Acta Geographica Sinica ›› 2019, Vol. 74 ›› Issue (2): 222-237.doi: 10.11821/dlxb201902002

Special Issue: 人口与城市研究

• Population and Urban Studies • Previous Articles     Next Articles

Driving mechanism of interprovincial population migration flows in China based on spatial filtering

GU Hengyu1(),SHEN Tiyan1(),LIU Ziliang2,MENG Xin1   

  1. 1. School of Government, Peking University, Beijing 100871, China
    2. School of Economics and Management, South China Normal University, Guangzhou 510006, China
  • Received:2017-11-09 Online:2019-02-25 Published:2019-01-29
  • Supported by:
    National Social Science Foundation of China, No.17ZDA055;National Natural Science Foundation of China, No.71473008, No.71733001


According to previous studies, not only does the conditional gravity model based on ordinary least squares often bring about poor fitting of migration flows in reality, but also there exists overdispersion in the extended Poisson gravity model. Simultaneously, network autocorrelation usually exists in population migration data (e.g., the spatial interaction among migration flows). The problems mentioned above result in biased estimation. In order to capture network autocorrelation and deal with the issue of overdispersion, we build an eigenvector spatial filtering negative binomial gravity model (ESF NBGM) based on the data of 1% national population sample survey in 2015, to analyze the driving mechanism of interprovincial population migration flows in China. The results are as follows: (1) Positive spatial spillover effect exists in interprovincial population migration flows, and ESF can capture network autocorrelation in data, so as to reduce the estimated deviation of the model. Furthermore, eigenvectors ranking top 1.4% can properly interpret the spatial pattern of high network autocorrelation in data. (2) There exists overdispersion in China's interprovincial migration flows. Considering this problem, a negative binomial regression model is more suitable for the estimation of driving mechanism for population migration, together with statistical enhancement. (3) Network autocorrelation leads to overestimation of distance variables and underestimation of non-distance variables. The results of the improved model reveal that: chief factors the affect driving mechanism are regional population characters, social network, economic development and education level. Meanwhile, living environment and road network gradually become one of the most crucial pulling factors that influence migration flows. (4) Compared to previous studies, social network (i.e. migration stock) plays a more significant role in population migration flows, while the impact of spatial distance keeps weakening.

Key words: interprovincial population migration flows, spatial filtering, negative binomial gravity model, driving mechanism, China