structural roles or positions in a network are said to be structurally equivalent (Asero et al., 2015) , and are grouped into the same block. Thereby, a network is divided into different "blocks". A "block" is a subnetwork embodied in the overall network, and the actors within a "block" are structurally indistinguishable because they have the same external relationships. This implies that the actor with the same role or position in different links can be interchangeable with one another. According to Borgatti et al. (2018) , the analysis for structural equivalence provides a high-level description of the links within a network. Moreover, structurally equivalent actors share other similarities as well; they show a certain amount of homogeneity (Borgatti et al., 2018) . Considering the competition in tourism market and alternative tourist flow routes, it is necessary to analyze structural equivalence when studying tourist flows (Asero et al., 2015) . The formulas of these indicators have been explained by Knoke and Kuklinski (1982) , Scott (1991) , Carrington, Scott, and Wasserman (2005) , Knoke and Yang (2008) , Luo (2012) , Borgatti et al. (2018) , among others. In this regard, we explained the above indicators in the context of international tourist flow (Table 1 ). The indicators of network analysis used in this study were calculated by UCINET 6.6. Network density indicates the extent to which countries/regions interact with other countries/regions in terms of international tourism. Here, ρ is the network density, ranging from 0 to 1 (0 denotes no connection, and 1 denotes that all countries/regions are connected); m is the number of actual connections among countries/regions; n indicates the number of countries/regions; and n (n − 1) refers to the potential maximum connections between countries/regions contained in this network. Degree centrality Measured by the number of links that an actor has. In a valued network, it takes the link value into account. It indicates the structural importance of a country/region in the international tourist flow network. A country/region with a high degree centrality value occupies a position that is important in the network. Degree centrality in a directed network includes outward (out) and inward (in) degree. Out-degree centrality refers to the number of links that an origin country/region sends for destination countries/ regions, while in-degree centrality is the number of links that the destination country/region receives from origin countries/regions. C n a n out degree C n a n in degree n is the number of countries/regions in the network and α ij (α ji ) indicates the edges from country/ region i (j) come in or out of country/region j (i), respectively. Blockmodel A network is divided into several blocks, and actors within a block have the same external links and network positions. It is suitable to explore tourist routes because destinations with the same role or position in different routes can be substitutable with one another. The Convergence of Iterated Correlations (CONCOR) algorithm is a method of hierarchical clustering for obtaining blockmodel structure in network analysis. It starts with a correlation matrix in which each pair of countries/regions is linked with the other ones. Then, extract each row of this matrix and make it linked with other rows to obtain similarity profiles of countries/ regions. On this basis, two blocks are detected through these similarity profiles. Furthermore, this procedure is repeated within each block until a satisfactory subsets partition is obtained. Besides, it provides the density of each block. Additionally, the trend of density was essentially consistent with that of international tourist arrivals as listed in Fig. 2 . After 2009, the number of international tourists and the network density of international tourist flows continued to increase significantly. Moreover, Gephi software was applied to visualize the distribution of international tourist flows. Fig. 3 and Fig. 4 show the international tourist flow networks in 1995 and 2018, respectively. The larger node size indicates that a particular country/region generate more international tourists, while the thicker line between countries/regions represents more outbound tourists. As shown in Fig. 3 , in 1995, the largest cluster for the international tourist flow network was Europe, followed by North America and East Asia, with several countries as the centre, including the United States, Germany, Canada, France, and the United Kingdom. A large number of countries/regions were at the edge of the international tourist network and had only a few travel links with the remaining countries/regions within this network in 1995. While in 2018, we can find that after 24 years, almost all countries/regions have strengthened travel links with other countries/regions, which suggests that the interconnectedness of the international tourist flow network has been largely improved. Although Europe, North America and East Asia were still the most important clusters, the number of countries/ regions covered in these clusters had increased significantly. Out-degree centrality was used to describe the role and function of a country/region in outbound tourism. The results of the out-degree centrality are summarized in Fig. 5 . During the study period, out-degree centrality values of most countries/regions were on the rise, with occasional fluctuations in 2001, 2003 or 2008, and maintained a relatively stable ranking. Thus, Table 2 only reports the values of outdegree centrality and rankings of countries/regions in 2018 because of space limitations. Given that countries/regions in the top 30 account for about 80% of the sum of out-degree centrality in the 221 countries/regions, we considered these countries/regions occupied a relatively important role and function in outbound tourism. Over the 24 years, Germany, the United Kingdom, France, Switzerland, Czech Republic, Italy, Belgium, Austria, Spain, the Netherlands, Ukraine, the Russian Federation, the United States, Canada, Mexico, China, Hong Kong SAR, Macao SAR, Taiwan, Japan and Australia were always in the top 30, playing a dominant role in generating international tourists to many destination countries/regions. In particular, among these 21 countries/regions, 12 of them, including Germany and the United Kingdom, belong to Europe; the United States, Canada and Mexico are countries in North America; China, Japan, Hong Kong SAR, Macao SAR, Taiwan are located in East Asia; and Australia belongs to Oceania. Specifically, during the period 1995-2018, Germany always ranked 1st, with an average out-degree centrality value of 107.444, indicating that Germany plays a leading role in global outbound tourism, taking into account the number of destination countries/regions and outbound tourists. In particular, before 2000, the out-degree centrality values of Germany were far higher than those of the United States, which ranked 2nd. The out-degree centrality values of the United States showed a relatively stable upward trend from 1995 to 2013 and a sharp increase after 2013. During 2002 and 2015, the out-degree centrality values of Hong Kong SAR surpassed those of the United States, ranking 2nd in the international tourist flow network, and maintained a slight upward trend after 2007. France, Canada and Italy maintained a relatively stable ability to interact with other destination countries/regions and showed a steady growth trend throughout this study period. Regarding the Russian Federation, its out-degree centrality value showed an increasing trend until 2013, after which it began to decrease sharply. This may be related to the sharp decline of oil price, the Ukraine crisis in 2014 and the sanctions imposed by Western countries, which make the economy stagnant in the Russian Federation and further have a negative impact on tourism (Dreger, Kholodilin, Ulbricht, & Fidrmuc, 2016) . It is worth noting that the out-degree centrality value of China continued to significantly increase from 1995 (4.836 of outdegree centrality) to 2018 (113.394 of out-degree centrality), with only a slight decrease in 2008 due to the financial crisis; moreover, the ranking of China showed an upward trend from 21st in 1995 to 3rd in 2018. Besides, after the financial crisis in 2008, the growth of out-degree centrality of most countries/regions slowed, while China showed a Y. Shao, et al. Tourism Management Perspectives 36 (2020) 100752 trend of unprecedented growth to generate outbound tourists to increasing destination countries/regions. In-degree centrality was used to describe the role and function of a country/region in the inbound tourism network. As shown in Fig. 6 (Table 2) , maintained or increased most connections with other origin countries/regions and had the strongest ability to attract international tourists compared with the rest of the world. Concerning other leading destination countries/regions, the in-degree centrality values and rankings of Spain, the United States, Italy, France and Poland have remained close to each other since 2001, and ahead of other countries/ regions, including the United Kingdom, which has almost always ranked around 7th out of 221 countries/regions since 1997. As the most important inbound tourism destination in the last century, Poland's indegree centrality values fell sharply between 1999 (89.070 of in-degree centrality) and 2002 (50.691 of in-degree centrality), and between 2007 (66.085 of in-degree centrality) and 2009 (53.597 of in-degree centrality). The United States showed a similar trend to Poland in terms of in-degree centrality, but more smoothly during the study period. It is worth noting that Southeast Asian countries, such as Thailand, Malaysia, Singapore, Vietnam and Indonesia, were at the forefront of the inbound tourist flow network, consistent with the national positioning created by these countries. Moreover, countries/regions with regional conflicts or infectious diseases had low in-degree centrality values, such as Sudan, Chad, Palestine and the Central African Republic. Tourists throughout the world rarely visit these countries/regions considering their safety. Furthermore, the rankings of out-degree centrality of countries/regions were relatively consistent with those of their in-degree centrality within the international tourist flow network. For example, countries/ regions with high out-degree centrality tended to have high in-degree centrality in the international tourist flow network, including but not limited to Germany, the United States, China, the United Kingdom, France, Japan, Canada, Hong Kong SAR, Italy, Spain, Macao SAR and the Russian Federation, which are concentrated in Europe, East Asia and North America. Moreover, the out-degree centrality in countries/ regions with small in-degree centrality also tended to be small, such as Sierra Leone, Montserrat, and Seychelles ( Table 2) . Most of the countries/regions mentioned above are located in Africa or are islands with small populations and territories. However, the majority of countries in Southeast Asia, including Thailand, Indonesia and Malaysia, had higher rankings in in-degree centrality than that of out-degree centrality during the years of the study, revealing that inbound tourism was a significant pillar of the growth strategies of these countries. Besides, except for countries/regions (e.g., French Guiana, Pakistan, Iraq) that did not have statistics on inbound tourist arrivals, the values of outdegree centrality in several countries/regions (e.g., Republic of Moldova, Belarus, Belgium) were higher than those of in-degree centrality, indicating that these countries/regions have a stronger ability to generate international tourists to many countries/regions than to attract tourists. The above subsections mainly reveal the role and function of a country/region in the international tourist flow network and do not allow for the importance of travel links between countries/regions to be Table 2 The centrality analysis of countries/regions in the international tourist flows in 2018. Y. Shao, et al. Tourism Management Perspectives 36 (2020) 100752 understood. Therefore, following the study of Asero et al. (2015) , this study used the CONCOR algorithm to estimate the structure corresponding to the country/region's role and position in the network. Countries/regions with the same tourist flow routes can be clustered into one block, indicating that countries/regions in the same block are structurally equivalent and can be substituted for each other (Luo, 2012) . Also, CONCOR algorithm provides the density of each of the blocks (Borgatti et al., 2018) , and allows for identifying the main links from the values of the density matrix (Asero et al., 2015) . Since the roles and positions of countries/regions in the international tourist flow network are relatively stable, countries/regions in each block have not changed significantly over the years. Therefore, we took the results of the 2018 CONCOR algorithm as an example (Table 3 and Table 4 ). Block 1 centred on the Russian Federation and Belarus, and mainly included countries/regions distributed around the Caspian Sea (e.g., Turkmenistan, Kazakhstan, Islamic Republic of Iran). The majority of countries/regions within this block had a higher value of out-degree centrality than that of in-degree centrality, especially Belarus and Republic of Moldova. Moreover, as shown in Table 4 , countries/regions in Block 1 were closely linked to each other (Density = 0.179) and interacted with countries/regions in other blocks except for Block 3. This suggests that most countries/regions within Block 1 have a strong ability to generate international tourists to both countries/regions in other blocks and its own block. The majority of countries/regions in Block 2 are concentrated in North and Central Africa (e.g., Algeria, Sudan, Djibouti) and Arabian Peninsula (e.g., Saudi Arabia, Kuwait, Yemen); that is, around the Red Sea. Except for Saudi Arabia, the degree centrality of countries/regions within this block generally ranked in the middle or lower among 221 countries/regions, indicating that these countries/regions have a medium performance in international tourism. As for Block 3, except for French Guiana and Guadeloupe, which are French overseas regions, other 16 countries, including Burundi and Mozambique, mainly belong to Central or Southern Africa. Most countries/regions in this block had low values of degree centrality and barely had travel links with other countries/regions, suggesting that these countries/regions are at the edge of international tourism flow network. According to Block 4, except for Suriname, 23 countries/regions in this block are around the Central or Western Pacific (e.g., China, Malaysia, Indonesia), and the remaining countries/regions are located around the Gulf of Guinea (e.g., Côte d'Ivoire, Liberia). Asian countries/regions in Block 4 generally ranked higher than those in Africa and Oceania in terms of degree centrality. Besides, the interactions between countries/regions within this block (Density = 0.252) were much higher than those with countries/regions in other blocks. Countries in Block 5 are located in Europe, such as Greece, the United Kingdom, Ukraine, Belgium, Germany and France. Generally, European countries have small territories, developed economies and high affluence rankings in the world. These countries not only generate international tourists but also have the ability to attract tourists from other countries/regions. Besides, its block density was the highest, reaching 0.328, revealing the close connections between European countries. In Block 6, countries/regions are mainly distributed along the Mediterranean Sea (e.g., Tunisia, Morocco, Egypt, Malta, Spain) and the West or North Indian Ocean (e.g., Seychelles, Madagascar, Maldives, Sri Lanka), while a few countries/regions are concentrated in the Gulf of Guinea (e.g., Togo, Congo). The majority of countries/regions in Block 6 had higher rankings in in-degree centrality than that of out-degree centrality, indicating that these countries/regions have a stronger ability to attract international tourists. Block 7 was dominated by the United States, Canada, and Mexico. Except for the above three countries, other 35 American countries/regions within this block possessed medium or small degree centrality, such as Cayman Islands, Aruba, Colombia. Other countries/regions in Block 7 are located in Asia (e.g., Qatar, Armenia, Israel, Philippines, Nepal), Africa (e.g., United Republic of Tanzania, Ethiopia), Oceania (e.g., Kiribati, French Polynesia) and Europe (i.e., Iceland). Countries/ regions within this block has established travel links with countries/ regions in the other seven blocks, especially with European countries in Block 5. Block 8 focused on countries/regions located in South America (e.g., Argentina, Brazil, Venezuela, Bolivia), and islands located in Oceania or Western Pacific, including but not limited to Australia, New Zealand, Japan, Cook Islands, Fiji, and Palau. As shown in Table 4 , countries/regions within Block 8 mainly interacted with other countries/regions in its block as well as Block 7 and Block 5. From the above analysis, countries/regions within the same block are mostly located on the same continent or are geographically close to each other. Geographic contiguity, language similarity or colonial links between two countries/regions increase the bilateral flow of tourists (Yang et al., 2018) . It is worth noting that countries/regions located in a block have the same external tourist flows, and the substitution effect refers to the structurally equivalent relationship between countries/ Shao, et al. Tourism Management Perspectives 36 (2020) 100752 regions. In the real situation, every country/region has uniqueness attributes in nature, culture and other aspects that cannot be replicated by other countries/regions within the same block. Recent years have seen the rapid development of international tourism. The number of international tourists and the amount of tourism revenues are measures of international tourism from the quantity point of view (e.g., Su & Lin, 2014; Balli et al., 2016; Liu, Li, and Parkpian, 2018) ; however, the network structure and evolution of international tourist flows lack attention. Essentially, international tourism involves cross-border activities (Deng & Hu, 2018) . Given the move toward globalization, the order of international tourism is constantly changing (Yang et al., 2018) and can be revealed by the movement of international tourists. Identifying the structure and evolution of international tourist flows is critical for understanding the changes in the past and for formulating effective strategies for future tourism development (Lew & McKercher, 2006) . In this regard, based on network analysis, this study empirically evaluates the evolution of international tourist flows between 221 countries/regions during the period 1995-2018 from the perspective of structure, rather than tourist arrivals or tourism revenues. Network analysis is an approach used to map and measure the flow paths of resources between actors within a network system (Zha, Shao, & Li, 2019) , which is suitable for exploring the movement of international tourists. Currently, scholars have applied this approach to the study of tourist flows (Zeng, 2018) . However, studies have been limited to a specific region, such as China (Leung et al., 2011) and Sicily (Asero et al., 2015) , and lack a global perspective with few exceptions (e.g., Lozano & Gutiérrez, 2018) . Moreover, these studies mainly centre on a specific year (e.g., Lozano & Gutiérrez, 2018; Zeng, 2018) . Great changes have taken place and are ongoing in the world order since the last century, which has also had a profound impact on international tourism (Yang et al., 2018) . Thus, this study applies network analysis to explore the roles, functions and evolutions of countries/regions over the world in tourism flow networks, thereby enriching the study of tourist flows from a global perspective. Understanding the structure and evolution of international tourist flows can be useful for improving market competitiveness and destination management. This study constructs the international tourist flow network and attempts to reveal the structure and evolution of this network from two levels: the whole network and actor. As for the whole network, the estimated results of the density indicator show that the international tourist flow network is a sparse network, but its density is on the rise. This is related to globalization (Keum, 2010) and government policies (Deng & Hu, 2018) , among other factors. This finding echoes the conclusion in the study of Friedman (2005) that the world is flat. Moreover, according to Var, Schlüter, Ankomah and Lee (1989) and Becken and Carmignani (2016) , globalization promotes international tourism around the world, whereas international tourism contributes to globalization, making tourism a real force for world peace. Moreover (Ritchie, 2008) , the financial crisis in 2008 (Hall, 2010) , the influenza A (H1N1) epidemic in 2009 (Lee, Song, Bendle, Kim, & Han, 2012) , and other factors. It should be noted that the studied period has seen several crisis events, including but not limited to the above-mentioned ones. However, only global crises, especially global public health crises, have an impact on the structure of the international tourist flow network. In this regard, we can forecast that the coronavirus disease 2019 (COVID-19), which continues to spread rapidly across the world, has led to a decline in the network density of international tourism flows. In terms of the actor, the role and function of a country/region in the international tourist flow network are identified utilizing the degree centrality indicator. The roles and functions of countries/regions within the outbound tourist network are a reflection of a country's economic development (Li et al., 2008) , the level of openness , price competitiveness index (Seetaram, Forsyth, & Dwyer, 2016) , government policy (Li, Harrill, Uysal, Burnett, & Zhan, 2010) and population (Li, Shu, Tan, Huang, & Zha, 2019) , while those of the inbound tourist flow network are related to tourism competitiveness (Mou et al., 2020) , tourist attractions (Su & Lin, 2014) and culture (Yang & Wong, 2012) , among others. Specifically, among these 221 countries/regions, Germany, Italy, the United Kingdom, France, Spain, Austria, the Russian Federation, the United States, Canada, Mexico, China, Hong Kong SAR and Macao SAR are among the top tourist-generating and receiving countries/regions from 1995 to 2018 and are regarded as the core actors within the international tourist flow network. This finding is consistent with the study of Lozano and Gutiérrez (2018) . These 13 countries/regions are concentrated in Europe, East Asia and North America, with vast territories (e.g., the Russian Federation, Canada, the United States), developed economies (e.g., Germany, the United States, France), relative political stability (e.g., Germany, China, the United Kingdom) or large populations (e.g., China, Mexico, the United States) on the whole (Li et al., 2008) . Germany, in particular, plays a leading role in the global outbound tourism market from 1995 to 2018, while China has acted as the dominating inbound tourism market since 2000 when considering the number of destination/origin countries/regions and international tourists. The Henley & Partners Visa Restriction Index shows that German passports are among one of the most valuable passports worldwide. For example, German passport holders can visit 176 countries worldwide visa-free in 2017 (Henley & Partners, 2017) . According to Wu et al. (2019) , China gave priority to inbound tourism from 1949 to 2008 for both political and economic reasons. Recently, China has developed government policies concerning tourism, such as the Belt and Road Initiative, largely enhancing the inbound tourism market and even changing China's inbound tourism market landscape . Moreover, other factors, including China's thriving history and culture (Lim & Pan, 2005) , cannot be ignored. Countries/regions that do not perform well within the international tourist flow network over the years are mainly located in Africa or on islands, such as Montserrat and Niue, with performances affected by safety concerns, transportation accessibility or small populations. This finding echoes the study of Li et al. (2008) . Besides, the majority of countries in Southeast Asia (e.g., Thailand and Malaysia) have relatively well-developed inbound tourism compared with outbound tourism due to the availability of abundant tourism resources, government support (e.g., the proposal of the Malaysia Tourism Transformation Plan) , a vast diversity of tourism products ) and a relatively low exchange rate (Seetaram et al., 2016) . Moreover, a small majority of countries/regions, such as the Republic of Moldova, Belarus and Sweden, play relatively more important roles and functions in outbound tourism than inbound tourism over the years. Regarding the structure corresponding to the country/region's role or position in the network, the CONCOR algorithm estimates the structurally equivalent countries/regions of the international tourist flow network flows in 2018. Most countries/regions with similar or the same external links in terms of international tourism are located on the same continent or are geographically close. This finding is in line with the study of Lozano and Gutiérrez (2018) that the clustered structure is determined by geographical factors. Geographically close countries/ regions have similar natural, cultural and political environments, which can affect the tourism industry (Yang et al., 2018) . For example, Narayan, Narayan, Prasad, and Prasad (2010) noted that Pacific Island countries, especially Fiji, the Solomon Islands and Papua New Guinea (in Block 8 of this study), have similar natural disasters and political instability, which can influence the choices of international tourists. Given the fierce competition in the international tourism market, countries/regions that are structurally equivalent need to provide different kinds of leisure products to be differentiated to international tourists. The policy implications are clear and of great significance. First, policymakers should analyze international tourist flows not only from the perspective of tourist arrivals and tourism revenues but also from the perspective of network structure. Future policies should be proposed, such as establishing partnerships with more countries/regions, to address the problems related to tourism in the increasingly globalized world. Second, policymakers should manage tourism routes, plan tourism facilities and define marketing strategies by identifying the roles and functions of countries/regions within the international tourist flow network. Third, according to the results of this study, countries/ regions that are geographically close have similar or the same international tourist flow structures. Thus, differentiated tourism products should be provided to create a unique and competitive tourism image for a country/region. It is important to note that this study has several limitations. First, the data used in this study is compiled by destination countries/regions, each of which may adopt distinct definitions of tourism and may collect tourist arrival data differently. Currently, there are 8 statistics definitions related to national borders or accommodation establishments, which may affect the accuracy of the data set used in this study. Second, due to the use of different tourism statistics systems, several countries/ regions only reported data for a subset of origin countries/regions, leading to missing values in the data set. Third, limited by the research goal, it is difficult to analyze every country/region in the world, which may ignore some important evaluations of individual countries/regions. Implications for future research involve a more in-depth exploration of the international tourist flow network. According to Welch, Welch, Young, and Wilkinson (1998) , as actors define various elements of the network and interact with the external environment, relationships between actors are constantly shifting. In other words, networks are dynamic, and links between countries/regions are both built and lost. Future research is needed to examine the factors (e.g., visa, air transportation) that affect the structure, and how the structure influences the socio-economic development of a country/region. Second, it is an exciting research endeavour to apply network analysis to establish relationships among different agents related to international tourism (e.g., air carriers, tourism service providers in destination). Third, considering the impact of crises on the tourism system, future research should focus on the impact of global crisis events (e.g., COVID-19) on the structure of international tourism (e.g., redistributing power and other resources in the network), and recovery measures. None. Zhiyong Li is a professor as well as the Dean of School of Tourism, Sichuan University, P.R. China. His research interests center on tourism marketing, outbound tourism and hospitality management. Mingzhi Luo is a lecturer in the School of Tourism, Sichuan University, P.R. China. His research interest includes tourism economics and tourism policy. He is currently working on tourism recovery affected by natural disasters. Y. Shao, et al. Tourism Management Perspectives 36 (2020) Implications of global climate change for tourism flows and seasonality Analysis of tourist behaviour based on the tracking data collected using a mobile communication instrument Building tourism networks through tourist mobility The impacts of immigrants and institutions on bilateral tourism flows Does tourism lead to peace? Analyzing social networks A cross-national analysis of international tourist flows in China Structural holes: The social structure of competition Models and methods in social network analysis Social network analysis in tourism Exploring the spatial patterns of car-based tourist travel in Loch Lomond Modelling China's outbound tourist flow to the "Silk Road": A spatial econometric approach Sentiment, mood and outbound tourism demand Between the hammer and the anvil: The impact of economic sanctions and oil prices on Russia's ruble Estimation of off-track visits in a nature reserve: A case study in Central Belgium The world is flat: A brief history of the twenty-first century The strength of weak ties Crisis events in tourism: Subjects of crisis in tourism An approach and technique for the study of information exchange Henley & Partners Passport Index Does the belt and road initiative stimulate China's inbound tourist market? An empirical study using the gravity model with a DID method Identifying the spatial structure of the tourist Y attraction system in South Korea using GIS and network analysis: An application of anchor-point theory Tourism flows and trade theory: A panel data analysis with the gravity model Network analysis Social network analysis Understanding tourist movement patterns in a destination: A GIS approach The impact of nonpharmaceutical interventions for 2009 H1N1 influenza on travel intentions: A model of goal-directed behavior The framework of tourism A social network analysis of overseas tourist movement patterns in Beijing: The impact of the Olympic Games Modeling tourist movements Estimating the size of the Chinese outbound travel market: A demand-side approach Spatial pattern of tourist flows among the Asia-Pacific countries: An examination over a decade Does the demographic structure affect outbound tourism demand? A panel smooth transition regression approach Inbound tourism developments and patterns in China Social forces social capital: A theory of social structure and action A panel data-based analysis of factors influencing market demand for Chinese outbound tourism Inbound tourism in Thailand: Market form and scale differentiation in ASEAN source countries A complex network analysis of global tourism flows Social network analysis Exploring spatio-temporal changes of city inbound tourism flow: The case of Shanghai Tourism and economic growth: A panel data analysis for pacific island countries A model of travel itineraries Intranational tourist flows in Malaysia Tourism disaster planning and management: From response and recovery to reduction and readiness Social network analysis. A handbook Destination networks Measuring price elasticities of demand for outbound tourism using competitiveness indices Network characteristics of drive tourism destinations: An application of network analysis in tourism Tracking tourists in the digital age Analysis of international tourist arrivals worldwide: The role of world heritage sites Tourism and world peace: The case of Argentina Small worlds: The dynamics of networks between order and randomness The importance of networks in export promotion: Policy issues Social structure from multiple networks. I. Blockmodels of roles and positions International tourism highlights World Tourism Organization. International tourism, receipts (current US$) Agent-based simulations of China inbound tourism network Modelling spatio-temporal movement of tourists using finite Markov chains The world is flatter? Examining the relationship between cultural distance and international tourist flows The influence of cultural distance on China inbound tourism flows: A panel data gravity model approach Pattern of Chinese tourist flows in Japan: A social network analysis perspective Linkage analysis of tourism-related sectors in China: An assessment based on network analysis technique The impacts of cultural values on bilateral international tourist flows: A panel data gravity model This study was supported by a grant from the National Social Science Fund of China (17XGL012).