How regional characteristics drive various human mobility types across Europe

Abstract

Spatial mobility of people between regions is critically shaped by socioeconomic landscapes, influencing and being influenced by regional characteristics, especially amid the on-going green and digital transitions. However, this is often studied through the lens of a single mobility type, overlooking the diverse factors driving different forms of movement. This study examines the distinct impacts of regional characteristics on seven types of mobility—permanent migration, long-term and short-term student mobility, seasonal work, long-distance and cross-border commuting, and multilocal living arrangements—across European regions. By integrating data from the Labor Force Survey, Erasmus + , and Twitter, we show how regional characteristics have different effects regarding each mobility type influenced by unique combinations of economic, environmental, and social factors. These differences underscore the limitations of a one-size-fits-all approach in addressing interregional spatial mobility, highlight the need for policies tailored both to specific types of mobility and regional characteristics. Considering these insights, we present avenues for targeted policy recommendations towards enhancing regional attractiveness and supporting balanced territorial development.

Using large-scale data on seven mobility types across Europe, this study moves beyond permanent migration to reveal how regional characteristics shape diverse forms of movement and to inform more targeted regional policy.

Introduction

Spatial mobility of people is a critical factor shaping socioeconomic landscapes in our globalized and interconnected world^1,2. Stemming from the new mobilities approach³, mobility is inseparable from ‘moorings’—fixed physical, infrastructural, and institutional settings. At the same time, patterns of movement can also transform these settings over time. At the regional level, people moving between regions and regional characteristics influence each other, creating a dynamic interplay that reshapes economic opportunities, social structures, and environmental conditions^4,5. While regional attributes such as labor markets, quality of life, and infrastructure can foster mobility^6,7, the resulting population shifts can, in turn, alter these very characteristics^8,9, impacting labor market dynamics, regional disparities, and the sustainability of public services in both origin and destination regions. Additionally, megatrends such as the green and digital transitions further amplify the evolving nature of human mobility, adding layers of complexity to understanding its full impact.

As regions across Europe differ in socioeconomic, environmental, and infrastructural characteristics, these attributes collectively create regions of varied attractiveness, which in turn influence the mobility patterns of individuals. For example, regions with robust economies, high-quality infrastructure, and ample amenities draw more migrants, reinforcing local growth, whereas other areas experience outmigration¹⁰. Thus, regional attractiveness enables some areas to continuously attract new residents, whereas others struggle to retain their populations. Understanding these dynamics requires examining how regional characteristics interact, not just as isolated factors but as an integrated system of spatial mobility patterns. This holistic perspective allows us to investigate how certain regions evolve into functional hubs for economic and social activity, while others risk being “left behind” as they face economic stagnation and population loss.

At the same time, human mobility between regions extends beyond solely permanent migration, encompassing a wide range of movement patterns driven by diverse motives and needs¹¹. In contemporary societies, mobility is increasingly dynamic, shaped by factors such as economic opportunities, educational aspirations, and lifestyle preferences. Long-distance commuters regularly travel extensive distances for work¹², while seasonal workers temporarily relocate to meet labor demands in specific industries, such as agriculture or tourism¹³. Cross-border commuters optimize their income regardless of additional legal and taxation obstacles¹⁴. Students, too, engage in temporary mobility, often moving abroad for educational exchanges that enrich their academic and cultural experiences¹⁵. Moreover, digitalization has enabled a multilocal lifestyle, empowering individuals to maintain more than one home and divide their time between urban and rural areas or across borders¹⁶. These mobility types are driven by distinct motivations and influenced by varying regional characteristics. For instance, economic conditions and labor market structure may play a critical role in attracting seasonal workers, while educational institutions and urban amenities may be key determinants for student mobility. The directionality of these flows may also vary significantly; regions experiencing outmigration due to permanent migration may simultaneously attract seasonal workers or multilocal residents drawn by tourism or natural amenities.

Despite extensive research on spatial mobility, only a limited number of studies examine how regional characteristics shape multiple types of mobility simultaneously, particularly beyond permanent migration and at a pan-European scale. Most existing work still treats specific mobility forms or individual regional factors in isolation, which constrains our understanding of how different types of movement respond to diverse territorial contexts. A key reason for this gap is the scarcity of harmonized, comparable data that capture different forms of mobility across regions. Our study addresses this challenge by assembling and integrating multiple large-scale datasets, including survey, administrative, and digital trace data, to construct a comprehensive picture of seven distinct mobility types. Analyzing these within a single, consistent framework allows us to reveal how their determinants converge or diverge across contexts, and how they collectively shape regional dynamics.

Background on permanent migration and influencing drivers. Permanent migration is shaped by economic, social, political, and environmental factors¹⁷. Economic conditions like employment opportunities and higher living standards are consistently motivating migration, especially from rural to urban regions^18,19. At the same time, social factors such as family ties and cultural settings influence individuals’ choices to move^20,21, and political drivers due to government policies and political turbulence can force or attract people to relocate^22,23. Not the least, environmental drivers, including natural disasters and climate change, further prompt populations to move seeking safer and more favorable living conditions^24,25.

Focusing on the economic perspective, studies have long investigated permanent migration. The push-pull framework, originating with Ravenstein’s Laws of Migration²⁶ and developed in gravity models²⁷, explains migration as driven by population size, distance, and wage disparities^28,29. Wage differences and employment prospects strongly affect mobility, while economic activity and labor demand within a region further determine migration flows^30–32. Labor market structures are also critical, as high- and low-skill jobs in advanced economies draw workers from different regions and skill levels, fostering interregional commuting and migration^33–35.

Beyond economic incentives, quality-of-life factors have also become prominent in understanding migration decisions. Regional amenities, public services, and low barriers to integration significantly impact an area’s attractiveness, particularly for highly educated individuals and the creative class^36–38. Research suggests that access to cultural and recreational facilities and a vibrant social scene can boost a region’s ability to retain and attract skilled populations^39–41. Regions offering improved living conditions and a high quality of life, thus, often emerge as hubs of human capital concentration and innovation.

Moreover, social networks are deeply intertwined with spatial mobility, shaping and sustaining ties across varying geographic scales. People tend to establish localized networks around places of residence or work, but as mobility increases, social networks spread out spatially, adjusting to new activity locations^42–44. Similarly, digitalization and digital social networks have increased the geographical extents of social networks beyond geographical constraints and influence migration decisions^45,46. Network dynamics vary with life stages, influencing the types of social interactions and roles that motivate mobility decisions^47,48. For example, young adults may move to urban areas for education or career opportunities, driven by peer networks and professional aspirations, while older individuals might relocate closer to family or seek retirement destinations with strong community ties. Social connections, family status, and cultural ties all contribute to the complexity of spatial mobility choices^49–51.

Political and environmental factors may also play a crucial role in shaping permanent migration, given short-term disruptions and long-term changes. Political instability, conflicts, and restrictive governmental policies can act as both push and pull forces, often compelling individuals to leave regions where governance is weak or human rights are compromised^22,23. For example, periods of political upheaval or state repression have historically led to waves of forced migration, as populations seek safety and stability in more secure regions⁵². Moreover, environmental drivers, ranging from acute natural disasters to long-term climate change, exert increasing pressure on communities. Extreme weather events such as floods, droughts, and hurricanes can render areas uninhabitable, forcing residents to relocate in search of improved living conditions^24,25. Gradual environmental degradation and resource scarcity have also emerged as significant contributors to migration flows, as affected populations adapt to changing ecological conditions⁵³. These political and environmental influences not only interact with economic opportunities and social networks but also contribute to complex, multifaceted migration dynamics that are becoming increasingly relevant in the context of global change^54,55.

Background on other types of mobility and influencing drivers. While permanent migration is one of the most studied forms of spatial mobility, it represents only a fraction of the diverse ways people move between regions. Other mobility types, such as student mobility, seasonal work, long-distance commuting, cross-border commuting, and multilocal living, are shaped by unique regional characteristics and remain relatively underexplored, particularly in a European-wide context. Here, we briefly describe several additional mobility types and their links to regional factors, as they have been identified by existing literature.

Student mobility, particularly within Europe, is driven by factors such as the reputation of universities, urban amenities, and cultural or climatic appeal^15,56. Despite the attention given to this mobility type, studies at the European scale primarily focus on long-term movements, with short-term exchanges like Erasmus exchange programs receiving less comprehensive analysis^57–59.

Another mobility type that closely relates to regional characteristics but still lacks investigation within the European context is seasonal work mobility. Seasonal work mobility involves temporary relocations in response to labor demand in specific industries, often tied to tourism and agriculture¹³. In this case, regional sectoral structure plays a key role in the attractiveness of a region as a destination for seasonal workers. Coastal regions around the Mediterranean, Nordic wilderness areas, and Alpine Mountain destinations exemplify regions benefiting from seasonal influxes of workers^60,61.

Long-distance commuting refers to regular travel between a place of residence and a distant workplace. This form of mobility has been extensively examined within individual countries, where it is often influenced by factors such as job opportunities concentrated in urban centers and more affordable housing options in peripheral areas^62–64. However, this type of mobility remains underexplored on a European scale. Addressing this gap is essential for understanding the broader patterns and drivers of long-distance commuting across diverse regional contexts within Europe.

Enabled by open-border policies and freedom of movement within the European Union (EU), cross-border commuting, regular travel between a place of residence in one country and a workplace in another⁶⁵, has grown significantly, particularly in regions with pronounced wage and cost-of-living differences. For instance, many individuals commute daily between Luxembourg and its neighboring countries, or between Switzerland and France, drawn by economic opportunities and lower cost of living on either side of the border^66,67.

Multilocal living, the practice of maintaining multiple residences and dividing one’s time among them, is a distinct mobility type relevant in the Nordic countries’ context, but also elsewhere in Europe⁶⁸. This mobility type is typically associated with retirees and professionals who have the privilege of working remotely, yet the phenomenon is more diverse⁶⁹. To date, studies on multilocal living are limited to a country or bilateral contexts without a European-wide perspective^68,70,71.

Major transitions and the changing patterns of spatial mobilities of people. Throughout history, major transitions—such as economic shifts, technological advancements, and societal transformations—have profoundly influenced how and why people move across regions. These changes have reshaped migration patterns, commuting behaviors, and the wider connectedness of regions, reflecting broader trends in human activity and adaptation.

Recent studies suggest that the green transition indirectly influences individual mobility and its spatial outcomes⁷². Specifically, the European Green Deal⁷³ is expected to lead to increased factor mobility and the redistribution of economic and social resources, concentration of green technologies, jobs, and innovation in regions more prepared for sustainable economic activities. While skilled migration is crucial to this process due to skill-biased technological changes associated with the green transition⁷⁴, its impact also extends to unskilled workers, affecting the entire workforce. The green transition impacts semi-skilled and unskilled workers, especially in traditional industries like fossil fuels, mining, and heavy manufacturing. These sectors may experience job losses due to automation and decarbonization efforts; however, the transition also holds potential for creating new jobs in renewable energy, environmental management, and sustainable manufacturing⁷⁵. Regions that are less developed may struggle to take advantage of these opportunities, potentially facing issues like unemployment, skill mismatches, and brain drain⁷⁶. The movement of both skilled and less-skilled individuals to more prosperous areas could worsen regional inequalities, leaving behind less-skilled workers and lower-productivity firms. This concentration of opportunities may further drive migration flows, deepening territorial divides and increasing social discontent in polarized societies^77,78.

At the same time, the digital transition is redefining regional attractiveness by changing how individuals access services and influencing migration trends. While there are concerns about the digital divide and the potential for marginalization, the rise of virtual services, such as telemedicine and e-government, brings economic, environmental, and social advantages by offering remote access to key services⁷⁹. However, rural regions may encounter obstacles in this shift, not only due to limited digital literacy or reluctance to embrace new technologies⁸⁰, but also because of “last mile challenges”. In many rural and peripheral areas, the per capita cost of laying down infrastructure rises significantly, making it less feasible to provide comprehensive digital access. Migration, often viewed as a driver of development through the transfer of skills and resources, adds further complexity. In Europe, the potential of migrants to stimulate innovation and foster local growth is a key topic, especially in disadvantaged or left-behind areas⁸¹. However, a spatial mismatch often arises, as regions most in need of both skilled and unskilled migrant labor are usually not the destinations that migrants prefer, creating further challenges for local development. Here, twin transition could support compensating permanent “brain-drain” by people with a multilocal lifestyle temporarily residing and engaging with the local community in rural regions⁸².

Finally, the recent COVID-19 pandemic, through travel restrictions, disrupted daily mobility patterns of people within cities and between regions for commuting to work and for leisure purposes^83,84. It has also motivated people escaping temporarily from cities to their second homes in the countryside or a neighboring country⁸⁵, and affected the urban-rural dynamics regarding permanent migration flows⁸⁶. The significant shift toward remote work due to COVID-19 restrictions has allowed many people to reconsider their preferences on where to live and explore relocating to more peripheral areas, thus sparking changes in real estate and migratory patterns across major cities in many countries around the globe^85,87,88. However, cities still remain vital locations as hubs of innovation and amenities, suggesting that urban centers retain their economic importance⁸⁹. This is further reinforced by corporate policies requiring employees to be present on-site for several days a week, ensuring continued demand for urban workspaces and services. Yet, little is known about how these societal transitions are reshaping mobility patterns beyond permanent migration and daily commuting.

Summary of the study. In this study, we first assess the effects of economic performance, living conditions, social fabric, environmental quality, and digital infrastructure on seven types of interregional mobility (permanent migration, long-term and short-term student mobility, seasonal work, long-distance and cross-border commuting, and multilocal living arrangements), highlighting the value of an integrated perspective for this type of analysis. Second, we provide insights to enhance our understanding of regional attractiveness resulting from spatial mobility choices. By investigating these interdependencies, our study provides input to support the development of targeted regional policies aimed at promoting balanced growth by mitigating the unbalanced effects of spatial mobility across Europe.

Results

Regional mobility flows in Europe

Our analysis of regional mobility flows across Europe highlights distinct patterns for the various mobility types (Fig. 1). Permanent migration data, based on Twitter, reveals a total of 24 million movements, with Germany, France, Spain, and Italy as top destinations, especially the capital regions like Île-de-France and Berlin. Long-term student mobility through Erasmus+ saw over 40,000 movements primarily to Germany, France, and the UK, with regions like Munich and Île-de-France drawing the most students. Seasonal work mobility, totaling about 22 million movements, primarily directed flows to Germany, France, and Italy, though Luxembourg and Norway saw notable international movement. Short-term student mobility involved nearly 1.7 million Erasmus participants, with Spain, Germany, and France as preferred destinations. Long-distance commuting data shows 250 million movements concentrated in Germany, France, and Austria, with Switzerland and Luxembourg receiving a high share of incoming commuters. Cross-border commuting, estimated at 22 million movements, was particularly significant in Luxembourg, Switzerland, and Norway, with notable flows between neighboring countries like France and Switzerland. Finally, multilocal living arrangements, identified through Twitter data, showed over 405,000 individuals maintaining multiple residences, especially within France, Germany, and Italy. The temporal evolution of these mobility types is presented in Supplementary Materials – Section 1, providing additional insight into how the intensity of different mobility forms has changed over time. Further details are available in the official report D2.1 of the MOBI-TWIN project⁹⁰.

Fig. 1. Spatial distribution of regional mobility flows across Europe.

Total mobility flows across Europe from 2012 to 2022 for various mobility types: a permanent migration, b long-term student mobility, c seasonal work mobility, d short-term student mobility, e long-distance commuting, f cross-border commuting, and g multilocal living arrangements. All maps are displayed using the ETRS89-extended/LAEA Europe projection (EPSG:3035).

Influence of regional characteristics on mobilities

Our analysis of regional characteristics explores their varied and sometimes contrasting effects on different mobility types across Europe. In the main text, we primarily focus on models 4 and 5 (Eqs. 4 and 5 in our Statistical Analysis Section), which incorporate dyadic origin–time and destination–time fixed effects to account for push and pull factors, respectively. We focus on these two models, as they control for multilateral resistance to mobility created by potential other destinations caused by different years and the heterogeneity of push and pull factors. However, for multilocal living, we rely on the first model since data for this mobility type covers only a single period (2012–2022), making time fixed effects unavailable. While models 2, 3, 4, and 5 produced relatively consistent results, model 1—where only time-related shocks were accounted for without controlling for multilateral resistance, bilateral relationships, or push and pull factors—yielded different outcomes. This divergence is especially evident when comparing model performance metrics. In particular, the constant lower pseudo R² values for model 1 (Table 1) underscore its limited explanatory power. Detailed results for each model and mobility type are available in Supplementary Materials – Section 2. It is important to note that while Model 1 captures cross-sectional differences between regions (e.g., higher Gross Domestic Product (GDP) vs. lower GDP), Models 2–5 estimate within-region effects over time, focusing on how changes in regional characteristics are associated with mobility flows. For a more detailed explanation of the models, see our Statistical Analysis Section. While these results provide valuable insights into how regional characteristics are statistically linked to different mobility patterns, they should be interpreted as associations rather than evidence of causation. Despite the inclusion of extensive fixed effects, omitted variable bias cannot be entirely ruled out, and we do not attempt to capture potential feedback dynamics through which mobility itself might influence regional characteristics. Therefore, the results presented here should be understood as robust correlational patterns that help identify important relationships and guide future causal research, rather than definitive statements about directional effects.

Table 1.

Pseudo R² values for each mobility type across five model specifications

Mobility	Model 1 Pseudo R2	Model 2 Pseudo R2	Model 3 Pseudo R2	Model 4 Pseudo R2	Model 5 Pseudo R2
Permanent migration	0.64	0.90	0.89	0.90	0.90
Long-term student mobility	0.25	0.39	0.66	0.41	0.41
Seasonal work	0.64	0.77	0.88	0.78	0.77
Short-term student mobility	0.45	0.64	0.83	0.64	0.64
Cross-border commuting	0.62	0.75	0.87	0.77	0.77
Long-distance commuting	0.68	0.78	0.89	0.78	0.78
Multilocal living	0.58	-	-	-	-

Economic factors are associated with mobility patterns in diverse ways across Europe (Fig. 2). An increase in GDP per capita at the origin is linked to an increase in long-term student mobility and permanent migration, and declining seasonal work and long-distance commuting flows, suggesting that growing regional prosperity may reduce the need to move for work but may provide opportunity for studying in other regions. At the destination, a higher GDP may attract cross-border commuters but also deter seasonal workers. Rising employment at the origin tends to reduce outflows, while higher employment at the . The production and weighting of the mobility data. Our workflow for extracting relevant mobility flows from the three primary datasets, and the order of the weighting schemes we used to produce the final mobility data used in this article. destination may draw in students and seasonal workers. Longer work hours at the origin and destination are associated with greater student mobility, while at the destination, they deter long-distance commuters. For multilocal living, flows concentrate in regions with higher GDP, longer work hours, and lower employment, reflecting cross-sectional differences rather than temporal changes.

Fig. 2. Economic, digitalization, and connectivity factors associated with mobility types.

Statistically significant coefficients (p < 0.05) for economy and labor market, digitalization, and access and connectivity variables across different mobility types. Mobility forms are color-coded according to the mobility categories defined in the “Methods” section: long-term (rose red and mustard yellow), short-term (blue and light blue), and circular (green, purple, and gray). Coefficients are based on model 4 (pull factors at the destination) and model 5 (push factors at the origin).

Digitalization factors are associated with mobility in different ways at the origin and destination (Fig. 2). Increases in average mobile internet speed at the origin are associated with higher rates of cross-border commuting, long-distance commuting, long-term student mobility, and multilocal living, suggesting that improved digital connectivity may support more flexible or temporary forms of movement. At the destination, however, faster mobile internet tends to reduce long-distance commuting and seasonal work. A growing share of high-tech employment suggests a decrease in permanent migration. At the destination, it attracts short-term students and cross-border commuters. For multilocal living, both digital connectivity and high-tech employment are positively associated with inflows, reflecting broader regional differences rather than temporal dynamics.

Increases in tourism and transport infrastructure are associated with mobility patterns differently across regions (Fig. 2). An increase in tourist arrivals at the origin and destination is associated with higher permanent migration and lower cross-border commuting. For multilocal living, tourist intensity at both origin and destination is positively associated with flows, reflecting the structural attractiveness of tourism regions for multilocal individuals. Increases in transport infrastructure at the origin and destination may reduce multilocal living.

Living conditions are significantly associated with mobility, with notable differences across mobility types (Fig. 3). Increases in population density at the origin are associated with higher outflows of seasonal workers and commuters, suggesting that factors like congestion or . The production and weighting of the mobility data. Our workflow for extracting relevant mobility flows from the three primary datasets, and the order of the weighting schemes we used to produce the final mobility data used in this article. housing pressure may prompt temporary and work-related mobility. At the destination, higher population density draws long-term students and commuters while discouraging permanent migration and seasonal work. Rising life expectancy at the origin may reduce seasonal worker outflows and short-term student mobility, whereas at the destination, it may attract seasonal workers and long-term student mobility, indicating that better living standards may enhance a region’s pull, especially for short-term movements. Higher crime rates at the origin are associated with a decrease in student mobility, and at the destination, it is linked to reduced inflows across several types. For multilocal living, regions with higher crime rates and greater population density tend to attract more inflows, whereas higher life expectancy is associated with lower levels of multilocal living.

Fig. 3. Living conditions and social fabric factors associated with mobility types.

Statistically significant coefficients (p < 0.05) for living conditions and social fabric variables across different mobility types. Mobility forms are color-coded according to the mobility categories defined in the “Methods” section: long-term (rose red and mustard yellow), short-term (blue and light blue), and circular (green, purple, and gray). Coefficients are based on model 4 (pull factors at the destination) and model 5 (push factors at the origin).

Social fabric at the regional level reveals nuanced associations with mobility (Fig. 3). An increase in median age at the origin may reduce permanent migration and cross-border commuting. At the destination, a higher median age is associated with increased student inflows but may decrease seasonal work and cross-border commuting. A higher women-to-men ratio at the origin may strongly increase all mobility types except permanent migration. At the destination, a rising female share may reduce short-term student mobility and cross-border commuting. A higher share of students per worker tends to increase student mobility from the origin regions and decrease it at the destinations, as expected. However, at the destination, it is also associated with greater levels of seasonal work and commuting. For multilocal living, regions with a higher share of women and a higher median age are more likely to attract multilocal residents.

In terms of regional typologies, an increase in the urban population share at the origin is strongly associated with reduced long-term student mobility, seasonal work, and cross-border commuting, but may lead to higher long-distance commuting and permanent migration (Fig. 4). At the destination, a higher urban share may increase student mobility while reducing seasonal work and cross-border commuting. Mountainous regions at the origin are linked to increased permanent migration, seasonal work, and short-term student mobility, and considerably lower cross-border commuting, whereas a higher mountainous population at the destination decreases long-term student and seasonal worker flows. A larger coastal population at the origin is positively associated with both student mobility and work-related movements. At the destination, however, regions with higher coastal populations tend to attract fewer long-term students but more cross-border commuters.

Fig. 4. Regional typology and environmental factors associated with mobility types.

Statistically significant coefficients (p < 0.05) for regional typology, landscape, and environment variables across different mobility types. Mobility forms are color-coded according to the mobility categories defined in the “Methods” section: long-term (rose red and mustard yellow), short-term (blue and light blue), and circular (green, purple, and gray). Coefficients are based on model 4 (pull factors at the destination) and model 5 (push factors at the origin).

Environmental and landscape characteristics show relatively weak associations across different mobility types (Fig. 4). Higher maximum temperatures at the origin are linked to lower permanent migration but greater short-term student mobility, while warmer destinations tend to attract more cross-border commuters. The effects of precipitation on mobility appear minimal.

Discussion

Our comprehensive analysis of various mobility types across European NUTS2 regions reveals significant insights into the factors driving spatial mobility and their implications for regional development. By integrating multiple datasets, including the Labor Force Survey (LFS), Erasmus+ student exchange data, and geotagged Twitter data, we have identified how regional characteristics are associated with patterns of permanent migration, student mobility, work-related mobility, and multilocal living arrangements. These findings not only highlight the association between the regional characteristics and regional mobility but also shed light on the broader socioeconomic effects on both developed and left-behind regions.

Our results demonstrate a pronounced concentration of mobility flows of different types toward economically prosperous, densely populated regions with favorable living conditions and advanced digital infrastructure. The concentration of mobility in already developed regions exacerbates regional disparities, posing significant challenges for less developed or rural areas^18,19,91, as well as social disparities within a region. For individuals, mobility provides opportunities for improved living standards, education, and career advancement. However, it can also entail challenges such as social isolation, cultural adjustment, and the costs associated with relocation^92,93. The challenges extend beyond individuals to the communities they leave behind. Less developed regions may struggle with talent outflow, reduced economic vitality, and declining populations, which can lead to a vicious cycle of underdevelopment^20,94,95. The outmigration of skilled workers and students deprives these regions of human capital essential for innovation, economic diversification, and competitiveness⁹⁶. The loss of young and educated individuals can weaken local economies, reduce consumer bases, and strain public services due to a shrinking tax base⁹⁷. This dynamic can lead to increased regional inequalities and social discontent, further hindering development efforts. On the contrary, the inflow of talent and labor to prosperous regions can strain urban infrastructures, inflate housing markets, and widen social inequalities^98,99. While these regions benefit from an influx of skills and economic activity, they may also face challenges related to overcrowding, environmental degradation, and social cohesion.

Our findings underscore the varying influence of regional characteristics on different mobility types, suggesting that a tailored approach to mobility policies could provide more nuanced solutions to regional disparities. Traditional strategies typically focus on reducing outmigration and attracting new residents to less developed regions, as well as managing the inflow of permanent migrants in prosperous areas. However, this narrow focus overlooks the potential of other mobility types to revitalize less developed areas, particularly given the increasing prevalence of flexible living arrangements facilitated by remote work, such as multilocal lifestyles and digital nomadism^{16,68,69,100,101}. Rather than solely aiming to attract permanent residents, regions could leverage their unique characteristics to foster seasonal, short-term, or multilocal mobility, bringing economic and social vitality in alternative ways, as some regions have already begun to do^102,103. For instance, a region with natural assets suitable for seasonal industries, like ski resorts, could concentrate on attracting tourists and seasonal workers rather than permanent migrants, thereby boosting the local economy and creating temporary but impactful population flows. Partnerships with higher education institutions for traineeships could further diversify these flows and strengthen ties to academic communities. While our results can offer valuable insights into the types of regional characteristics associated with different mobility forms, they should not be interpreted as direct policy prescriptions. Instead, they highlight possible avenues that warrant further investigation before informing concrete interventions.

Based on our analysis and following the mobilities approach³, we assume that every type of mobility can result in heterogeneous outcomes, which can have either positive or negative impacts for a region. Consequently, mobility-related policies should be aligned with a region’s specific strengths, constraints, and social priorities, while recognizing that not all forms of mobility are equally desirable or sustainable. Rather than treating increased mobility flow as an objective in itself, policymakers may need to consider which forms of mobility are compatible with local well-being, social equity, and environmental sustainability^104–106, and which may intensify pressures such as exclusion, housing stress, or uneven dependence on external flows¹⁰⁷. In this sense, regional attractiveness should be interpreted cautiously: some regional assets may support valuable forms of mobility, but they may also generate selective access and uneven territorial outcomes^78,107. Therefore, the policy relevance of our findings lies less in maximizing mobility than in identifying how different mobility types relate to regional characteristics, and in using that knowledge to design more context-sensitive and socially balanced interventions.

For example, remote and left-behind areas are often border regions located far from national capitals. However, borders can serve as valuable local resources¹⁰⁸. In the EU, differences between neighboring countries, such as wage levels, living costs, taxation, and availability of goods, foster various types of cross-border mobility, including commuting for work, shopping, leisure, and visiting friends and relatives⁶⁷. These interactions contribute to the vitality of border regions, making them more attractive and potentially offsetting permanent outmigration¹⁰⁹.

There are some potential considerations and limitations related to this study. First, the models examining different types of mobility included varied numbers of observations due to data availability on regional characteristics. Some flows had to be excluded from the analysis because of missing information on the regional characteristics related to the origin or the destination of the movement, which limited the overall dataset size. Additionally, the choice of variables was constrained by data availability and the need to address multicollinearity to maintain model robustness. Consequently, other relevant factors, such as housing market dynamics or indicators of economic agglomeration, influencing mobility may not have been included.

Second, while we controlled for regional area in our statistical analyses, differences in the size of NUTS2 regions can significantly impact the observed number of movements. Smaller regions may exhibit higher movement counts simply due to their limited geographical scope, potentially skewing results and interpretations. Conversely, larger regions may obscure internal movements within their boundaries while still recording higher inflows and outflows, as they often encompass multiple cities or, in some cases, entire countries, such as Estonia. While the regional size effects were captured by fixed effects in our analysis, and thus, not directly estimated, improved data availability at a finer spatial resolution, like the NUTS3 level, could enhance the accuracy of future analyses. A further limitation concerns the classification of multilocal living. We use a 50% movement frequency threshold to separate frequent commuting from less frequent multilocal arrangements. While this threshold is conceptually reasonable and captures the behavioral difference between routine travel and maintaining multiple residences, the lack of ground-truth data means we cannot validate or test alternative cut-offs. As a result, this classification should be seen as an informed approximation rather than a definitive boundary. We performed a sensitivity analysis with different thresholds (25%, 50%, and 75%) to assess the sensitivity of our approach to identifying multilocal mobility patterns, and found 50% to perform well (see Supplementary Materials – Section 3).

Third, given the relatively large number of explanatory variables and the use of multiple fixed-effects specifications, there remains a potential risk of overfitting, despite our efforts to mitigate it through theory-driven variable selection, multicollinearity checks, and the use of high-dimensional fixed effects. Furthermore, because origin–destination flows are inherently spatial processes, neighboring regions may share unobserved characteristics that lead to spatial autocorrelation in explanatory variables or residuals. Our modeling framework addresses this by incorporating origin–time and destination–time fixed effects, which capture evolving regional shocks, as well as dyadic fixed effects, which account for persistent bilateral structures such as geographic proximity, historical ties, and network linkages. This approach ensures that model identification relies primarily on within-region temporal variation, thereby limiting bias from spatial dependence. Nevertheless, some residual spatial autocorrelation may persist, reflecting spatial interdependencies beyond the scope of fixed-effects adjustment. Moreover, while some regional outcomes may respond to mobility with delays, our models capture only contemporaneous associations; exploring lagged dynamics remains a direction for future research.

A key limitation of our study lies in the inherent biases and constraints of the data sources used. Despite growing interest in spatial mobility at the regional level, even basic origin–destination data for permanent migration remain unavailable at the regional level in many European countries. As such, we relied on the best data sources currently accessible, each with its own limitations. The LFS data, while comprehensive, exhibits inconsistencies across countries and years, such as geographical disparities in the reported home and work locations, which were intended to be at least at the NUTS2 level. Additionally, some countries, such as Germany, are overrepresented in the dataset. A significant limitation of the LFS dataset is the lack of unique identifiers, preventing the tracking of individuals over time or across countries, which would have greatly enhanced the analysis. Twitter data, meanwhile, represents only a subset of the population, predominantly younger individuals, which may not fully reflect broader mobility patterns. Although a weighting scheme was applied to account for the market share of Twitter users across countries and age groups, differences in mobility behavior across age groups may still introduce biases into the models. Similarly, Erasmus+ data does not capture student mobility taking place outside the Erasmus framework, such as mobility supported by bilateral agreements or other mobility frameworks like Nordplus or the Turing Scheme. These limitations underscore the need for more comprehensive and representative datasets at the regional level to enhance the robustness and generalizability of future analyses.

To conclude, our study emphasizes the need to move beyond traditional analyses of permanent migration to fully understand regional mobility patterns and their implications for regional attractiveness. By examining multiple mobility types, such as student mobility, seasonal work, commuting, and multilocal living, we reveal that different forms of movement are shaped by distinct regional characteristics across economic, societal, environmental, and digital dimensions. This nuanced understanding is essential for designing policies that help regions adapt and leverage the benefits from ongoing transitions, including the green and digital transitions and the rise of remote work arrangements.

Our findings highlight that mobility types do not respond uniformly to regional characteristics but are shaped by distinct combinations of assets and attributes. This underscores the importance of examining diverse mobility flows to understand how regions might strategically align their strengths with the mobility forms most compatible with their development trajectories. For example, policies informed by these associations, such as enhancing digital connectivity, leveraging natural amenities, or fostering educational partnerships, could help diversify mobility flows and contribute to a more dynamic regional context, albeit with further research required before practical implementation.

Rather than focusing solely on preventing outmigration or attracting permanent residents, policymakers may benefit from a broader, asset-based perspective that considers how different mobility types relate to regional characteristics. However, encouraging mobility should not be treated as an unquestioned goal. Different forms of mobilities can create opportunities for some regions and populations while generating exclusionary, unequal, or environmentally unsustainable outcomes for others. For this reason, policies informed by our findings should aim not to maximize mobility, but to assess which forms of mobility are socially equitable, territorially appropriate, and environmentally defensible in particular contexts. Future policy design must therefore address the inherent tension between fostering economic vitality and preventing social exclusions, such as gentrification or the marginalization of those with lower mobility. In this way, we encourage future research to emphasize a more differentiated understanding of how every type of mobility can support more balanced and context-sensitive regional development across Europe.

Methods

Mobility data

In this study, we explore seven types of mobility, which we organize into three primary categories: long-term, short-term, and circular mobility. Long-term mobility includes permanent migration and long-term student mobility, where individuals stay in the destination region for over 12 months. Short-term mobility encompasses short-term student mobility and seasonal work, characterized by stays between three and 11 months. Lastly, circular mobility involves habitual, frequent movement between origin and destination regions, typically including return trips to the origin.

We sourced the mobility data from three primary datasets: the Labor Force Survey (LFS), Erasmus student exchange data, and geotagged Twitter data (now X) from the study period between 2012 and 2022. Each data source required specific preprocessing methods to extract relevant mobility types (Fig. 5). To maintain consistency across the various datasets, the geographical unit of analysis is the NUTS2 region, as defined by the Nomenclature of Territorial Units for Statistics (NUTS) classification system¹¹⁰. The data were collected as part of the MOBI-TWIN project, which investigates the impact of the green and digital transitions on regional mobility across Europe.

Fig. 5. The production and weighting of the mobility data.

Our workflow for extracting relevant mobility flows from the three primary datasets, and the order of the weighting schemes we used to produce the final mobility data used in this article.

Labor Force Survey (LFS)

The LFS data were used to analyze three types of mobility: long-distance commuting, seasonal work, and cross-border commuting. The data were processed annually to capture these mobility types, with population weights provided by the LFS used for aggregation at the annual level. To ensure comparability across Europe, additional country-level weights were applied based on the working-age population for each year. We further enhanced the dataset using ancillary sources to address gaps and improve comparability across regions (see Supplementary Materials – Section 4 for details).

Erasmus student exchange data

The Erasmus student mobility data covered the period 2014–2022, detailing individual student exchanges¹¹¹. To focus on meaningful mobility, we selected exchanges lasting at least three months. For this analysis, we used the georeferenced version of the Erasmus dataset¹¹¹.

Twitter data (X)

Geotagged tweets from Twitter were used to derive patterns of permanent migration and multilocal living. The raw data were extracted from an archive of historical Twitter data that consists of the activities of all users who tweeted at least once from within the bounding box of the European¹¹². Historical raw data was collected via the Twitter Academic API¹¹³. Each tweet in the dataset was timestamped and geolocated using H3 spatial indexing^114,115. To ensure data quality, we applied a multi-step filtering process. The top 5% of users with the highest number of tweets were removed, as these accounts were likely to represent automated bots or atypical usage patterns¹¹². In addition, we excluded users with fewer than 180 days (six months) of activity, ensuring sufficient temporal coverage for reliably identifying residential locations and mobility behaviors. Each user was assigned a unique identifier to track movements and identify key locations, such as the region of residence.

We implemented a weighting procedure to improve representativeness and comparability across regions because Twitter users are not representative of the general population, with younger, urban, and male users overrepresented, and platform penetration varying widely across Europe. First, we combined Eurostat population data by age group with external estimates of Twitter usage by age bracket to approximate the likely distribution of Twitter users within each country. This adjustment accounted for differences in age structure, mitigating systematic underrepresentation of regions with older populations. Second, for each NUTS2 region, we calculated a regional penetration rate by dividing the number of identified Twitter users with home locations by the age-adjusted population. This rate was applied to the mobility outcomes to generate more realistic estimates of movement patterns for permanent migration and multilocal living.

While these steps mitigate some of the biases inherent in social media data, it is important to emphasize that Twitter-based estimates primarily reflect digitally active and mobile populations. Consequently, they should be interpreted as relative indicators of mobility patterns, rather than as precise counts of all movements. To further assess the validity of our approach, we compared the estimated totals of permanent migration at the country level with official statistics from Eurostat. Although the absolute counts of movements derived from Twitter were not directly comparable, we observed a high correlation between our estimates and Eurostat values, indicating that the data reliably capture relative differences across countries (see Supplementary Materials – Section 3).

For comparability of mobility flows, we produced an origin–destination matrix connecting all NUTS2 regions for each mobility type, using the centroids of each region to create lines representing the flow between them. To enhance the readability of the visualizations, we applied an edge-bundling technique to the raw line data to cluster lines with similar directions into bundled paths¹¹⁶. For this, we used pre-processed edge-bundled layers developed for NUTS2 regions across Europe^117,118.

Regional characteristics data

We use regional characteristics relevant for different mobility types based on the literature (discussed in the “Results” section) that we grouped into seven themes: social fabric, living conditions, economy and labor market, access and connectivity, digitalization, landscape and environment, and regional typology. The data at the NUTS2 regional level were primarily collected from open Eurostat and European Spatial Planning Observation Network (ESPON) databases, with additional sources alongside the preprocessing steps, and maps of these data from 2012 to 2022 are provided in Supplementary Materials – Section 5.

Defining mobility types

We classified and examined seven distinct types of mobility, encompassing permanent, short-term, and circular mobility forms.

Permanent migration

Permanent migration was identified using Twitter data by analyzing changes in users’ primary residential locations. For each user, significant locations at the NUTS3 level were determined annually. A location was considered significant if it ranked in the top 75th percentile for tweet frequency and if the user had spent at least 30 days there within the year. The location with the longest dwell time was designated as the user’s primary residence. A migration event was recorded when a user’s primary residence changed within two years, classifying it as a permanent migration from the origin to the destination. For consistency with other datasets, the data was aggregated from NUTS3 to NUTS2 levels.

Long-term student mobility

Long-term student mobility data was extracted from the Erasmus dataset. We filtered the dataset to include only student movements lasting longer than 11 months, distinguishing long-term mobility from shorter exchanges.

Seasonal work mobility

Seasonal work mobility was identified using LFS data. Respondents were classified as seasonal workers based on two criteria: explicit self-identification as seasonal workers (using variables such as JATTACH and SEEKWORK) or having a short-term contract (3–11 months in duration, defined by TEMP and TEMPDUR variables). Workers were included in the dataset if they worked in a region (REGION_W) different from their region of residence (REGION).

Short-term student mobility

Short-term student mobility, also derived from the Erasmus dataset, represents individual student exchanges lasting between 3 and 11 months. This filtering ensures a clear distinction between short-term mobility (lasting one or two semesters) and longer-term exchanges.

Long-distance commuting

Long-distance commuting was identified by filtering LFS respondents whose region of residence (REGION) differed from their region of work (REGION_W). To capture long-distance commuters, only those whose commute exceeded 100 kilometers (Euclidean distance between NUTS2 region centroids) were included. However, the varying sizes of NUTS2 regions make accurately capturing long-distance commuting a challenge. Large regions, such as Northern and Eastern Finland, cover vast areas, while smaller regions, such as Inner London, cover much less, leading to potential overrepresentation of countries with smaller NUTS2 regions.

Cross-border commuting

Cross-border commuting is a subset of long-distance commuting, where the region of work is located in a different country (COUNTRY_W) than the region of residence (COUNTRY). Unlike long-distance commuting, no minimum distance filter was applied; any commute between two regions in different countries was considered cross-border commuting.

Multilocal living

Multilocal living was identified by tracking significant locations for each Twitter user at the NUTS3 level. Here, a location was deemed significant if it fell in the top 75th percentile for tweet frequency and if the user spent at least 90 days there within a five-year window (two years before and after the given year). To ensure that identified locations reflect meaningful residential ties rather than incidental visits, we additionally required that a location appear as significant in more than two years within any five-year window.

We then examined movement patterns between each user’s significant locations to distinguish multilocal living from long-distance commuting. To further reduce the likelihood that short-range local trips or repeated visits to non-residential places were captured, we only considered movements between significant locations that were at least 100 km apart. Specifically, for each unordered pair of significant locations involving the primary residence, we calculated the share of movements represented by that pair. If a single pair accounted for more than 50% of all movements (p_max > 0.5), we classified this relationship as commuting, reflecting frequent back-and-forth mobility typical of work-related travel. To evaluate the robustness of this classification rule, we tested a range of alternative p_max thresholds and documented the resulting variations in Supplementary Materials – Section 3. By contrast, if no pair exceeded this threshold, the movements were classified as multilocal living, consistent with less frequent travel between multiple residential bases.

Locations were classified as primary or secondary based on dwell time, but this classification is approximate due to the temporal limitations of Twitter data. The final dataset aggregated these movements from NUTS3 to NUTS2 levels. This approach allowed us to identify multiple places of residence within the same temporal frame (one year), in contrast to permanent migration, which was defined by a change in primary residence across years.

Statistical analysis

To examine the association between regional characteristics and regional mobility flows, we employ a Poisson pseudo-maximum likelihood (PPML) regression model. This approach is particularly well-suited for modeling count data and addresses common issues such as the presence of zero flows and heteroskedasticity¹¹⁹. To account for multilateral resistance to migration and time-related shocks, we estimate five PPML specifications, progressively refining fixed effects to improve estimation accuracy.

The first specification includes time fixed effects ($D_t$) to account for common disturbances across all regions in each period (Fig. 6a), while using origin and destination characteristics ($X_{jkt}$) as predictors (Eq. 1)¹²⁰. However, this model does not fully address multilateral resistance to migration, defined as the influence exerted by alternative destination countries on bilateral migration flows¹²¹ (Fig. 6b), potentially leading to biased estimates. To account for multilateral resistance, the second specification incorporates monadic origin and destination fixed effects ($D_j$, $D_k$) along with time fixed effects, controlling for unobserved time-invariant characteristics of each region (Eq. 2)¹²². The third specification replaces monadic effects with dyadic origin–destination fixed effects ($D_{jk}$) while keeping time fixed effects, capturing persistent bilateral relationships (Fig. 6c), such as historical migration networks and geographic constraints (Eq. 3)¹²³. The fourth specification employs dyadic origin–time fixed effects ($D_{jt}$) alongside destination fixed effects ($D_k$), fully capturing push factors (Fig. 6d) while accounting for multilateral resistance due to migration preferences at the origin (Eq. 4)¹²³. The fifth specification applies dyadic destination–time fixed effects ($D_{kt}$) along with origin fixed effects ($D_j$), controlling for all pull factors (Fig. 6e) and time-varying destination heterogeneity (Eq. 5)¹²⁴.

$$m_{jkt}=\alpha +D_t+{\beta }_1{X}_{jkt}+e_{jkt}$$ 1

$$m_{jkt}=D_j+D_k+D_t+{\beta }_1{X}_{jkt}+e_{jkt}$$ 2

$$m_{jkt}=D_{jk}+D_t+{\beta }_1{X}_{jkt}+e_{jkt}$$ 3

$$m_{jkt}=\alpha +D_{jt}+D_k+{\beta }_1{X}_{kt}+e_{jkt}$$ 4

$$m_{jkt}=\alpha +D_{kt}+D_j+{\beta }_1{X}_{jt}+e_{jkt}$$ 5

where $m_{jkt}$ represents the number of migrants from NUTS2 region j to NUTS2 region k at time t, α is the intercept capturing baseline migration levels, and ${\beta }_1$ is the coefficient vector representing the effects of the explanatory variables on migration flows. The term $e_{jkt}$ is the error component accounting for unobserved factors that may influence migration beyond the specified independent variables.

Fig. 6. Conceptual overview of fixed effects and structural controls in the mobility models.

Each panel illustrates key model components: a time shocks, b multilateral resistance via competing regions, c persistent bilateral relationships, d origin-based push factors, and e destination-based pull factors. Note that these depict the effects controlled in the models rather than their explicit presentation; for example, while multilateral resistance is shown in (b), it is also addressed in models 3, 4, and 5 via various fixed effect structures.

Prior to estimation, we standardize the independent variables. This normalization not only enhances interpretability by placing all predictors on a common scale but also mitigates potential numerical issues during model estimation. Model diagnostics are performed using three metrics, including Wald Chi², log-pseudolikelihood, and pseudo R². We ran the statistical analysis using Stata MP 18 software¹²⁵.

Our independent variables encompass a wide array of regional characteristics, grouped into several categories for analysis (Table 2). In addition to the variables in Table 2, for each movement, we included a variable indicating whether the movement is international or not. Multicollinearity diagnostics showed that all variables had variance inflation factor (VIF) values below 5, except two (population density and transport infrastructure), which remained under 10 (Supplementary Materials – Section 6), ensuring that multicollinearity is not a substantial concern in our models. The transport infrastructure variable, however, was time-invariant due to data limitations and was therefore dropped in Models 2–5, where region fixed effects absorb such constant variation. For the same reason, the area variable was also excluded from these models. In addition, in Model 3, which includes dyadic origin–destination fixed effects, the variables international move and distance were omitted, as they are time-invariant within each regional pair. Prior to standardization, the distance between regions was log-transformed using the natural logarithm to further enhance model robustness and interpretability. In these models, we use the characteristics of destination and origin regions, or primary and secondary residences in the case of multilocal living, as predictors, and the dependent variable is the movement count between regions. For a more detailed description of each factor and further data availability information, please refer to Supplementary Materials – Section 5.

Table 2.

Summary statistics of independent variables

Factor	min	Mean	Median	max	SD
Digitalization
Mobile speed (Mb/s)	4.75	10.76	10.03	31.99	3.23
High-tech employment (%)	0.20	3.64	3.25	13.55	2.16
Economy and labor market
GDP per capita (1000 €)	2.92	27.51	26.48	214.75	18.62
Employment rate (15–64)	28.30	66.10	67.94	86.50	10.22
Weekly work hours	28.55	37.85	37.20	55.05	3.30
Landscape and environment
Max annual temp (°C)	−4.10	15.44	14.96	30.75	4.16
Annual precipitation (mm)	1.04	6.94	6.37	28.84	2.72
Living conditions
Population density (/1000 km²)	0.00	0.41	0.12	11.52	1.13
Life expectancy (<1 yr)	69.70	80.40	81.10	85.80	2.71
Police crime rate (robbery and theft per million)	2.87	129.02	112.42	753.61	100.20
Regional typology
Urban population (%)	0.20	0.65	0.64	0.99	0.18
Mountainous population (%)	0.00	0.16	0.00	1.0	0.31
Coastal population (%)	0.00	0.05	0.00	1.0	0.15
Access and connectivity
Tourist arrivals accommodations (in million)	0.00	3.20	2.09	40.46	3.80
Transport infrastructure (/1000 km²)	0.00	436.31	279.18	6809.74	654.01
Social fabric
Median age	17.60	41.88	42.60	52.10	4.92
Women per 100 men	88.60	103.67	103.50	119.60	3.82
Students in tertiary (20–64) (per 1000 workers)	0.05	1.03	0.85	25.68	1.40
Spatial
Distance between NUTS2 regions (log km)	2.80	1406.91	1267.37	5924.00	846.24
Area (100k km²)	0.00	0.18	0.11	2.27	0.22

A sensitivity analysis was performed separately for NUTS2 regions within the EU-15, meaning the EU member countries prior to the 2004 enlargement. The results of these analyses are available in Supplementary Materials – Section 7.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

M.M.: Conceptualization, Methodology, Data curation, Formal analysis, Software, Visualization, Investigation, Writing – original draft, Writing – review & editing. T.V.: Conceptualization, Methodology, Data curation, Software, Visualization, Writing – review & editing. A.P.: Writing – original draft, Writing – review & editing, Project administration, Funding acquisition. V.R.: Formal analysis, Writing – review & editing, Funding acquisition. O.J.: Conceptualization, Supervision, Data curation, Writing – review & editing, Project administration, Funding acquisition.

Peer review

Peer review information

Nature Communications thanks Alexis Gumy, Ana María López-Villuendas, Junjun Yin, and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

The full harmonized dataset, including all mobility types and regional characteristics, is hosted on Zenodo under the MOBI-TWIN repository accessible at 10.5281/zenodo.14228376¹²⁶. While the dataset at the time of publication is under project-level embargo, the Zenodo link is permanent, and the data will become publicly accessible once the embargo period concludes. Researchers interested in accessing the data before the embargo is lifted may contact the corresponding author for further information or collaboration inquiries. The student mobility datasets at both the NUTS3 and NUTS2 levels are available through the study of Väisänen et al.¹¹¹

Competing interests

The authors declare no competing interests

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-73288-6.