Different roads take me home: the nonlinear relationship between distance and flows during China's Spring Festival

Human mobility patterns have garnered significant academic attention, with research employing physics-based modeling and social science explanations. However, the nonlinear relationship between flows and distance in intercity mobility, and regional variations in this relationship, remain poorly understood. China's Spring Festival, characterized by massive long-distance travel, provides a unique context to explore this. This study utilizes Tencent's big data to analyze intercity mobility during the Spring Festival, employing a GBDT model to uncover the nonlinear dynamics. The research aims to understand the uneven development context behind China's mobility patterns by introducing nonlinear relationships and inter-provincial differences. The study hypothesizes three distinct distance-mobility intensity relationships: Plateau (minimal distance decay within a province), Drop (sharp decline at provincial borders), and Rebound (revival beyond a certain distance due to more developed provinces).

Existing research on China's internal migration highlights the role of rapid urbanization, particularly in coastal regions, driven by industrial development and economic opportunities. National policies, especially concerning provincial capitals, also shape migration patterns. High housing costs and the hukou (household registration) system create significant barriers for migrants seeking urban settlement. The Spring Festival migration emphasizes these urban-rural disparities and the challenges they pose to China's social fabric. Previous studies utilizing survey data and limited granular data from sources like national censuses and the China Migrants Dynamic Survey provide insights into mobility, but big data offers more nuanced analysis. Physics-based models, including network analysis, explore universal laws, while social science approaches focus on socioeconomic factors. Explanatory machine learning offers a combined approach, with the GBDT method particularly promising for understanding nonlinear relationships.

This research uses Tencent's Migration Dataset, a geolocation data source derived from smart device users, to analyze inter-city population movements. Data from 2018, encompassing approximately 40,289 daily records, were cleaned and aggregated to 20,155 prefectural-level city population flow records. The GBDT model was chosen for its ability to handle nonlinear relationships and large datasets. The dependent variable is inter-city connection strength (Flow), and the key independent variable is the distance between city centers. Control variables include city population, GDP, public service levels (fiscal, technological, educational expenditures), environmental quality (air quality, PM2.5), and industrial structure. Logarithmic transformations were applied to mitigate data skewness. Hyperparameter tuning was performed using RandomizedSearchCV to optimize the GBDT model. Partial dependence plots (PDPs) were used to visualize the nonlinear relationships between distance and mobility intensity, allowing for a detailed analysis of how distance affects population flows.

Analysis of the 2019 Spring Festival migration reveals significant population outflows from eastern coastal cities (like Beijing, Shenzhen, Guangzhou, Shanghai) and inflows into central and western provinces. Visualizations show a pronounced movement from east to west, mirroring broader economic trends. Provincial-level analysis shows variations in migration patterns: Guangdong exhibits significant intra-provincial movement, while Hubei shows less outbound but considerable eastward migration. Sichuan maintains consistent eastward flows, and Heilongjiang exhibits substantial return migration. The GBDT model shows geographical distance as the most significant factor (57.26%) influencing inter-city connection strength. PDPs reveal three types of nonlinear relationships between distance and connection strength across provinces: Plateau (high initial values followed by a plateau in eastern provinces), Drop (sharp decline at provincial borders, especially in central and northeastern provinces), and Rebound (increase in connection strength beyond a certain distance threshold, primarily in western provinces). Spatial visualization reveals Guangdong's extensive inter-provincial connectivity, contrasted with Hubei's stronger internal connectivity and limited external influence. These findings highlight the influence of economic development levels and geographic locations on population mobility.

The findings address the research question by demonstrating the nonlinear relationship between distance and intercity mobility during China's Spring Festival and highlighting significant regional variations. The three identified patterns (plateau, drop, rebound) reflect the complex interplay of economic development, geographic location, administrative boundaries (and the hukou system), and infrastructure. The contrasting patterns observed in Guangdong (extensive national influence) and Hubei (strong regional influence) underscore the importance of considering regional economic strategies and development levels when analyzing population mobility. These results offer insights into China's urban-rural dynamics and highlight the ongoing separation of employment and settlement, especially concerning the hukou system.

This study identifies three distinct nonlinear relationships between distance and intercity mobility during China's Spring Festival: plateau, drop, and rebound. Regional differences highlight the uneven development and administrative influences on mobility. The findings emphasize the need for policies promoting equal access to public services and balanced regional development to mitigate the impacts of the hukou system and address the challenges of seasonal migration. Future research could explore the evolutionary patterns of these relationships over time and conduct international comparisons to identify universal laws governing distance-flow dynamics.

The study's limitations include the use of aggregated intercity data from Tencent, preventing the separation of tourists from those returning home. The lack of long-term historical data restricts the analysis of the evolution of mobility patterns. International comparative studies are necessary to establish the universality of the identified nonlinear relationships.