2017年中国地理信息科学理论与方法学术年会优秀论文专辑

区域人口迁移时空溢出效应与动力机制分析

  • 赵心怡 ,
  • 蒲英霞 , *
展开
  • 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
*通讯作者:蒲英霞(1972-),女,副教授,博士,研究方向为空间数据分析与集成。E-mail:

作者简介:赵心怡(1993-),女,硕士生,研究方向为空间数据挖掘。E-mail:

收稿日期: 2018-01-02

  要求修回日期: 2018-04-04

  网络出版日期: 2018-06-20

基金资助

国家自然科学基金项目(41271388、41771417)

江苏高校优势学科建设工程资助项目

江苏省地理信息资源开发与利用协同创新中心资助项目

Space-time Spillover Effects and Driving Forces of Regional Migration Process

  • ZHAO Xinyi ,
  • PU Yingxia , *
Expand
  • 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
*Corresponding author: PU Yingxia, E-mail:

Received date: 2018-01-02

  Request revised date: 2018-04-04

  Online published: 2018-06-20

Supported by

National Natural Science Foundation of China, No.41271388, 41771417

Priority Academic Program Development of Jiangsu Higher Education Institutions (PAPD)

Jiangsu Center for Collaborative Innovation in Geographical Information Resource Development and Application

Copyright

《地球信息科学学报》编辑部 所有

摘要

区域人口迁移流的规模不仅取决于迁出地与目的地的“双边”要素,也与前期迁移流和周边迁移流息息相关。传统重力模型揭示了区域人口迁移过程的“推-拉”机制,但受制于对时空维度的忽视,无法有效表达迁移流之间的时空依赖关系,因而难以度量区域要素变化对迁移流产生的时空溢出效应。本文引入多种形式的时空依赖结构,构建迁移流时空重力模型,并采用贝叶斯马尔可夫链蒙特卡洛(MCMC)方法进行估计。在此基础上,结合时空效应框架量化区域要素对迁移流的影响,定量分析人口迁移过程的时空溢出效应与动力机制。本文以1985-2015年中国省际人口迁移为例,通过与非空间的动态重力模型估计结果比较,初步表明时间依赖、空间依赖以及时空扩散依赖在区域人口迁移过程中不容忽视;时空维度上,区域要素变化在初期对迁移网络的溢出效应超过对该区域迁移流的直接影响;逐渐衰减的时空溢出效应维持了区域人口迁移规模发展的相对稳定,与动态重力模型估计结果形成了鲜明对比。区域人口规模、人均GDP水平及其时空溢出效应共同驱动中国省际人口迁移系统的发展。耦合时空维度依赖关系的时空重力模型能更好地理解区域人口迁移过程的演化特征,为促进区域人口均衡发展提供科学的决策依据。

本文引用格式

赵心怡 , 蒲英霞 . 区域人口迁移时空溢出效应与动力机制分析[J]. 地球信息科学学报, 2018 , 20(6) : 817 -826 . DOI: 10.12082/dqxxkx.2018.180015

Abstract

Interregional migration is a significant component of regional population growth as well as a major driving force in urbanization process. The evolution of migration flows is not only related to the characteristics of origin and destination regions, but also the past and surrounding migration flows. Most empirical migration studies based on traditional gravity models have failed to capture space-time spillover effects during the migration process due to ignoring time or spatial dependence among migration flows. By introducing several space-time interaction effects, this paper constructed the space-time gravity model of interprovincial migration flows in China over the period of 1985-2015 and estimated the model using Bayesian Markov Chain Monte Carlo (MCMC) method. The space-time spillover effects evaluation framework further explained the space and time dynamics in the evolution of interprovincial migration associated with changes in regional GDP per capita and population size. The preliminary results are as follow: firstly, the estimates of time, spatial and space-time diffusion dependence are all significant, which can provide powerful means for exploring complex and systematic behaviors among regional migration flows. Secondly, regional population size dominates the Chinese interprovincial migration process more than twice the influence of regional GDP per capita. Thirdly, the spillover effects of regional socio-economic factors play a quite significant role during regional migration process, which are greater than the corresponding origin and destination effects in the short term. More importantly, the decaying spillover effects through the whole space-time network will help the migration system stay at an equilibrium state over the long term. All in all, the coupled space-time gravity model contributes to capture the space-time spillover effects and driving forces during the regional migration process, which provides a scientific basis for predicting future migration trends and promoting balanced regional population development.

1 引言

人口迁移作为区域人口变化的重要组成之一,对重塑区域人口布局、促进区域经济社会发展与转型起到日趋关键的作用[1,2]。从系统的角度来看,区域和区域之间的人口迁移流构成一个时空迁移网络系统,其中每个区域可视为迁出地(Origin,O)和目的地(Destination,D)网络节点,而区际迁移流作为网络中的O-D流,不仅受迁出地和目的地的影响,还受过去迁移流和周边迁移流的影响[3,4]。在迁移网络系统中,某区域要素的变化,既影响与本区域节点直接相关的迁出流或迁入流,也会通过时空依赖结构对整个网络系统产生时空溢出效应,引起一系列的连锁反应,共同驱动人口迁移系统的发展与演变。传统的重力模型由于能够较好地表达迁出地和目的地之间的“推-拉”作用和距离衰减效应,被广泛应用于区域人口迁移演变规律与作用机制研究[5,6]。在此基础上,通过进一步对时空依赖结构和依赖关系的表达,量化区域要素变化对迁移流产生的时空溢出效应,从系统的角度评价宏观经济社会变量对人口迁移的影响,可为未来区域人口迁移系统的调控提供一定的借鉴和参考。
随着空间统计学和空间计量经济学的发展,国内外研究人员开始对不同迁移流在空间或时间上的相互影响与相互机制进行探究与表达[7,8,9,10]。Black[11]以美国人口迁移为例,利用全局自相关Moran’I指数证实了迁移流之间存在空间依赖。Griffith等[12]针对重力模型的不合理,利用因变量空间滞后形式表示O-D流之间的空间依赖关系;LeSage等[3]进一步提出了表达多种空间自相关形式的空间O-D模型,包括空间自回归模型(SAR)、空间自相关模型(SAC)、空间杜宾模型(SDM)等;蒲英霞等[13]利用SAR模型对中国省际迁移过程进行模拟,发现迁移流之间存在显著的空间效仿和竞争行为;Fan[14]将表征时间滞后的迁移存量引入到重力模型中,发现迁移存量对人口迁移起到显著的促进作用,揭示了迁移过程中时间依赖的重要性。王桂新等[15]发现改革开放以来,中国不同时期的省际人口迁出规模分布以及人口迁入规模分布之间均具有较强的相关性,相邻时期的相关性达到最大值。上述研究或从空间维度,或从时间维度考察迁移流之间的依赖关系,一定程度上丰富了我们对人口迁移系统的理解。然而,这2种方式仅是从单一的空间维度或时间维度考虑迁移流之间的相互作用,割裂了迁移流在时空上的依赖关系,对整个网络系统中时空溢出效应的理解和认识有待进一步加强。
本文在传统重力模型基础上,同时考虑多种形式的时空依赖(时间依赖、空间依赖和时空扩散依赖)关系,构建时空重力模型,将区域要素的空间效应在时间维度上加以分解,形成二维多向的时空溢出效应评价框架,以更好地探究区域人口迁移过程的时空动力机制。具体以1985-2015年中国省际人口迁移面板数据为例,构建中国省际人口迁移时空重力模型,并采用贝叶斯马尔可夫链蒙特卡洛(Markov Chain Monte Carlo,MCMC)方法对模型系数和效应进行估计。通过与非空间的动态重力模型估计结果比较,进一步考察区域经济社会要素变化如何通过时空依赖结构对中国省际人口迁移规模产生影响,以期对区域人口迁移过程的时空动力机制和系统演变规律做出合理解释。

2 研究方法

2.1 重力模型

传统的重力模型侧重于表达迁出地与目的地的“推-拉”作用以及距离衰减效应对O-D流的影响,具体表达式如下[3,4]
y = α ι N + x o β o + x d β d + γg + ε (1)
式中:yn×n的O-D流矩阵按列排列的N×1(N=n2)列向量;ιN为所有元素均为1的N×1列向量;xoxd)为迁出地(目的地)经济、社会等要素构成的解释向量,假设xn×k的解释变量矩阵,xoxd)可以通过克罗内克积$\otimes$运算生成,即 x o = x ι n ( x d = ι n x );gn×n距离矩阵按列排列的N×1列向量;ε表示均值为0,方差为常数的误差向量;αβoβdγ为待估系数或向量,分别表示常数项、迁出地与目的地因素以及距离因素的系数。
在重力模型中,解释变量的系数可以看作弹性系数,定量解释区域要素变化对O-D流的影响程度。若某变量的βo为正,表明该要素增加会产生更多迁出流;同理,若βd为正,表明该要素增加会吸引更多迁入流[3]。因此,O-D流会同时受到迁出地的“推力”和目的地的“拉力”作用。而距离因素通常对O-D流的产生起到阻碍作用,其系数γ一般为负。

2.2 时空重力模型

在整个网络系统中,任意一条O-D流不仅受迁出地和目的“双边”要素影响[12],还依赖于时空尺度上前期O-D流和周边O-D流的影响,即时空溢出效应。因此,时空重力模型是在传统重力模型的基础上继续引入时间依赖和空间依赖,同时考虑时间和空间交互作用的一类模型。具体地,在时间维度上,通常假设各个时期的观测值(yt)仅受前一时期观测值(yt-1)的影响,而与后续时期无关;在空间维度上,参考LeSage等[3,16]提出的迁出地空间依赖、目的地空间依赖和迁出地-目的地空间依赖来描述O-D流之间的复杂空间依赖关系。在具体建模过程中,通常分别采用Wo,WdWw描述这3种空间关系:Wo=W$\otimes$IN,Wd=IN$\otimes$W,Ww=W$\otimes$W[17,18]。其中,W是对角线元素为0且经过行标准化处理的n×n空间权重矩阵,描述横截面上n个区域之间的空间关系。
结合时间滞后效应以及3种空间依赖关系,对式(1)进行拓展,构建覆盖T个时期的时空重力模型,具体如式(2)所示,其中 η = μ N × T + ε N × T
Y = ρ o ( I T W o ) Y + ρ d ( I T W d ) Y + ρ w ( I T W w ) Y + ϕ Y - 1 + θ o ( I T W o ) Y - 1 + θ d ( I T W d ) Y - 1 + θ w ( I T W w ) Y - 1 + α ι N × T + X o β o + X d β d + γG + η (2)
式中:因变量 Y = ( y 1 ' , , y T ' ) ' y t = ( y 1 t , y Nt ) ' 是第 t时期O-D流矩阵按列排列的N×1列向量。相应地, Y - 1 = ( y 0 ' , , y T - 1 ' ) ' 表示Y的一阶时间滞后项。 ITT维单位矩阵,IT$\otimes$Wo,IT$\otimes$WdIT$\otimes$Ww通过结合时间和空间上的依赖关系,表达T个时期内的时空权重矩阵。(IT$\otimes$WoY,(IT$\otimes$WdY和(IT$\otimes$WwY描述同一横截面上的O-D流在不同空间自相关作用下的多种影响形式;同理,(IT$\otimes$WoY-1,(IT$\otimes$WdY-1和(IT$\otimes$WwY-1是O-D流的一阶时间滞后项在不同空间自相关作用下对当前O-D流的多种影响形式。ιN×T为所有元素均为1的NT×1列向量。 X o = ( x o 1 ' , , x oT ' ) ' Xd= ( x d 1 ' , , x dT ' ) ' T个时期的迁出地(目的地)的解释向量,组织形式类似于Y G = ( g 1 ' , , g T ' ) ' T个时期起讫点间的距离向量。此处时空重力模型中误差项被分解为个体随机效应μN×T和随机误差项εN×T,其中μN×T各个时期上的μN值相同,反映N条O-D流之间不随时间变化只随个体变化的个体异质性,且μi(1≤iN)服从正态分布 N ( 0 , σ μ 2 ) 。同时,假设εN×TμN×T无关,且其中各个误差项εj(1≤jN×T)服从正态分布 N ( 0 , σ ε 2 ) 。此外,(ρo,ρd,ρw),ϕ和(θo,θd,θw)分别表示空间依赖,时间依赖以及时空扩散依赖的程度[19]
对于中国省际迁移过程而言,时空重力模型 可以描述迁移流之间二维多向的复杂依赖关系 (图1)。相较于传统重力模型,该模型不仅考虑了同一横截面上相邻迁移流之间的空间相关性,还 从动态连续的角度考虑各个横截面之间的纵向关系[8,20],包含更丰富的信息,一定程度上提高了模型估计结果的准确性[21]
Fig. 1 The chart for spatial dependence, time dependence and space-time diffusion dependence structure of Chinese interprovincial migration process

图1 中国省际迁移过程的时间依赖、空间依赖以及时空 扩散依赖结构图

当前研究文献大多利用最大似然(ML)或准最大似然(QML)方法对结合空间关系的模型进行估计,在对似然函数求导的过程中往往需要很大的计算量,影响模型估计效率。为了避免多维数值积分产生的复杂计算量,本文采用贝叶斯马尔可夫链蒙特卡洛(MCMC)抽样方法。具体采取基于参数联合后验分布的贝叶斯更新机制实现模型参数的轮流循环抽样和估计[22,23,24]。各待估参数联合后验分布的派生结果如下:
p α , β o , β d , γ , σ ε 2 , σ μ 2 , ρ o , ρ d , ρ w , ϕ , θ o , θ d , θ w | Y f Y | α , β , μ , σ ε 2 , ρ , ϕ , θ p μ | σ μ 2 p ( σ μ 2 ) p ( α ) p ( β ) p ( ρ ) p ( ϕ ) p ( θ ) p ( σ ε 2 ) (3)
式中: β = ( β o ' , β d ' , γ ), ρ = ( ρ o , ρ d , ρ w ) , θ = ( θ o , θ d , θ w ) f Y | . 表示似然函数; p ( . ) 表示各个参数的后验分布。从各参数的条件后验分布中进行抽样,并对抽样序列进行均值以及其他统计指标的计算,最终得到参数估计结果以及显著性条件。
具体的参数先验分布设定为:
α ~ N ( α 0 , M α - 1 ) β ~ N ( β 0 , M β - 1 ) σ ε - 2 ~ G ( v 0 / 2 , S 0 / 2 ) σ μ - 2 ~ G ( v 1 / 2 , S 1 / 2 ) (4)
由于αβ、存在先验不确定性,因此对其设置非信息先验,即设α0=0,β0=0以及 M α - 1 = 10 - 12 , M β - 1 = 10 - 12 I 2 k + 1 。同样,Gamma先验的参数v0, v1, S0S1都设置为0.001。由于α, σ ε - 2 σ u - 2 的条件分布的形式已知,采用Gibbs抽样方法进行抽样。而对于时空依赖参数ρ,ϕθ,规定满足区间(-1, 1)内的均匀分布[25,26],由于该分布不是已知的标准形式,本文选择Metropolis-Hastings(M-H)抽样方法,从均匀分布中进行抽样并伴随调和随机游走过程产生候选值,以完成参数估计。

2.3 时空溢出效应

在时空重力模型中,某区域要素变化产生的时空溢出效应以迁出地效应、目的地效应、网络效应的形式遍布于整个迁移网络系统。其中,迁出地效应(oe)包括2个部分:① 对从该区域出发的所有迁出流产生的直接影响;② 因该区域要素变化而引起其他迁移流变化反过来对该区域迁出流产生的反馈作用。类似地,目的地效应(de)不仅包含该区域要素变化对抵达该区域的所有迁入流产生的直接影响,还包括因该区域要素变化而引起其他迁移流变化反过来对该区域迁入流产生的反馈作用;网络效应(ne)则是对其他区域迁移流受到的时空溢出效应的汇总;总体效应(te)是上述效应之和[4,27]
为简便起见,首先考虑特定横截面上特定区域要素变化对O-D流产生的影响,即同期迁出地效应、目的地效应、网络效应和总体效应。具体地,利用nn×n的偏导矩阵TE记录所有区域的第r个变量的变化在空间交互作用下对所有O-D流产生的影响[20,28]
TE = T E 1 T E 2 T E n = Y 1 / X 1 r Y 2 / X 2 r Y n / X n r = B - 1 J d 1 β d r + J o 1 β o r J d 2 β d r + J o 2 β o r J d n β d r + J o n β o r (5)
式中:B表示空间过滤 ( I n 2 - ρ o W o - ρ d W d - ρ w W w ) Jdi是第i行值为1,其余元素为0的n×n矩阵;类似地,Joi是第i列值为1,其余元素为0的n×n矩阵。因此,n×n矩阵TEi Y i / X i r )的行向量表示对抵达区域i的所有迁入流的影响,列向量表示对从区域 i出发的所有迁出流的影响(1≤ini为整数)。
根据 te = ( 1 / n 2 ) ι n 2 TE ι n 计算整个网络中所有O-D流的累积影响的均值,得到总体效应。
迁出地效应和目的地效应,可以通过类似方式求得: oe = ( 1 / n 2 ) ι n 2 OE ι n , de = ( 1 / n 2 ) ι n 2 DE ι n 。其中,OEDE的定义与TE类似,分别由nOEiDEi构成。
网络效应一般通过nete-oe-de计算得到。在非空间的动态重力模型中,ρo, ρdρw的值均为0,即不存在空间溢出效应。
在此基础上,进一步考虑O-D流的时间溢出效应。具体地,特定区域的第r个变量在第t时期的变化会对第t+T时期的O-D流产生不同程度的影响。此时,TE新的计算方式为:
TE = T E 1 T E 2 T E n = Y 1 ( t + T ) / X 1 ( t ) r Y 2 ( t + T ) / X 2 ( t ) r Y n ( t + T ) / X n ( t ) r = D T J d 1 β d r + J o 1 β o r J d 2 β d r + J o 2 β o r J d n β d r + J o n β o r
D T = ( - 1 ) T ( B - 1 C ) T B - 1 (6)
式中:C表示时空扩散过滤 - ( ϕ I n 2 + θ o W o + θ d W d + θ w W w ) [29]。据此计算该变量在第t+T时期的teoedene
相应地,计算t~t+TT个时期内累积的总体效应时,TE如下:
TE = T E 1 T E 2 T E n = Y 1 ( t ~ t + T ) / X 1 ( t ) r Y 2 ( t ~ t + T ) / X 2 ( t ) r Y n ( t ~ t + T ) / X n ( t ) r = s = 0 T D s J d 1 β d r + J o 1 β o r J d 2 β d r + J o 2 β o r J d n β d r + J o n β o r (7)
T趋近于无穷大时,区域要素变化对O-D流产生趋于稳定的长期效应。此时, s = 0 D s = B + C - 1 ,进而可以计算长期的teoedene

3 中国省际迁移流时空动态解释

为进一步探索改革开放以来我国省际人口迁移过程中复杂的时空动力机制,本文利用全国31个省(市、自治区,不包括港澳台地区)1985-2015年 6个时期(5年为一个时期)的5580条省际迁移流,构建中国省际人口迁移时空重力模型,并利用MCMC方法进行估计。具体数据来自3次全国人口普查(“四普”、“五普”和“六普”)和3次全国1%人口抽样调查(1995、2005和2015年)[30,31,32,33,34,35]。为确保迁移数据的完整性与连续性,重庆市1985-1990年和1990-1995年的缺失数据根据四川省相应时期迁移数据分配得到,西藏地区1985-1990年的缺失数据通过前后时期迁移数据插值得到。在解释变量方面,选取省会城市之间的铁路里程、人均GDP水平和人口数量表征地理、经济以及人口因素对省际迁移的影响[36,37,38]。由于铁路里程受时间影响较小,默认各个时期铁路里程相等。考虑到因变量是某时期内的迁移总量,解释变量数据来自该时期中间年份(如2010-2015年时期为2013年)的《中国统计年鉴》[39,40,41,42,43,44]。其中,人均GDP是以1980年GDP水平作为不变价,消除价格波动影响。此外,模型中的因变量和解释变量都进行了对数处理。
空间权重矩阵W基于空间邻接关系构造,若两省之间存在公共边界,则对应的元素设为1,否则为0。在此基础上进行行标准化生成Wo,WdWw。本文不考虑省内迁移流,故删除矩阵对角线元素,得到新的矩阵: W ˜ o , W ˜ d W ˜ w 。结合选择的具体变量,同时考虑时间、空间和时空扩散依赖关系,构建中国省际人口迁移时空重力模型,具体表达式为:
Y = ρ o ( I T W ˜ o ) Y + ρ d ( I T W ˜ d ) Y + ρ w ( I T W ˜ w ) Y + ϕ Y - 1 + θ o ( I T W ˜ o ) Y - 1 + θ d ( I T W ˜ d ) Y - 1 + θ w ( I T W ˜ w ) Y - 1 + α ι N × T + X o _ GDP β o _ GDP + X o _ Popu β o _ Popu + X d _ GDP β d _ GDP + X d _ Popu β d _ Popu + γG + η (8)
式中:Xo_GDPXo_Popu分别表示迁出地的人均GDP水平和人口数量;Xd_GDPXd_Popu分别表示目的地的人均GDP水平和人口数量;βo_GDP,βo_Popu,βd_GDPβd_Popu为相应变量的系数; η = μ N × T + ε N × T
此外,为进一步考察区域人口迁移过程中空间依赖对迁移流的影响,本文同时构建了非空间的动态重力模型作为基准,具体表达式如下:
Y = ϕ Y - 1 + α ι N × T + X o _ GDP β o _ GDP + X o _ Popu β o _ Popu + X d _ GDP β d _ GDP + X d _ Popu β d _ Popu + γG + η (9)
式中: η = μ N × T + ε N × T

3.1 模型估计

本文在对动态重力模型进行估计时,共进行5000次MCMC抽样,其中前2000次作为预热被舍弃;对于时空重力模型,考虑到待估参数的增加导致模型复杂度提高,将抽样次数增加到35 000次(前25 000次被舍弃),以确保其余抽样最终生成平稳的马尔可夫链。中国省际人口迁移动态重力模型和时空重力模型的系数估计和相应的显著程度详见表1
Tab. 1 Coefficients estimates for dynamic gravity model and space-time gravity model

表1 中国省际迁移动态重力模型和时空重力模型的系数估计结果

动态重力模型 时空重力模型
变量 系数 p 系数 p
α -1.0335 0.0918 -0.6042 0.0395
γ -0.9795 0.0000 -0.2147 0.0000
βo_GDP -0.0665 0.0440 -0.0249 0.0786
βo_Popu 0.8746 0.0000 0.2222 0.0000
βd_GDP 0.7947 0.0000 0.1713 0.0000
βd_Popu 0.5094 0.0000 0.1084 0.0000
ρo - - 0.4381 0.0000
ρd - - 0.3165 0.0000
ρw - - 0.0670 0.0000
ϕ 0.0563 0.090 0 0.5935 0.0000
θo - - -0.2282 0.0000
θd - - -0.1301 0.0000
θw - - -0.2282 0.0000
R2 0.6760 0.7687
修正R2 0.6757 0.7684
AIC -37 679 -39 245
BIC -37 641 -39 207
表1可看出,时空重力模型在数据拟合方面明显更优,其R2 和修正R2高于非空间的动态重力模型且AICBIC也更小。在该模型中,时间依赖系数ϕ显著为正,表明省际迁移在时间维度上存在较强的关联。例如,过去一个时期从湖南到广东的人口迁移会促进当前从湖南到广东的人口迁移,并对未来两省之间的迁移产生潜在影响。同时,迁出地和目的地空间依赖系数 ( ρ o , ρ d ) 也显著为正,表明迁出地之间的空间依赖关系以及目的地之间的空间依赖关系对省际人口迁移起到了重要的作用。例如,从湖南到广东的人口迁移在一定程度上会促进从湖南周边省份(如湖北)到广东的人口迁移(即迁出地空间依赖),也会促进从湖南到广东周边省份(如福建)的人口迁移(即目的地空间依赖)。相比较,基于迁出地-目的地的空间依赖系数ρw数值较小,对迁移流的影响较弱。而显著为负的时空依赖系数 ( θ o , θ d , θ w ) 说明人口迁移过程中存在一定程度的相互制约机制。例如,过去一个时期从湖南到广东的人口迁移会对当前从湖南周边省份(如湖北)到广东的人口迁移起到一定的抑制作用,类似地,也会抑制当前从湖南到广东周边省份(如福建),乃至从湖南周边省份到广东周边省份的人口迁移,并对未来这些省份之间的人口迁移产生潜在的制约作用。
相较于动态重力模型,迁移流之间的时间依赖程度在时空重力模型中得到显著增强。这主要是由于非空间的动态重力模型忽视了迁移流之间的空间依赖,一定程度上放大了解释变量的影响力,导致时间依赖系数估计偏小。事实上,在省际迁移过程中,关于目的地工作机会等对迁移人口产生吸引力的信息主要依靠社会网络(如家庭成员、亲戚、朋友、老乡)获得[8]。即使举家迁移,家庭成员往往也不同步,说明省际人口迁移存在时间上的“路径依赖”。同时,较强的时间依赖关系也进一步验证了中国改革开放以来形成的“由西向东”的宏观区域迁移模式具有一定的稳定性[15]
距离因素与迁移流之间呈显著的负相关,说明长距离会在一定程度上抑制人口迁移活动。在考虑时空交互之后,时空依赖结构起到了类似于距离变量的作用,削弱了距离因素的影响。值得注意的是,由于引入时空依赖关系,变量的系数估计不能直接用来分析区域经济社会要素变化对迁移流的影响[4],需要进一步考虑时空溢出效应,才能更准确全面地考察区域要素对整个迁移网络产生的“连锁反应”,从而对区域人口迁移的动力机制做出更合理的解释。

3.2 效应分析

表2记录了动态重力模型和时空重力模型的迁出地效应、目的地效应、网络效应以及总体效应,全面描述区域要素变化对中国省际迁移影响的时空变化过程。表中标有“0(同期)”的行表示区域要素变化对当前迁移流产生的瞬时响应。对后续时期迁移流产生的影响记录在相应时期的行中。最后一行反映区域要素变化对迁移流的影响趋于稳定时的长期累积结果。表中“均值”列是各时期的效应均值,“累积和”列则是对应时间范围内效应累积之和。
Tab. 2 Estimates of the spillover effects from dynamic gravity and space-time gravity models

表2 中国省际迁移动态重力模型和时空重力模型的溢出效应估计

时间T 动态重力模型 时空重力模型
人均GDP 人口数量 人均GDP 人口数量
均值 累积和 均值 累积和 均值 累积和 均值 累积和
迁出地效应(oe
0(同期) -0.0665** -0.0665** 0.8746*** 0.8746*** -0.0214 -0.0214 0.4111*** 0.4111***
1 -0.0060* -0.0724** 0.0784*** 0.9531*** -0.0288** -0.0502 0.2047*** 0.6158***
2 -0.0006 -0.0730** 0.0073** 0.9604*** -0.0181** -0.0683 0.1277*** 0.7436***
3 -0.0001 -0.0730** 0.0007* 0.9611*** -0.0119** -0.0802 0.0835*** 0.8271***
4 -0.0000 -0.0730** 0.0001 0.9612*** -0.0080** -0.0882 0.0563*** 0.8834***
5 -0.0000 -0.0730** 0.0000 0.9612*** -0.0055** -0.0936 0.0386*** 0.9219***
长期 - -0.0730** - 0.9612*** - -0.1067 - 1.0142***
目的地效应(de
0(同期) 0.7947*** 0.7947*** 0.5094*** 0.5094*** 0.2363*** 0.2363*** 0.1776*** 0.1776***
1 0.0713*** 0.8661*** 0.0456*** 0.5550*** 0.1068*** 0.3431*** 0.0619*** 0.2394***
2 0.0067** 0.8727*** 0.0043** 0.5593*** 0.0562*** 0.3993*** 0.0319*** 0.2713***
3 0.0006* 0.8734*** 0.0004* 0.5597*** 0.0301*** 0.4293*** 0.016 5*** 0.2878***
4 0.0001 0.8734*** 0.000 0 0.5597*** 0.0163*** 0.4457*** 0.0087*** 0.2966***
5 0.0000 0.8734*** 0.0000 0.5597*** 0.0090*** 0.4546*** 0.0046*** 0.3012***
长期 - 0.8734*** - 0.5597*** - 0.4655*** - 0.3055***
网络效应(ne
0(同期) - - - - 0.6980*** 0.6980 1.3208*** 1.3208
1 - - - - -0.0904 0.6075 -0.2313 1.0895
2 - - - - -0.0560*** 0.5516 -0.1592*** 0.9303
3 - - - - -0.0350*** 0.5166 -0.1101*** 0.8202
4 - - - - -0.0200*** 0.4965 -0.0724*** 0.7477
5 - - - - -0.0117*** 0.4848 -0.0486*** 0.6991
长期 - - - - - 0.4662*** - 0.5884***
总体效应(te
0(同期) 0.7283*** 0.7280*** 1.3840*** 1.3840*** 0.9128*** 0.9128*** 1.9095*** 1.9095
1 0.0654*** 0.7936*** 0.1241*** 1.5081*** -0.0124 0.9005*** 0.0353 1.9448
2 0.0061** 0.7998*** 0.0116** 1.5197*** -0.0179* 0.8826*** 0.0004 1.9452
3 0.0006* 0.8003*** 0.0011* 1.5208*** -0.0169*** 0.8657*** -0.0101* 1.9351
4 0.0000 0.8004*** 0.0001 1.5209*** -0.0117*** 0.8541*** -0.0074*** 1.9277
5 0.0000 0.8004*** 0.0000 1.5209*** -0.0082*** 0.8458*** -0.0055*** 1.9222
长期 - 0.8004*** - 1.5209*** - 0.8250*** - 1.9080***

注: ******分别表示1%、5%和10%水平上显著

表2可看出,随着时间推移,各类效应的瞬时效应逐渐衰减并趋近于0,而累积效应也逐渐接近于长期效应,达到较为稳定的状态。变量的迁出地效应,无论是短期还是长期,在时空重力模型中都发生了较大变化:首先,在时空溢出作用下,区域人口规模对省域外迁的影响更为凸显。显然,一个地区的人口迁出规模与其人口数量之间存在一定的正相关关系。同时,由于时空溢出,周边迁移流的反馈作用在一定程度上进一步促进该区域的人口迁移。具体地,若区域人口数量增长10%,长期来看,该省份人口迁出量将累积增长10.14%,超过区域人口增长量。其次,区域人均GDP在短期内的迁出地效应随时间推移显著为负。区域经济发展将逐渐对后续时期的人口迁出起到显著的抑制作用,有助于留住本地人口,改善人才流失状况。最后,区域变量的作用期限在时空重力模型中明显拉长,这意味着在考虑时空交互作用之后,较强的时空溢出效应激发了区域变量变化对迁移流持续且长期的影响。
在时空重力模型中,区域人均GDP和人口数量对人口迁入均起到了积极的促进作用。从欠发达地区迁往发达地区是人口迁移的基本格局。随着我国城镇化进程的加速,省际人口迁移逐渐发展为向人口稠密区的集聚性迁移[45]。结合迁入地效应估计结果,若人均GDP水平提高10%,跨省外来人口数量同期增加2.36%,长期迁入人口增长4.66%;若区域人口数量增长10%,跨省外来人口数量同期增长1.78%,长期增长3.06%,略低于区域经济发展水平对迁移人口的吸引作用。相较于动态重力模型估计结果,人均GDP和人口数量的目的地效应均有所减小,时空溢出效应对平衡较强的目的地“拉力”有一定作用。总体上,区域经济发展水平高有利于吸引外来人口迁入,而区域人口规模大是促使本地人口外迁的重要推力。
网络效应主要体现了区域要素在迁移网络上的时空溢出现象。无论是短期还是长期,较强的时空交互作用使得区域人均GDP和人口数量的变化潜在地影响全国范围内其他省际间的人口迁移,起到“牵一发而动全身”的作用。具体地,若区域人均GDP提高10%,其他省份间的迁移人口同期平均增长6.98%,长期将增长4.66%;若区域人口数量增加10%,其他省份间的迁移人口同期平均增长13.21%,超过人口数量增长,长期增幅稳定在5.88%。图2表明,区域要素变化初期,网络效应明显强于其迁出地和目的地效应,表明区域要素对网络上其他迁移流累积的间接影响超过对该地区迁入流以及迁出流的直接影响,这意味着时空交互作用增强了迁移流之间的联系,尤其是短期内与周边迁移流之间的横向依赖。值得注意的是,长期网络效应均小于相应的同期效应。这意味着,迁移流之间的相关性不仅随着时间的推移而减弱,而且在后续时期,迁移流会受到周围迁移流的抑制作用,致使网络效应由正转负。网络效应累积和随时间逐渐减弱,在一定程度上避免了区域人口迁移规模发展的失控,使迁移行为更加有序。与此同时,逐步增强的迁出地和目的地效应在人口迁移进程中开始呈主导地位,起到推动迁移网络系统发展的作用。
Fig. 2 The chart of space-time spillover effects trends for space-time gravity model

图2 中国省际迁移时空重力模型时空溢出效应趋势图

总体效应从全局角度,考察了区域要素变化对所有迁移流的累积影响。具体地,区域人均GDP每提高10%,整个迁移网络上的人口迁移量同期增长9.13%,长期将稳定在8.25%。相对于人均GDP变化所产生的影响,人口因素对省际人口迁移起到更重要的作用。总体看来,区域人口数量每增长10%,迁移人口同期增长19.10%,长期将保持在19.08%,约为人均GDP作用的两倍。纵观整个迁移过程,人口规模对迁移流起到更显著的作用,促进省际间的人口迁移与流动;而长期来看,区域人均GDP对缓解迁移规模的过度增长更为有效。

4 结论

时空维度下的区域人口迁移网络系统构成一个有机整体。除了迁出地和目的地的“推-拉”作用,区域要素的时空溢出效应对整个迁移网络产生潜在的影响,是人口迁移过程的内在驱动力。改革开放以来,随着区域社会经济的快速发展,我国省际人口迁移活动日趋活跃。本文利用1985-2015年间6个时期的省际迁移面板数据,构建省际迁移时空重力模型,并利用贝叶斯MCMC方法对模型进行估计。通过二维多向的效应评价框架,有效地解释了时空环境下区域要素变化对人口迁移的直接与间接影响,以探究中国省际人口迁移活动的时空溢出效应和动力机制。整体来看,区域要素变化及其时空溢出作用共同驱动区域人口迁移的发展,从时空视角描述了人口迁移过程的内在特性与演变机制。初步得到以下结论:
(1)区域人口迁移过程存在时空上的“路径依赖”。时间上,前期迁移流会对当前迁移流以及未来迁移流产生正向带动效应;空间上,迁移流主要受到迁出地和目的地周边迁移流的辐射作用,且来自迁出地周边迁移流的影响更为强烈。从时空交互的长期过程来看,迁移流之间也存在一定的竞争行为,周边地区过去的迁移流会对当前迁移流产生抑制作用。区域人口迁移系统实质上是一个在时空尺度上相互联系且相互制约的网络系统。
(2)耦合时空依赖关系的时空重力模型能够有效评估区域要素对人口迁移的作用机制。以中国省际人口迁移为例,区域经济发展水平和人口数量是影响省际人口迁移进程的重要因素。其中,区域人口规模的影响尤为凸显,超过区域经济发展水平。从迁出地来看,区域人口规模较大是促进人口外迁的重要原因,而区域经济发展水平的提高将会在一定程度上抑制人口外迁;从目的地来看,区域经济发展状况是迁移者选择目的地的重要因素,同时固有的社会联系也起到了类似的牵引作用。
(3)区域经济社会要素的变化对迁移流产生显著的时空溢出效应。尤其在区域要素变化初期,网络溢出效应明显强于相应的迁出地和目的地效应,反映了区域迁移流之间的紧密联系。随着时间推移,逐渐衰减的溢出效应将在一定程度上削弱区 域要素对周边迁移流的辐射影响,但仍起到不可忽视的抑制作用。在时空交互作用下,区域人口迁移系统存在内在的调控与自适应过程,这与非空间的动态重力模型中持续增长的各类效应形成了鲜明对比。
本文构建的区域人口迁移时空重力模型和效应评价框架,对揭示区域人口迁移过程时空溢出效应与动力机制具有一定的意义。为进一步探究中国省际人口迁移的发展过程与演变趋势,今后需要在以下几方面加以改进:① 从就业环境、教育水平以及自然条件等方面引入更多合适的解释变量,对模型进一步完善,以期更为全面地探究省际人口迁移的影响机制;② 分时期对迁移流进行回归分析,并与整个研究时期的总体回归模型加以比较,以更深入地探究不同时期省际人口迁移的发展机制与演变规律;③ 着眼于当前人口迁移现状,对未来15-30年内人口迁移发展趋势进行短期与长期预测,并对其预测结果进行评价与分析。

The authors have declared that no competing interests exist.

[1]
Bijak J.Forecasting international migration in Europe: A Bayesian view[M]. Berlin: Springer, 2010.

[2]
White M J.International handbook of migration and population distribution[M]. Berlin: Springer, 2016.

[3]
LeSage J P, Pace R K. Spatial econometric modeling of origin-destination flows[J]. Journal of Regional Science, 2008,48(5):941-967.Downloadable (with restrictions)! Author(s): James P. LeSage & R. Kelley Pace. 2008 Abstract: Standard spatial autoregressive models rely on spatial weight structures constructed to model dependence among "n" regions. Ways of parsimoniously modeling the connectivity among the sample of "N"="n"-super-2 origin-destination (OD) pairs that arise in a closed system of interregional flows has remained a stumbling block. We overcome this problem by proposing spatial weight structures that model dependence among the "N" OD pairs in a fashion consistent with standard spatial autoregressive models. This results in a family of spatial OD models introduced here that represent an extension of the spatial regression models described in Anselin (1988). Copyright (c) 2008, Wiley Periodicals, Inc.

DOI

[4]
LeSage J P, Thomas-Agnan C. Interpreting Spatial econometric origin-destination flow models[J]. Journal of Regional Science, 2015,55(2):188-208.Spatial interaction or gravity models have been used to model flows that take many forms, for example population migration, commodity flows, traffic flows, all of which reflect movements between origin and destination regions. We focus on how to interpret estimates from spatial autoregressive extensions to the conventional regression-based gravity models that relax the assumption of independence between flows. These models proposed by LeSage and Pace (2008, 2009) define spatial dependence involving flows between regions. We show how to calculate partial derivative expressions for these models that can be used to quantify these various types of effect that arise from changes in the characteristics/explanatory variables of the model.

DOI

[5]
Sen A, Smith T E.Gravity models of spatial interaction behavior[M]. Berlin: Springer, 1995.

[6]
王亚平,蒲英霞,马劲松,等.基于空间OD模型的中国省际人口迁移机制分析[J].西北师范大学学报(自然科学版),2015,51(3):89-97.基于网络自相关理论与建模方法,利用全国第六次人口普查省际人口迁移数据和相关资料,分析了迁移流之间存在的几种网络空间依赖形式,据此构建空间OD模型,并对2005—2010年中国省际人口迁移的空间模式及其动力机制进行初步分析.结果表明,中国省际人口迁移流之间存在显著的空间依赖关系;分别选择网络权重矩阵Wo,Wd和Ww,利用Morans I对网络自相关程度进行测度,结果为0.502 5,0.286 7和0.107 8,且均在1‰水平上显著,说明抵达相同目的地(D)的人口迁移流的迁出地之间存在空间集聚现象;由同一迁出地(O)出发的迁移流的目的地之间也存在空间集聚现象;目的地和迁出地均相邻的人口迁移流之间也存在相互影响与制约.几种不同空间OD模型的拟合度均优于重力模型,其中迁出地的网络自相关效应更为显著,表明迁出地在地域空间上更为集中,抵达同一目的地的人口迁移流之间的相互作用更强.同时考虑3种不同网络自相关效应的空间OD模型更优,拟合度更好,该模型可以更加准确全面地揭示经济、就业、人口和距离等因素对人口迁移的影响作用.

DOI

[ Wang Y P, Pu Y X, Ma J S, et al.Mechanism analysis of inter-provincial migration flow s in China based on spatial OD models[J]. Journal of Northwest Normal University (Natural Science), 2015,51(3):89-97. ]

[7]
Etzo I.The determinants of the recent interregional migration flows in Italy: A panel data analysis[J]. Journal of Regional Science, 2011,51(5):948-966.ABSTRACT This study investigates the determinants of interregional migration flows in Italy in the light of the upsurge occurred in 1996, after two decades of decreasing internal migration rates. The fixed-effect vector decomposition estimator (FEVD) is applied to a gravity model using bilateral migration flows for the period 1996–2005. It is shown that the FEVD improves the estimates with respect to the traditional panel data estimators. The GDP per capita and the unemployment rate appear to be the key determinants whose changes push migrants out from their regions and direct them to “better off” destinations. Migrants leaving the regions in the Center-North respond differently to the push and pull forces compared to southern migrants. The dynamic model provides evidence for the presence of social networks, which in this analysis take place for each pair of regions.

DOI

[8]
Mitze T.On the mutual dynamics of interregional gross migration flows in space and time[C]. In: Patuelli R, Arbia G. Spatial Econometric Interaction Modelling. Switzerland: Springer International Publishing, 2016:415-440.

[9]
Rainer H, Siedler T.Social networks in determining migration and labour market outcomes: Evidence from the German Reunification[J]. Econ Transit, 2009,17(4):739-767.This paper empirically examines social network explanations for migration decisions in the context of German reunification. Using longitudinal data from the German Socio-Economic Panel, we first show that the presence of a family in West Germany is an important predictor for the migration hazard rate of East Germans. We then explore whether pre-migration networks have a discernible impact on the economic and social assimilation of East German immigrants in West Germany. We find that East German immigrants are more likely to be employed and to hold higher paying jobs when socially connected to the West prior to emigrating.

DOI

[10]
Chun Y, Griffith D.Modeling network autocorrelation in space-time migration flow data: An eigenvector spatial filtering approach[J]. Annals of the Association of American Geographers, 2011,101(3):523-536.Gravity-type spatial interaction models have been popularly utilized in modeling cross-sectional migration data, but their misspecification also has been raised in the literature. This misspecification issue principally concerns an insufficient accounting of underlying effects of spatial structure, including the presence of network autocorrelation among migration flows. Recent studies reveal that spatial interaction models are significantly improved by incorporating network autocorrelation in log-linear or Poisson regression estimation techniques, which are common estimation methods for spatial interaction models. However, when migration flows are structured as a panel data set from multiple time periods, the data set is likely to display temporal correlation within each measurement unit (here, each flow between a dyad of an origin and a destination) as well as network autocorrelation within each time period. Hence, spatial interaction models should be explicitly specified to account for these two different types of correlation structure. Using the eigenvector spatial filtering technique, this article outlines how to model network autocorrelation among migration flows structured through multiple time spans in either a linear or a generalized linear mixed model. An analysis of annual U.S. interstate migration data reported by the U.S. Internal Revenue Service shows that incorporation of two different types of autocorrelation leads to an improvement of model fitting and more intuitive parameter estimates.

DOI

[11]
Black W R.Network autocorrelation in transport network and flow systems[J]. Geographical Analysis, 1992,24(3):207-222.

[12]
Griffith D A, Jones K.Explorations into the relationships between spatial structure and spatial interaction[J]. Environment and Planning A, 1980,12:187-201.

DOI

[13]
蒲英霞,韩洪凌,葛莹,等.中国省际人口迁移的多边效应机制分析[J].地理学报,2016,71(2):205-216.

[ Pu Y X, Han H L, Ge Y, et al.Multilateral mechanism analysis of interprovincial migration flows in China[J]. Acta Geographica Sinica, 2016,71(2): 205-216. ]

[14]
Fan C C.Modeling interprovincial migration in China, 1985-2000. Eurasian Geography & Economics, 2005,46(3):165-184.Using data from China's 1990 and 2000 censuses, this paper examines interprovincial migration by describing its spatial patterns and estimating models based on the gravity approach. Between the censuses, interprovincial migration increased considerably in size and became more unidirectional and concentrated. Modeling results highlight the role of regional economic disparity and migration stock, as well as the gravity variables of population and distance, in explaining migration flows. Over time, the effect of regional disparity has increased, while that of distance decreased. Findings suggest that models based on experiences of countries lacking migration control are increasingly relevant for China. Journal of Economic Literature, Classification Numbers: C10, O15, O18. 3 figures, 3 tables, 45 references.

DOI

[15]
王桂新,潘泽瀚.中国人口迁移分布的顽健性与胡焕庸线[J].中国人口科学,2016(1):2-13.人口迁移在三大人口变动中最活跃,对人口分布影响最大。然而,文章通过对省际人口迁移规模分布、迁移强度分布、人口迁移流分布的考察,发现中国改革开放以来的省际人口迁移分布具有明显的顽健性。胡焕庸线历时80年,特别是改革开放以来持续频繁的人口迁移仍未能改变中国人口分布的基本格局,影响其稳定性。中国人口分布的基本格局及胡焕庸线,主要决定于地理位置固定不变的第一自然;人口迁移分布的顽健性与第一自然的决定作用密切相关。只要第一自然不变化,中国的人口分布及胡焕庸线就会保持高度稳定,人口迁移分布也不太可能失去顽健性而导致中国人口分布基本格局及胡焕庸线的改变。

[ Wang G X, Pan Z H.Into digitization: Some concepts and methods of Chinese historical geographic information system[J]. Historical Geography, 2002(1):2-13. ]

[16]
Elhorst J P.Spatial econometrics: From cross-sectional data to spatial panels[M]. Berlin: Springer, 2014.

[17]
Chun Y.Modeling network autocorrelation within migration flows by eigenvector spatial filtering[J]. Journal of Geographical Systems, 2008,10(4):317-344.Although the assumption of independence among interaction flows frequently is engaged in spatial interaction modeling, in many circumstances it leads to misspecified models and incorrect inferences. An informed approach is to explicitly incorporate an assumed relationship structure among the interaction flows, and to explicitly model the network autocorrelation. This paper illustrates such an approach in the context of U.S. interstate migration flows. Behavioral assumptions, similar to those of the intervening opportunities or the competing destinations concepts, exemplify how to specify network flows that are related to particular origin–destination combinations. The stepwise incorporation of eigenvectors, which are extracted from a network link matrix, captures the network autocorrelation in a Poisson regression model specification context. Spatial autocorrelation in Poisson regression is measured by the test statistic of Jacqmin-Gadda et al. (Stat Med 16(11):1283–1297, 1997). Results show that estimated regression parameters in the spatial filtering interaction model become more intuitively interpretable.

DOI

[18]
LeSage J P, Llano C. A Spatial interaction model with spatial structure origin and destination effects[J]. Journal of Regional Science, 2013,15(3):265-289.We introduce a Bayesian hierarchical regression model that extends the traditional least-squares regression model used to estimate gravity or spatial interaction relations involving origin-destination flows. Spatial interaction models attempt to explain variation in flows from n origin regions to n destination regions resulting in a sample of N 02=02 n 2 observations that reflect an n by n flow matrix converted to a vector. Explanatory variables typically include origin and destination characteristics as well as distance between each region and all other regions. Our extension introduces latent spatial effects parameters structured to follow a spatial autoregressive process. Individual effects parameters are included in the model to reflect latent or unobservable influences at work that are unique to each region treated as an origin and destination. That is, we estimate 2 n individual effects parameters using the sample of N 02=02 n 2 observations. We illustrate the method using a sample of commodity flows between 18 Spanish regions during the 2002 period.

DOI

[19]
Bhargava A, Sargan J D.Estimating dynamic random effects models from panel data covering short time periods[J]. Econometrica, 1983,51:1635-1659.This paper advocates the use of simultaneous equations estimators (especially LIML) to estimate dynamic random effects models from panel data. The methods are found to perform quite satisfactorily in Monte Carlo experiments. The LIML procedures are also extended to the case where some of the regressors are correlated with the effects and a theorem on identification is proved. Finally, the Michigan Panel is used for some illustrations.

DOI

[20]
Parent O, LeSage J P. A space-time filter for panel data models containing random effects[J]. Computational Statistics & Data Analysis, 2011,55(1):475-490.A space–time filter structure is introduced that can be used to accommodate dependence across space and time in the error components of panel data models that contain random effects. This specification provides insights regarding several space–time structures that have been used recently in the panel data literature. Markov Chain Monte Carlo methods are set forth for estimating the model which allow simple treatment of initial period observations as endogenous or exogenous. The performance of the approach is demonstrated using both Monte Carlo experiments and an applied illustration.

DOI

[21]
Griffith D A, Fischer M M, LeSage J P. The spatial autocorrelation problem in spatial interaction modeling: A comparison of two common solutions[J]. Letters in Spatial & Resource Sciences, 2017,10:75-86.Spatial interaction models of the gravity type are widely used to describe origin-destination flows. They draw attention to three types of variables to explain variation in spatial interactions across geographic space: variables that characterize the origin region of interaction, variables that characterize the destination region of interaction, and variables that measure the separation between origin and destination regions. A violation of standard minimal assumptions for least squares estimation may be associated with two problems: spatial autocorrelation within the residuals, and spatial autocorrelation within explanatory variables. This paper compares a spatial econometric solution with the spatial statistical Moran eigenvector spatial filtering solution to accounting for spatial autocorrelation within model residuals. An example using patent citation data that capture knowledge flows across 257 European regions serves to illustrate the application of the two approaches.

DOI

[22]
Parent O, LeSage J P. Spatial dynamic panel data models with random effects[J]. Regional Science & Urban Economics, 2012,42(4):727-738.

[23]
Chib S, Carlin B P.On MCMC sampling in hierarchical longitudinal models[J]. Kluwer Academic Publishers, 1999,9(1):17-26.

[24]
Elhorst J P.Spatial panel data models[C]. In: Fischer MM. Handbook of applied spatial analysis. Berlin: Springer, 2010:377-407.

[25]
LeSage J P, Pace R K. Introduction to spatial econometrics[M]. Boca Raton: CRC press, 2009.

[26]
LeSage J P. Spatial econometric panel data model specification: A Bayesian approach[J]. Spatial Statistics, 2014,9:122-145.A Bayesian model uncertainty argument is advanced that logically implies we can simplify this task by focusing on only two model specifications. One of these, labeled the spatial Durbin model (SDM) implies global spatial spillovers , while the second, labeled a spatial Durbin error model (SDEM) leads to local spatial spillovers . A Bayesian approach to determining an appropriate local or global specification, SDEM versus SDM is set forth here for static panel variants of these two models. The logic of the Bayesian view of model uncertainty suggests these are the only two specifications that need to be considered. This greatly simplifies the task confronting practitioners when using static panel data models.

DOI

[27]
Golgher A B, Voss P R.How to interpret the coefficients of spatial models: Spillovers, direct and indirect effects[J]. Spatial Demography, 2016,4:175-205.Abstract This paper briefly reviews how to derive and interpret coefficients of spatial regression models, including topics of direct and indirect (spatial spillover) effects. These topics have been addressed in the spatial econometric literature over the past 5鈥6 years, but often at a level sometimes difficult for students new to the field. Our goal is to overcome this handicap by carefully presenting the mathematics behind these spatial effects and clearly illustrating how they work using two small fictive datasets and one large dataset with real data. The motivation for the paper is primarily pedagogical. Theoretical and conceptual impediments associated with the application of procedures are discussed.

DOI

[28]
Debarsy N, Ertur C, LeSage J P. Interpreting dynamic space-time panel data models[J]. Statistical Methodology, 2012,9:158-171.There is a vast amount of literature regarding the asymptotic properties of various approaches to estimating simultaneous space–time panel models, but little attention has been paid to how the model estimates should be interpreted. The motivation for the use of space–time panel models is that they can provide us with information not available from cross-sectional spatial regressions. LeSage and Pace (2009) [7] showed that cross-sectional simultaneous spatial autoregressive models can be viewed as a limiting outcome of a dynamic space–time autoregressive process. A valuable aspect of dynamic space–time panel data models is that the own- and cross-partial derivatives that relate changes in the explanatory variables to those that arise in the dependent variables are explicit. This allows us to employ parameter estimates from these models to quantify dynamic responses over time and space as well as space–time diffusion impacts. We illustrate our approach using the demand for cigarettes over a 30 year period from 1963–1992, where the motivation for spatial dependence is a bootlegging effect where buyers of cigarettes near state borders purchase in neighboring states if there is a price advantage to doing so.

DOI

[29]
Parent O, LeSage J P. A spatial dynamic panel model with random effects applied to commuting times[J]. Transportation Research Part B, 2010,44:633-645.A space ime filter is set forth for spatial panel data situations that include random effects. We propose a general spatial dynamic specification that encompasses several spatiotemporal models previously used in the panel data literature. We apply the model to the case of highway induced travel demand. The theory of induced travel demand asserts that increased highway capacity will induce growth in traffic for a number of reasons. Our model allows us to quantify the spatial spillover impacts of increased highway capacity at one location in the network on travel times in neighboring locations and in future time periods.

DOI

[30]
国务院人口普查办公室,国家统计局人口和就业统计司.中国1990年人口普查资料[M].北京:中国统计出版社,1993.

[ Population Census Office under the State Council, Department of Population and Employment Statistics under the National Bureau of Statistics (NBS). Tabulation on the 1990 Population Census of the People's Republic of China[M]. Beijing: China Statistics Press, 1993. ]

[31]
国务院人口普查办公室,国家统计局人口和就业统计司.中国2010年人口普查资料[M].北京:中国统计出版社,2012.

[ Population Census Office under the State Council, Department of Population and Employment Statistics under the National Bureau of Statistics (NBS). Tabulation on the 2010 Population Census of the People's Republic of China[M]. Beijing: China Statistics Press, 2012. ]

[32]
国务院人口普查办公室,国家统计局人口和就业统计司.中国2010年人口普查资料[M].北京:中国统计出版社,2012.

[ Population Census Office under the State Council, Department of Population and Employment Statistics under the National Bureau of Statistics (NBS). Tabulation on the 2010 Population Census of the People's Republic of China[M]. Beijing: China Statistics Press, 2012. ]

[33]
国务院人口普查办公室,国家统计局人口和就业统计司.1995年全国1%人口抽样调查资料[M].北京:中国统计出版社,1997.

[ Population Census Office under the State Council, Department of Population and Employment Statistics under the National Bureau of Statistics (NBS). Tabulation on the 1995 1% Population Sampling Survey of the People's Republic of China[M]. Beijing: China Statistics Press, 1997. ]

[34]
国务院人口普查办公室,国家统计局人口和就业统计司. 2005年全国1%人口抽样调查资料[M].北京:中国统计出版社,2007.

[ Population Census Office under the State Council, Department of Population and Employment Statistics under the National Bureau of Statistics (NBS). Tabulation on the 2005 1% Population Sampling Survey of the People's Republic of China[M]. Beijing: China Statistics Press, 2007. ]

[35]
国务院人口普查办公室,国家统计局人口和就业统计司.2015年全国1%人口抽样调查资料[M].北京:中国统计出版社,2016.

[ Population Census Office under the State Council, Department of Population and Employment Statistics under the National Bureau of Statistics (NBS). Tabulation on the 2015 1% Population Sampling Survey of the People's Republic of China[M]. Beijing: China Statistics Press, 2016. ]

[36]
刘生龙. 中国跨省人口迁移的影响因素分析[J].数量经济技术经济研究,2014,31(4):83-98.

[ Liu S L.Factors analysis on interprovincial migration in China[J]. The Journal of Quantitative & Technical Economics, 2014,31(4): 83-98. ]

[37]
刘晏伶,冯健.中国人口迁移特征及其影响因素——基于第六次人口普查数据的分析[J].人文地理,2014,29(2):129-137.

[ Liu Y L, Feng J.Characteristics and impact factors of migration in China: Based on the analysis of the sixth census data[J]. Human Geography, 2014,29(2):129-137. ]

[38]
王桂新,潘泽瀚,陆燕秋.中国省际人口迁移区域模式变化及其影响因素水-基于2000和2010年人口普查资料的分析[J].中国人口科学,2012(5):2-13.文章通过对2000和2010年两次人口普查数据的比较分析发现,20世纪90年代以来中国省际人口迁移的区域模式具有相当的稳定性,但也出现了一些明显的局部性变化。如人口迁出和迁入的分布都呈一定集中趋势,且迁入集中趋势尤为明显;人口迁出呈“多极化”,迁入则更加集中化;一些重要的人口迁出地和迁入地的迁出和迁入强度表现出强者恒强、强者更强的特征。人口迁入重心北移,长三角都市圈取代珠三角都市圈成为21世纪以来省际人口迁入的主要地区。省际人口迁移区域模式及其变化,主要是迁入地城镇收入水平等经济因素的吸引作用明显增强所致,并与空间距离等恒定因素的影响相对弱化有关。

[ Wang G X, Pan Z H, Lu Y Q. China's inter-provincial migration patterns and influential factors: Evidence from year 2000 and 2010 population census of China[J]. Chinese Journal of Population Science, 2012(5):2-13. ]

[39]
国家统计局.中国统计年鉴(1988)[M].北京:中国统计出版社,1988.

[ National Bureau of Statistics (NBS). China statistical yearbook (1988)[M]. Beijing: China Statistics Press, 1988. ]

[40]
国家统计局.中国统计年鉴(1993)[M].北京:中国统计出版社,1993.

[ National Bureau of Statistics (NBS). China statistical yearbook (1993)[M]. Beijing: China Statistics Press, 1993. ]

[41]
国家统计局.中国统计年鉴(1998)[M].北京:中国统计出版社,1998.

[ National Bureau of Statistics (NBS). China statistical yearbook (1998)[M]. Beijing: China Statistics Press, 1998. ]

[42]
国家统计局.中国统计年鉴(2003)[M].北京:中国统计出版社,2003.

[ National Bureau of Statistics (NBS). China statistical yearbook (2003)[M]. Beijing: China Statistics Press, 2003. ]

[43]
国家统计局.中国统计年鉴(2008)[M].北京:中国统计出版社,2008.

[ National Bureau of Statistics (NBS). China statistical yearbook (2008)[M]. Beijing: China Statistics Press, 2008. ]

[44]
国家统计局.中国统计年鉴(2013)[M].北京:中国统计出版社,2013.

[ National Bureau of Statistics (NBS). China statistical yearbook (2013)[M]. Beijing: China Statistics Press, 2013. ]

[45]
杨云彦. 中国人口迁移的规模测算与强度分析[J].中国社会科学,2003(6):97-107.

[ Yang Y Y.Into digitization: Some concepts and methods of Chinese historical geographic information system[J]. Historical Geography, 2002(6):97-107. ]

文章导航

/