Measuring communities' efficiencies within the global tourism network

The study situates international tourism within accelerating globalization, highlighting tourism’s role in global mobility, economic development, and complex inter-organizational networks. While prior work has described macro-level structures (e.g., small-world characteristics) and micro-level node attributes (e.g., degree, centrality), meso-level analysis of community structures remains underdeveloped. The paper addresses the lack of systematic, quantitative methods to analyze communities within global tourism networks, whose formation cannot be explained by geography alone and whose membership evolves over time. The study’s purpose is to (1) detect and track the evolution of communities in the global tourism network (1995–2021), (2) propose and operationalize efficiency metrics at the community level (structural, performance, functional), and (3) identify factors influencing these efficiencies using a mixed-effects modeling framework. Research questions:

1. Which communities/clusters exist within the global tourism network and how have they evolved over time?
2. How can we measure the characteristics and efficiencies of communities within real-world complex networks?
3. What factors impact the efficiencies of communities, particularly how individual characteristics affect communities within the tourism network?

Early international tourism studies predominantly modeled demand using gravity-type approaches emphasizing economic scale and distance, with determinants including income, prices, exchange rates, infrastructure, and proximity. Additional factors such as migration, linguistic networks, neighboring-country spillovers, special events, and negative impacts from crime, conflict, geopolitics, and risk have been documented. With advances in computation and graph theory, network science has been increasingly applied to tourism, revealing uneven global flow distributions, small-world properties, decentralization trends, and the predictive value of network structures for tourism flows. Scholars have used community detection, clustering, and modularity-based methods to identify tourism clusters and seasonal typologies, and to link network structure to performance and behavior using diverse data (e.g., mobile traces, travel notes). However, most work emphasizes global (macro) structures or node-level (micro) properties; meso-level community analysis remains limited, often focusing only on delineation rather than systematic metrics for community structure, function, and outcomes. The literature calls for methods to quantify community evolution mechanisms, integrate multi-source data, and link community insights to policy and management.

Research framework: Five steps—(1) compile international tourist flow data and construct annual directed, weighted global tourism networks (1995–2021); (2) detect communities via modularity optimization and code communities; (3) measure year-to-year community similarity and segment network evolution into stages; (4) develop and compute three community efficiency metrics (structural, performance, functional) and track their trends; (5) test effects of 11 factors on efficiencies using mixed-effects models. Data: International tourist flows from UNWTO (over 200 countries/regions; 1995–2021). Indicators prioritized: TFR and VFN (visitor arrivals), with TCER and THSR used when needed. Name harmonization and segmentation for territorial changes (e.g., Serbia and Montenegro) were applied. Final dataset: 322,877 records of international tourist statistics (1995–2021). Additional factors (Table 1): GDP, GDP per capita, total population (TP), urban population (UP), exports of goods and services (EGS), air passengers carried (APC) from World Bank; UNESCO counts of World Heritage (WH) and Intangible Cultural Heritage (ICH); wars/conflicts (WC) from Wars in the World; political violence and protest events (PVPE) from ACLED; geographical proximity (GP) computed by authors via Google Earth (bordering/maritime adjacency within 1000 km). Network construction: Annual directed weighted networks G={N,w}; nodes are countries/regions; edge weight w_ij equals the number of tourists from i (origin) to j (destination) in a given year. Community detection: Louvain/Blondel modularity optimization (Q in [−1,1]) partitions the network, iteratively aggregating nodes/communities to maximize Q, revealing hierarchical community structure. Community similarity: Year-to-year community membership similarity quantified using a Jaccard-style index J(G_i, G_{i+t}) = |Gi ∩ Gi+t| / |Gi ∪ Gi+t| to track membership stability and identify structural breaks. Community efficiency metrics:

• Structural efficiency (S): scale-adjusted measure of internal connectivity tightness using small-worldness-type ratio S = (C_G/C_random) / (L_G/L_random), referencing ER networks of equal size to control for scale effects. Higher S implies greater internal connectivity efficiency.
• Performance efficiency (P): extent to which a community achieves its objective of facilitating internal tourist flows, computed as the share of internal flow relative to total flow through member nodes (0 to 1), with higher values indicating higher internal performance.
• Functional efficiency (F): external influence of a community based on outbound flow from the community to the rest of the network, adjusted for community size; higher values (0 to 1) indicate stronger impact on other communities. Mixed-effects modeling: Community efficiencies (S, P, F) are dependent variables; 11 factors are independent variables (averaged over members within each community-year). Multicollinearity tested via VIF; UP removed (VIF>10). Final models include year as a random effect to account for temporal variation and handle unbalanced panels/missingness. Significance assessed for factor effects on each efficiency dimension.

Network growth and structure:

• Global international tourists increased from 540 million (1995) to 6.6 billion (2022), a 12.2-fold increase; network nodes and edges expanded through 2019 (234 nodes; 17,048 directed edges), then contracted sharply in 2020 due to COVID-19.
• Community counts by period (via modularity): 8 communities (1995–2001), 7 (2002–2004), 6 (2005–2021), except 2020 (7). Modularity cycles indicate higher cohesion in 1995–1996, 2001–2004, 2009–2013, and 2020. Evolutionary stages (via similarity and counts): Four stages—Stage I (1995–1998), Stage II (1999–2003), Stage III (2004–2019), Stage IV (2020–2021). Low similarity around 1999–2001 marks a break. Community compositions (coded A–I):
• A: Former Soviet Union countries (e.g., Russia, Belarus, Ukraine).
• B: Central/southern African countries (e.g., South Africa, Tanzania, Zimbabwe).
• C: Middle Eastern countries (e.g., Saudi Arabia, Qatar).
• D: Oceania and Indian/Pacific coastal countries (e.g., Australia, New Zealand; at times East/Southeast Asia/Indian Ocean neighbors).
• E: China, Hong Kong, Macau, Taiwan, DPRK; merged into D (2004).
• F: Western/most European countries, esp. Mediterranean; some Nordic countries bridge F and A.
• G: South American countries (e.g., Brazil, Chile); merged into H (2002).
• H: North and Central America.
• I: Emerged from F during the pandemic (e.g., Germany, Poland, Italy, Slovakia), reflecting COVID-19’s impact on European tourism. Inter-community flows:
• Stage I (1995–1998): Largest flow F→A (avg 59.513 million); second H→F (14.349 million).
• Stage II (1999–2003): A→F surged then stabilized at high levels.
• Stage III (2004–2019): Largest A→F (47.062 million), then H→F (32.954 million), underscoring Mediterranean Europe as a major destination.
• Stage IV (2020–2021): Main flow I→F; A→H and A→I increased; flows from A to E/H/I comprised 49.86% of all A’s outbound flows. Community efficiency metrics (Table 3 summaries):
• Structural efficiency (S): Highest average C (3.297), followed by E (3.023) and B (2.889). Variance high in C and B; overall downward trend over time.
• Performance efficiency (P): Highest average E (0.427), then B (0.407), D (0.371)/F (0.372)/I (0.376). F shows notable dips in 2014–2015 and 2020.
• Functional efficiency (F): Table 3 averages show E (0.037) and F (0.020) highest; others lower (e.g., A 0.014; B 0.002; H 0.009). Annual trend generally upward (except Stage IV). The narrative highlights F as functionally dominant, reflecting strong extra-community impact in Europe/China-linked clusters. Determinants of efficiencies (mixed-effects results):
• Structural efficiency (S): Significant positive effects of GP (p<0.05) and EGS (p<0.05); significant negative effect of APC (p<0.05); PVPE negative at p<0.1 with the largest magnitude coefficient (−6.132), indicating political violence strongly damages structural cohesion.
• Performance efficiency (P): APC positive (p≈0.066); GDP negative and significant (p<0.05), suggesting higher average GDP within a community associates with lower internal performance efficiency.
• Functional efficiency (F): TP positive (p≈0.08), APC positive (p<0.05), GDP negative and highly significant (p<0.001), indicating populous, air-connected communities exert more external impact, while higher average GDP correlates with lower functional efficiency. Comparisons to prior work: Community counts differ from Chung et al. (2020) and Seok et al. (2021) due to full coverage of countries/regions rather than top-destination subsets.

The findings address the three research questions by delineating stable yet evolving community structures, proposing and computing three efficiency metrics at the meso-level, and identifying key macro factors shaping these efficiencies. The four-stage evolution aligns with major global events (Soviet Union aftermath, China’s WTO accession, COVID-19), demonstrating that community composition and cohesion respond to geopolitical and economic milestones. Inter-community flow analysis highlights persistent centrality of Mediterranean Europe and strong Europe–Eurasia and Americas–Europe linkages. Methodologically, the structural (S), performance (P), and functional (F) efficiency indices enable robust meso-level assessment that is less sensitive to membership volatility. Empirically, results suggest that proximity and trade intensity bolster internal cohesion, whereas air traffic scale and political violence can weaken structural efficiency; at the same time, air connectivity enhances both internal performance and external influence, and higher average GDP may dampen both internal performance and external diffusion efficiency at the community level. These community-level effects differ from some country-level findings in the literature, underscoring scale-dependent mechanisms. For policy and management, recognizing communities as operative clusters helps international organizations and governments tailor cooperation, marketing, and connectivity strategies, considering proximity, export ties, security, and air transport capacity. Data-driven delineations also show that conventional continental splits (e.g., UNWTO/WTTC) can miss cross-continental affinities (e.g., North Africa with Europe/Middle East; East Asia with Oceania).

This study constructs annual global tourism networks (1995–2021), detects and codes communities, segments evolution into four stages, and proposes three community efficiency metrics (structural, performance, functional). It quantifies community trends, inter-community flows, and the impacts of macro factors via mixed-effects models. Main contributions include: (1) a meso-level paradigm for community-focused tourism network analysis; (2) a similarity index for tracking community membership stability; (3) scalable measures of community efficiencies; and (4) empirical evidence that proximity, trade, security, air connectivity, population, and GDP shape community efficiencies in distinct ways. Future research should further validate and refine the similarity and efficiency metrics across contexts; incorporate additional variables (e.g., language, visa regimes, cultural and policy factors); compare alternative modeling frameworks (e.g., panel SEM, Bayesian hierarchical models); and extend analyses to post-2021 dynamics to capture long-run post-pandemic reconfiguration.

• The proposed similarity and efficiency metrics, while theoretically grounded, require further empirical validation and sensitivity testing.
• Potentially important factors such as language, visa policies, and cultural distance were not included due to data limitations and may affect community structures and efficiencies.
• Only mixed-effects models were employed; alternative econometric/network modeling approaches could yield different insights.
• Analysis ends in 2021; ongoing post-COVID-19 changes in global tourism networks warrant continued observation and evaluation.