地理学报

地理学报 ›› 2021, Vol. 76 ›› Issue (12): 2964-2977.doi: 10.11821/dlxb202112007

• 人口与城市研究 • 上一篇    下一篇

不确定性视角下的中国省际人口迁移机制分析

蒲英霞1,2,3(), 武振伟1, 葛莹4, 孔繁花1   

  1. 1.南京大学地理与海洋科学学院,南京 210023
    2.江苏省地理信息技术重点实验室,南京210023
    3.江苏省地理信息资源开发与利用协同创新中心,南京 210023
    4.河海大学地球科学与工程学院,南京 211100
  • 收稿日期:2020-10-23 修回日期:2021-05-20 出版日期:2021-12-25 发布日期:2022-02-25
  • 作者简介:蒲英霞(1972-), 女, 山东日照人, 博士, 副教授, 主要从事GIS与空间数据分析集成、区域人口迁移建模与复杂地理计算。E-mail: yingxiapu@nju.edu.cn
  • 基金资助:
    国家自然科学基金项目(41771417);国家自然科学基金项目(41771029);江苏高校优势学科建设工程;江苏省地理信息资源开发与利用协同创新中心资助项目

Analyzing the spatial mechanism of interprovincial migration in China under uncertainty

PU Yingxia1,2,3(), WU Zhenwei1, GE Ying4, KONG Fanhua1   

  1. 1. School of Geography and Ocean Science, Nanjing University, Nanjing 210023, China
    2. Jiangsu Provincial Key Laboratory of Geographic Information Science and Technology, Nanjing 210023, China
    3. Jiangsu Center for Collaborative Innovation in Geographical Information Resource Development and Application, Nanjing 210023, China
    4. School of Earth Science and Engineering, Hohai University, Nanjing 211100, China
  • Received:2020-10-23 Revised:2021-05-20 Published:2021-12-25 Online:2022-02-25
  • Supported by:
    National Natural Science Foundation of China(41771417);National Natural Science Foundation of China(41771029);Priority Academic Program Development of Jiangsu Higher Education Institutions;Jiangsu Center for Collaborative Innovation in Geographical Information Resource Development and Application

RichHTML

PDF (PC)

赞

摘要：

人口迁移过程具有内在的不确定性。贝叶斯模型平均方法（BMA）为不确定性问题提供了行之有效的解决方案。然而,当前该方法多用于线性回归模型在变量选择时出现的模型不确定性问题,很少用于空间建模。本文以2010—2015年中国省际人口迁移流为例,将BMA方法应用于空间OD模型,在考虑网络空间结构的基础上选取迁出地和迁入地各7个解释变量及距离因素,利用马尔可夫链—蒙特卡罗模型综合方法（MC3）进行模型抽样,以后验模型概率为权重计算相应变量的迁出地、迁入地和网络效应等,定量分析不确定性背景下省际人口迁移影响因素和空间机制。结果表明：① BMA模型估计结果更为稳健可靠。与单一模型相比,BMA中变量效应估计的90%可信区间明显缩小,不确定性程度显著降低,结果更为精确;② 区域经济社会发展对省际迁移至关重要。经模型空间抽样后,迁出地人口规模和GDP、迁入地教育水平和迁移存量等的变量后验包含概率大于90%;③ 网络效应在省际迁移过程中不可忽视。所有变量的网络效应占总体效应的40%以上,其中工资、城镇化率、教育和迁移存量等的网络效应（绝对值）大于各自的迁出地和迁入地效应;④ 若不考虑迁移建模中的不确定性,绝大多数区域经济社会变量对省际迁移的影响会被高估。

关键词: 人口迁移, 空间OD模型, 贝叶斯模型平均, 后验模型概率, 变量后验包含概率, 网络效应, 中国

Abstract:

Population migration process has an innate uncertainty with the increasing complexity of regional socioeconomic development. Bayesian model averaging (BMA) provides a feasible solution to the uncertainty of linear regression models. However, model uncertainties are seldom considered in spatial modeling. To reduce the uncertainties in migration modeling, this paper incorporates BMA approaches with spatial origin-destination (OD) models to quantify the spillover mechanism of interprovincial migration in China, 2010-2015. Specifically, we specified network dependence for migration flows and selected origin/destination's population size, gross domestic product (GDP), real wage, urbanization rate, the number of beds in health facilities per 1000 persons, the number of people over college level per 100000 persons in 2010 and migration stocks between 2005 and 2010 as well as railway travel time between provincial capitals in 2010 as explanatory variables. Among 2615 unique models based on 300000 samples using Markov chain Monte Carlo model combination (MC3), 58 models with posterior probability greater than 0.1% were chosen to estimate explanatory variables' origin effects, destination effects, and network effects. Some findings are as follows: (1) BMA model estimates are more robust and reliable. Compared with results from the single spatial origin-destination (OD) model, the widths of 90% credible interval of different explanatory variables are markedly reduced, indicating the degree of model uncertainty has been greatly decreased. (2) Regional population size, quality, and migration stocks have a significant influence on interprovincial migration processes. After model sampling, the inclusion probabilities of population size and GDP at origins, education level and migration stocks at destinations as well as distance variable are beyond 90%. (3) Network effects of most variables are statistically significant, accounting for more than 40% of their corresponding total effects. Moreover, the spillover effects of real wage, education level, and migration stocks are even greater than their corresponding origin and destination effects. (4) The impacts of most explanatory variables on interprovincial migration would be overestimated without considering uncertainties in modeling migration processes.

Key words: population migration, spatial OD models, Bayesian model averaging (BMA), posterior model probability, posterior probability of variable inclusion, network effects, China