Mechanism Analysis Of Interprovincial Migration Flows In China Based On Bayesian Model Averaging Methods

Since the reform and opening-up policy,the process of urbanization in China has been promoted rapidly,and the migration flows have been increasingly active.The research on the migration has been greatly expanded in breadth and depth,which has experienced the development changes from qualitative to quantitative.However,most studies have not paid attention to the effect of network autocorrelation by focusing on visual factors such as economy,society and culture and the spatial dependence in the network of migration flows.Even if some scholars introduce the spatial dependence of migration flows to the model,they often explain the model directly from the perspective of parameter estimation and do not pay much attention to the influence of network autocorrelation effect.In this paper,Chinese interprovincial migration flow in 1985-2015 years is taken as the object of study.Firstly,three types of network weight matrices are constructed to characterize the network adjacency relationships between migration flows,and the Moran's I index is used to analyze network autocorrelation and migration patterns.Secondly,besides traditional socio-economic explanatory variables,this paper introduces network autocorrelation effect to the traditional migration model and constructs Chinese interprovincial migration spatial OD model to capture spatial dependence in migration flows network.Thirdly,considering the variable uncertainty in the modeling process,this paper applies Bayesian model averaging method to the spatial OD model,and uses the posterior probability obtained by model space sampling to make the weighted average of the results of alternative models to obtain more robust results.Forthly,the effects estimation method is introduced to explain the model from the perspective of space spillover effects.Finally,this paper analyzes the mechanism of Chinese interprovincial migration flows in 1985-2015 years in terms of destination effect,origin effect,spatial network effect and total effect.The preliminary empirical results are as follows:(1)The mechanism of Chinese interprovincial migration has undergone profound changes in different periods.Firstly,from the perspective of posterior probability,the posterior probabilities of the same variable have changed at different stages.Some variables with larger posterior probabilities are smaller in other periods,such as wage variable.Some variables with smaller posterior probabilities are larger in other periods,such as urbanization variable.Secondly,from the effects of explanatory variables,the effects have changed significantly in different periods.For example,the origin-based effects of wage variable weakened in 1985-2005 and strengthened slowly in 2005-2015,and the origin-destination-based network effects of urbanization variable steadily increase in different periods.(2)The spatial OD model considering spillover effects better explains the machanism of Chinese interprovincial migration than the traditional gravity model.Firstly,the spatial spillover effects of some variables are very significant in different peroids,which means that the change of the variable not only affects the migration flows from or to the region,but also indirectly influences the migration flows from or to the surrounding regions through the spatial adjacency relationship,such as population variables.Secondly,from the perspective of spillover effects intensity,the spillover effects of most explanatory variables account for more than 25%of the total effects overall(only wage variable in 1995-2000 and GDP variable in 2000-2005 are exceptions,11%and 20%respectively).If the spillover effects are ignored,the influence of explanatory variables on Chinese interprovincial migration may be underestimated.More importantly,the traditional gravity model ignoring the results of spatial dependence can not explain the spatial mechanism of Chinese interprovincial migration.(3)The spatial model optimization and effects estimation method based on BMA provide a good tool for mechanism analysis of Chinese interprovincial migration.The traditional model comparison and optimization methods are based on the measurement of model fitting degree,select a single "optimal model”through certian evaluation criteria(such as R^2,AIC,likelihood ratio test,etc.),which greatly underestimates the uncertainty in the process of model optimization and causes the limitation of population migration prediction and related decision.The Bayesian model averaging method,based on the model uncertainty,uses the Bayesian statistics principle to calculate the posterior probability as the weight to obtain a weighted average model to obtain the effects estimation of the variables.The weighted average model based on posterior probability covers all possible models with large probability in the whole model space and it is more inclusive than a single model in the process of model construction.(4)Chinese interprovincial migration flows in 1985-2015 years show significant origin-,destination-,and origin-destination-based network autocorrelation effects.Overall,the origin-based network autocorrelation effects are the strongest,and the destination-based network autocorrelation effects are the weakest.The origin-,origin-destination-based network and total effects of population remained at a relatively high level for a long time,only the destination-based effects are relatively weak,reflecting that population pressure is always an important factor influencing people's migration decision.The origin-,destination-,origin-destination-based network and total effects of economic factor are in the forefront for a long time.Most noteworthy,the intensity of actual wage rate has weakened and that of GDP gradually strengthened.The migration people no longer only consider the wage level,but also pay more attention to the overall economic development and even its future development trend and potential.The origin-,destination-,origin-destination-based network and total effects of urbanization rate increased gradually,which reflects that people pay more and more attention to the development of urbanization.All effects of proportion of college students are relatively weak.All effects of migration stock are relatively moderate and increased gradually.The long-term and profound impact of social network on Chinese interprovincial migration is permeated with a wave of decades of migration,related to every family seeking a better life in a strange city.