| Since the reform and opening-up policy,the research on migration has gained much attention from researchers at home and abroad.Simultaneously,research results have been gained in terms of theoretical and empirical respectives.However,most studies are based on the traditional gravity model,which stresses the characteristics of origin(Origin,O)factors,destination(Destination,D)factors and distance factors.It fails to explicitly capture the spatial dependence relationship and express the feedback and spillover effects among multi-regions.Firstly,this paper investigates the network autocorrelation effects of Chinese inter-provincial migration flows over the 2005-2010 period,including network structure and effects analysis.The purpose of network structure analysis is to study the important nodes from the viewpoint of the migration network.On the basis of the network weight matrix,network effects analysis is to investigate the existence of the network autocorrelation effects among flows from global and node perspectives.Secondly,the spillover effect models considering the network effects are specified and estimated using maximum likelihood method.Monte Carlo simulation method is applied to estimate the spillover effects of model variables,including the total effects,direct and indirect effects.The effect mechanism of migration network system is analyzed according to spillover effects estimation results.Finally,the Akaike Information Criterion(AIC),the Likelihood Ratio(LR)test and Monte Carlo simulation are used to optimize model analysis,and evaluate the applicability of the models.The results show that:(1)There is a significant network dependence effects between inter-provincial migration flows in China.The values of Moran's I with the corresponding network weight matrix Wo,Wd and Ww are used to measure network autocorrelation degree.The estimates are positive and statistically significant.It shows that the origins of migration flows arrived at the same destination(D),are spatially agglomerated,and so are the destinations of migration flows departed from the same origin(O).In addition,population migration flows surrounding the same origins and destinations are mutually dependent.(2)In terms of seven different spatial OD models,the indirect effects of most models variables are significantly different from zeros.Indirect effects reflect the network effects of migration flows,characterizing a change of any given explanatory variable in a single region will affect migration flows of surrounding regions.Simultaneously,the direct effects are much stronger than indirect effects.It means that the impacts from a variable change of some regions on its own migration flows are larger than impacts on migration flows of surrounding areas.This is due to that these spillover effects take on a decreasing trend with the increase of the order of network weight matrix.(3)Spillover effects model without constraint conditions is the best.It not only considers every regional role as a desitination and origin,but also the impacts from the surrounding areas.This model is suitable for the research of the entire migration networking mechanism in China.Double and single constraint condition models are more focused on regional population migration mechanism analysis.(4)The Monte Carlo simulation results show that the specification of spillover effects model is proper and effective,especially for the specification of network weight matrix.It is assumed that when prior knowledge of the specific data generation process of migration flows is unknown,the "real" spatial spillover effects models are identified in the context of different network weight matrices and different degrees of network autocorrelation. |
PDF Full Text Request