Spatial Autocorrelation Analysis Of The Patterns And Mechanism Of Interprovincial Migration Flows In China

Migration flows exhibit some spatial patterns in the origin and destination places. However, bunches of existed research on Chinese migration overlook this phenomenon, by employing interprovincial migration data from2000National Census and some social and economic data in Chinese Statistical Yearbook, we attempt to explore the spatial patterns of inter-provincial migration flows between1995and2000and explain the mechanisms behind the observed patterns. In this manner, we aim to draw attention on the spatial autocorrelation in Chinese migration system.In the section of analysis on spatial mode, both traditional and specialized statistical are employed to explore the spatial pattern in1995-2000Chinese intermigration system, and furthermore qualifies the level of spatial autocorrelation as well as identify the local mode. First, statistics of in-migration and outmigration are computed, in the way of comparing these statistics for each province, the general trend and spatial plot are under revealing. By this mean, we find that1995-2000Chinese migration system present significantly spatial clusters, such as the vast region of central China was the major origin and however, the eastern China absorbed the majority of migrants. Then, we use global spatial autocorrelation statistics of Moran’s I to investigate the spatial autocorrelation in the origin and destination places. The spatial autocorrelation analysis of both in-migration and outmigration flow from some specified province is conducted and iterated one by one at the province level, and the result comes that1995-2000Chinese migration system bears origin-based and destination-based spatial autocorrelation. Finally, statistics of Gij are conducts to identify the local plot of spatial autocorrelation for both in-migration and outmigration. Five provinces are found existed local high-high spatial autocorrelation in their in-migration systems, and nine provinces present local high-valued spatial clusters in their out-migration system.In the section of analysis on migration mechanisms, we construct the traditional gravity and spatial O-D models to characterize the spatial spillover effects among migration flows and make a comparison between the traditional gravity and spatial O-D models. The empirical results indicate that (1) the distance decay effect in gravity model is statistically significant at the significance level of1%; population, FDI and education level in the destination places have expected positive sign in the estimated model and are statistically significant at the significance level of1%; population, landuse, sex rate in the origin places do have expected positive sign and are statistically significant at the significance level of1%.(2) Based on the spatial analysis of outmigration and immigration flows, we construct the spatial O-D model of lagged dependent variables to fit the spatial interaction among migration flows. In comparison with the traditional gravity model, the goodness-of-fit of the spatial O-D model is much better, which has smaller AIC value. At the same time, the estimated parameters of origin and destination places are also smaller than those of traditional gravity model, which indicates that the influences of origin and destination places may be exaggerated in the traditional analysis.In this paper, while we model the spatial dependence of migration flow by spatial lagged dependent variable, further research is needed to investigate the specification of spatial lagged independent and spatial autoregressive error models.