For example, Lardy (1975, 1978) argues that regional development ...... Bao, S., Henry, M. S. & Barkley, D. (1995) RAS: A Regional Analysis System Integrated.
GIS AND SPATIAL MODELING IN REGIONAL DEVELOPMENT STUDIES: A CASE OF GREATER BEIJING
by
Danlin Yu
A Dissertation Submitted in Partial Fulfillment of the Requirements for the Degree of
Doctor of Philosophy in Geography
at
The University of Wisconsin – Milwaukee May 2005
GIS AND SPATIAL MODELING IN REGIONAL DEVELOPMENT STUDIES: A CASE OF GREATER BEIJING
by
Danlin Yu
A Dissertation Submitted in Partial Fulfillment of the Requirements for the Degree of
Doctor of Philosophy in Geography
at
The University of Wisconsin – Milwaukee May 2005
Major Professor
Date
Graduate School Approval
Date
ii
ABSTRACT GIS AND SPATIAL MODELING IN REGIONAL DEVELOPMENT STUDIES: A CASE OF GREATER BEIJING by
Danlin Yu The University of Wisconsin – Milwaukee, 2005 Under the Supervision of Yehua Dennis Wei
This research focuses on the application of GIS and spatial analysis to urban/regional development studies. Geographic analysis has recently been widely used in regional studies. Application in China is limited due partly to data availability and methodological development. In this research, I initiate the attempt to establish an analytical framework of GIS and spatial data analysis in China’s regional development research. The study employs a multi-scale, multi-methodology approach in analyzing geographic data (geographic information). A general GIS and spatial data analysis framework is set through a study at provincial level in China. Focus of the study then turns to analyze data at county level, based on the case of Grater Beijing, China. Three types of intensive GIS and spatial data analysis are carried out. First, I investigate the spatial patterns and dynamics of regional development through GIS and exploratory spatial data analysis (ESDA). In particular, the Moran’s Index of per capita GDP is discussed in detail for regional/urban development
iii
pattern analysis. The analysis reveals significant positive spatial autocorrelation in Greater Beijing from 1978 – 2001. Dynamic analysis indicates such spatial autocorrelation is increasing. Furthermore, local spatial autocorrelation analysis finds a spatial urban-rural two-tier structure, which agrees with the common wisdom of China’s regional development patterns. Such spatial structure is shattered during the reform. Instead, a south-north divide in Great Beijing emerges. Second, spatial regression is employed for understanding global regional development mechanisms from a spatial econometric perspective. Data from 1995 and 2001 are used. Results reveal that the spatial models perform better than the ordinary least squares model. Specifically, in 1995 and 2001, the signs of all the development factors’ coefficients remain the same. However, their magnitude and significance change. In particular, per capital foreign direct investment and fixed investment become less influential in the spatial models. Per capita local financial expense emerges to be more influential. Third, geographically weighted regression models are developed to investigate spatially varying development mechanisms. The analysis reveals significant spatially varying development mechanisms. Influences of globalization and domestic investment seem to complement each other in space. It indicates regional development has strong local characteristics.
Major Professor
Date
iv
TABLE OF CONTENTS
CHAPTER 1 INTRODUCTION .................................................................................... 1 1.1 BACKGROUND .......................................................................................................... 1 1.2 RESEARCH OBJECTIVES ............................................................................................ 3 1.3 RESEARCH ON CHINA’S REGIONAL DEVELOPMENT ................................................... 4 1.4 REGIONAL DEVELOPMENT THEORIES: A CRITICAL REVIEW ...................................... 8 1.4.1 Convergence: the neo-classical theories and inverted-U models...................... 9 1.4.2 Divergence: cumulative causation and the backwash argument ..................... 13 1.4.3 Growth pole and product life cycle theories ................................................... 15 1.5 ORGANIZATION OF THE RESEARCH ......................................................................... 17 CHAPTER 2 A GIS/SPATIAL ANALYSIS ENVIRONMENT FOR REGIONAL DEVELOPMENT IN CHINA ...................................................................................... 19 2.1 MULTI-SCALAR PATTERNS OF REGIONAL DEVELOPMENT ....................................... 21 2.2 INVESTIGATING REGIONAL DIVERGENCE WITH GLOBAL MORAN’S I ....................... 33 2.3 ANALYZING SPATIAL ASSOCIATION WITH THE MORAN SCATTERPLOT .................... 36 2.4 IDENTIFYING GEOGRAPHICAL CLUSTERING WITH LOCAL MORAN’S I ...................... 40 2.5 UNDERSTANDING REGIONAL DEVELOPMENT IN CHINA ........................................... 45 2.6 CONCLUSION .......................................................................................................... 49 CHAPTER 3 REGIONAL DEVELOPMENT IN GREATER BEIJING ................. 53 3.1 INTRODUCTION ....................................................................................................... 53 3.2 GREATER BEIJING: SETTING AND DEVELOPMENT PROCESS ..................................... 57 3.3 GLOBAL SPATIAL PATTERNS AND DYNAMIC SPATIAL PROCESSES IN GREATER BEIJING ........................................................................................................................ 62 3.4 LOCAL SPATIAL PATTERNS AND NON-STATIONARITY ............................................. 66 3.4.1 Moran scatterplot and statistical characteristics of local Moran’s Ii ............. 66 3.4.2 Empirical analysis of Greater Beijing’s local pattern, 1978 - 2001 ............... 68 3.5 CONCLUSION AND DISCUSSION ............................................................................... 76 CHAPTER 4 DEVELOPMENT MECHANISMS IN GREATER BEIJING – A SPATIAL ECONOMETRIC PERSPECTIVE ........................................................... 80 4.1 INTRODUCTION ....................................................................................................... 80 4.2 DEVELOPMENT IN THE GREATER BEIJING: STATE, GLOBALIZATION, AND MARKETIZATION .......................................................................................................... 82 4.3 MODEL SPECIFICATION AND SPATIAL ECONOMETRIC METHODOLOGY .................... 85 4.3.1 Model specification......................................................................................... 85 4.3.2 Spatial econometric methodology ................................................................... 86 4.4 RESULTS AND DISCUSSION ..................................................................................... 89 4.5 CONCLUSION .......................................................................................................... 95 CHAPTER 5 SPATIALLY VARYING MECHANISMS – A FURTHER INVESTIGATION OF DEVELOPMENT MECHANISMS VIA GEOGRAPHICALLY WEIGHTED REGRESSION ANALYSIS ........................... 97
v
5.1 INTRODUCTION ....................................................................................................... 97 5.2 SPATIALLY VARYING MECHANISMS: TECHNIQUES OF GWR ................................... 99 5.3 SPATIALLY VARYING DEVELOPMENT MECHANISMS IN GREATER BEIJING .............. 103 5.4 CONCLUSION AND DISCUSSION ............................................................................. 114 CHAPTER 6 DISCUSSION AND CONCLUSION .................................................. 117 REFERENCES ............................................................................................................ 123 CURRICULUM VITAE ............................................................................................. 138 EDUCATION ................................................................................................................ 138 RECENT PUBLICATIONS .............................................................................................. 138 Peer-reviewed articles (in English) ....................................................................... 138 Proceeding paper (in English)............................................................................... 139 Papers in review (in English) ................................................................................ 139 Software packages authored and contributed ........................................................ 139 CONFERENCE PRESENTATIONS ................................................................................... 139 HONORS AND AWARDS............................................................................................... 140
vi
LIST OF FIGURES Figure 2.1 China and its regions ...................................................................................... 23 Figure 2.2 Interprovincial and interregional CV in China from 1978-2000 ..................... 24 Figure 2.3 Changes of LQs of the three municipalities.................................................... 25 Figure 2.4 Changes of LQs of the five coastal provinces ................................................ 26 Figure 2.5 Changes of LQs of the five central provinces and Hebei, Hainan .................. 27 Figure 2.6 Changes of LQs of the three northeastern and two central provinces ............. 27 Figure 2.7 Changes of LQs of the five northwestern provinces ....................................... 28 Figure 2.8 Changes of LQs of the four southwestern provinces and Guangxi ................. 29 Figure 2.9 Per capita GDP in 1978 .................................................................................. 30 Figure 2.10 Per capita GDP in 2000 ................................................................................ 31 Figure 2.11 Change of global Moran’s I from 1978 – 2000 ............................................ 35 Figure 2.12 Moran scatterplot of per capita GDP in 1978 ............................................... 38 Figure 2.13 Moran scatterplot of per capita GDP in 2000 ............................................... 39 Figure 2.14 Moran scatterplot map in 2000 ..................................................................... 41 Figure 2.15 Local Moran’s I in 1978 ............................................................................... 42 Figure 2.16 Local Moran’s I in 2000 ............................................................................... 43 Figure 3.1 Location of Greater Beijing ............................................................................ 58 Figure 3.2 Per capita GDP in Greater Beijing, 1978 (a) and 2001 (b) ............................. 60 Figure 3.3 Changing LQs of provinces in Greater Beijing and southeastern China ......... 62 Figure 3.4 Changing global Moran’s I in Greater Beijing, 1978-2001 ............................ 66 Figure 3.5 Moran scatterplot of Greater Beijing, 1978 .................................................... 70 Figure 3.6 Moran scatterplot in Greater Beijing 2001 ..................................................... 71
vii
Figure 3.7 Moran scatterplot maps in 1978 (a), 2001 (b)................................................. 72 Figure 3.8 Local spatial pattern maps, 1978 (a), 2001 (b) ............................................... 72 Figure 4.1 Distribution of per capita FDI in 1995 (a) and 2001 (b) ................................. 93 Figure 5.1 Fixed (a) and adaptive (b) weighting schemes in GWR…………………….102 Figure 5.2 Spatially varying mechanisms in 1995 for Greater Beijing………………...108 Figure 5.3 Spatially varying mechanisms in 2001 for Greater Beijing………………...110
viii
LIST OF TABLES Table 2.1 Grouping of China's provinces based on change of LQs ................................. 24 Table 2.2 Growth rates of the provinces in China, 1978 – 2000 ...................................... 32 Table 2.3 Result of regression analysis for the 1990 – 2000 period................................. 47 Table 3.1Characteristics of alternative spatial weight matrixes ....................................... 64 Table 3.2 Significance test (p-values) of Moran’s I for different weighting strategies .... 65 Table 4.1 OLS results for 1995 and 2001 ........................................................................ 89 Table 4.2 LM tests for the OLS models under different weighting strategies .................. 90 Table 4.3 Spatial autoregressive error models for 1995 and 2001 ................................... 91 Table 4.4 State-owned enterprises in Greater Beijing...................................................... 95 Table 5.1 Global OLS regression results for 1995 and 2001 in Greater Beijing ............ 104 Table 5.2 Comparison between OLS and GWR models in 1995 and 2001 ................... 105 Table 5.3 Spatial stationarity tests for individual variables in 1995 and 2001 ............... 106 Table 5.4 Stationary part of the 1995 mixed GWR model ............................................. 106
ix
1
Chapter 1 Introduction 1.1 Background Regional development is one of the primary concerns of geographers and a basic subject for geographical research. Regional development in China has attracted scholarly attentions for the last decades, especially the development issues during the reform period. Among the various research themes, regional inequality in China receives specific attention. This is mainly due to regional inequality in current China is a burning issue (Wei 1996, 1999, 2000; Yu and Wei, 2003; Wei and Ye 2004). The studies yield a large body of fruitful findings. However, due largely to data issues, previous studies on China’s regional development mainly focus on provincial level spatial units and policy analyses. Recent trends tend to emphasize multi-scale, multi-mechanism understanding in China’s regional development issues (Wei 2002). In particular, when various provinces issued their temporal county-level socioeconomic data during the past several years, studies on China take this advantage quickly. Scholars argue that better understanding could be achieved by in-depth analysis utilizing data at county-level (Wei and Fan 2000; Wei and Kim 2002). As the data became more and more readily available, more sophisticated means to manipulate the information are also required for in-depth data analysis. Traditional means in China’s regional development studies mainly focus on non-spatial statistical and empirical analysis. Geographical information systems/sciences (GIS) is employed mainly as a visual instead of an analytical means, though the employment of GIS and spatial methodology on other parts of the world (mainly North America, Europe) booms in the
2 literature (for instance, Rey and Montouri 1999; Fotheringham et al. 2002; Gallo and Ertur 2003). The recent development of GIS and spatial data analysis, especially the development of exploratory spatial data analysis (ESDA) and spatial modeling techniques during the past decade and the convergence of GIS and spatial data analysis (Goodchild and Hanning, 2004), lend power for fulfilling such tasks. In addition, although studies on county-level China recently take off, they mainly focus on the southern/southeastern coastal part of China, especially Jiangsu (see Wei 2000; Wei and Fan 2000; Long and Ng 2001; Wei and Kim 2002), Zhejiang (Wei and Ye 2004) and Guangdong (Lin 1997; Weng 1998; Gu et al. 2001). Systematic research on other provinces and regions remains limited (Wei and Ye 2004). There are only very few case studies on other provinces of China, such as Fujian (Lyons 1997), Sichuang (Bramall 1993). Given the massive scale of China and the tremendous differentials within the country, more studies of other representative provinces/regions are needed (Wei and Ye 2004). Studies on Greater Beijing intend to fill in the blank. This paper intends to explore the regional development issues in Greater Beijing, China, during the reform period (1978-2001). I adopt a multi-scalar, multi-methodology approach in studying the regional development issue in this area. The study focuses on the county-level data analyses and intensively relies on GIS and spatial data analysis methodology during the investigation. One of the goals of this paper is trying to quantitatively investigate the inherent geographical variation as well as spatial interaction in regional development, through GIS/spatial modeling and spatial statistical approaches. The other goal of this study is to examine the characteristics of regional development in different locales, in a hope to complement the current wisdom of China’s regional
3 development. This research intends to contribute to the literature of regional development studies in China in three aspects: first, it initiates an effort to set a GIS/spatial data analysis environment for regional development studies in China, specifically at the subprovincial level. In particular, this study promotes the application of exploratory spatial data analysis (ESDA) in understanding the dynamic patterns of regional development. Second, in an effort to understand regional development mechanisms, this study stresses the importance of spatial dependence and heterogeneity in the process. Instead of studying development mechanisms based on the non-spatial ordinary least squares regression methods, this study discusses the utilization of spatial econometric and local analysis techniques in mechanism study. Specifically, spatial regression and geographically weighted regression methods are detailed for better analyzing crosssectional data in geographic units. Third, through the case study in Greater Beijing, which is quite different from the often-studied southeastern China provinces in terms of economic and ownership structure, reform policies, and industrial bases, this study intends to provide a more complete view of China’s regional development, and identify regional specific development characteristics. In the following sections of this chapter, the research objectives of this study will be briefly outlined in section 2. A review of current studies on China’s regional development is presented in section 3. Section 4 discusses some of the theoretical aspects of regional development. Section 5 outlines the organization of this study. 1.2 Research Objectives As aforementioned, this study intends to fulfill two primary objectives, which include studying regional development through GIS and providing more complete view
4 of China’s regional development. Specifically, first, the promotion of GIS and spatial data analysis in China’s regional development will be the major theme of the study. The entire study is based on the idea that in analyzing cross-sectional data on geographic units, spatial effects will be considered explicitly as an inherent factor instead of as an exogenous input. In other words, the analyses are “all about geography”. In order to achieve the objective, I invest a reasonably large amount of time to develop new analyzing algorithms and programs in conjunction with the commercial GIS software packages. The study intends to bring robust spatial data analytical methodology into the China’s regional development research community. Second, the study takes the initial step to turn the research focus from the oftenstudied southern/southeastern coastal provinces in China to the northern China, taking Greater Beijing as the case study. In addition, although provincial level studies are still carried out, the focus of the study is on data analysis at county-level. It is hoped that the practice of this study will bring better understanding of China’s regional development from a more complete viewpoint, and promote future research to pay more attention on other parts of the vast country. 1.3 Research on China’s Regional Development When the People’s Republic of China (PRC) was established in 1949, the government inherited a very imbalanced China due both to the historical context and geographical differentiation. This imbalanced heritage determines China’s spatial development pattern, and it further leads the early constructors of the young Republic to put enormous efforts in trying to reduce regional inequality (for example, Chinese leader Mao Zedong once discussed the imbalanced regional development in his well-known
5 “planning-guiding” essay of “Discussion on the Ten Great Relationships” in 1956, Wei 2000). Regional equilibrium was the central idea of regional development in that time (Li 1999). There have been plethora of debates on whether and to what extent the imbalanced regional development was ameliorated during the pre-reform period (see Wei 1999, 2000 for a review). For example, Lardy (1975, 1978) argues that regional development converged during the Mao’s period, due primarily to the implementation of government policies, particularly the policies relating transferring various resources, including financial, human capital and investment efforts to the less developed regions. While other scholars also argue that since China hold the regional autarky policies, the unfair price structure between raw materials and final products, and the low efficient and poor investment-return projects – the “three frontiers construction” during the 1960s, regional development diverged during Mao’s period (Donnithorne 1972; 1976; Kueh 1989; Tsui 1991; Wei 2000). Those debates all capture a side view of China’s regional development during that specific time period, but the research on this period is usually impeded by the lack of systematic and comparable data. Beginning after the 1990s, the Chinese government has publicly released a series of relatively accurate and comparable socioeconomic datasets, which provides opportunities for scholars to re-examine the research of this period. By using the datasets, most recent literature suggests that the efforts of reducing regional inequality during the pre-reform period China have proved to be largely ineffective (Bo 1993; Fan 1995; Wei 2000; Wei 2002, Yu and Wei 2003, to name but a few). As more accurate and complete datasets are available, the literature generally agrees upon that during the first “FiveYear-Plan” (FYP, 1953-1957), regional inequality was reduced through the industrial
6 policies and spatial resources relocation. However, the premature advance action taken by the central government in the “Great Leap Forward” followed the first FYP, the Korean War, the break-up with the former Soviet Union in the later 1950s and early 1960s, the feeling of war-threat from its northern neighbor, the following “great famine”, together with the catastrophic “cultural revolution”, nearly destroyed the nation’s economic as well as social bases. During this catastrophic period, the national economy status decreased dramatically, and the regional gaps enlarged in the meantime. Regional development of China in the reform period has become the focus of most recent literature (see Wei 1999 for a review). After the Chinese government launched its open-door policies in 1978, regional development in China began to deviate from its former version. Decentralization, privatization and more importantly, the marketization process in the early 1990s, became the central ideas for the government’s regional development plots. Differing from the experiences of the Eastern European countries and former Soviet Union, China did not take the change with a gulp, but gradually, one step after another. The unprecedented success of China’s economic and social achievement during the reform period attracts various scholarly as well as political interests. However, as also noticed by a number of scholars (see Wei 1999, 2000, 2002; Lu and Wang 2002, Yu and Wei 2003 for a full review), the success and prosperity of the reform are concentrated on the eastern coastal provinces in China. The vast territory of the rest of the nation remains underdeveloped (Yu and Wei 2003). For instance, in 2001, per capital GDP in the richest region, Shanghai is more than 4 times in the poorest province, Guizhou. Unfortunately, this spatial disparity of economic development does not reveal itself as a great threat to the national economic system until after the early
7 1990s, especially under the cloak of the government’s early slogan “let some regions get rich first”. The Tiananmen Square incident in 1989 brought crucial adjustments in China’s consideration on regional development and related policy issues. Various efforts were again made to try to tackle the issue of inland-region development and cooperation between the core and periphery regions (Li 1999). In both the Ninth FYP (1996-2000) and the tenth FYP (2000-2005), the enlarging gap of regional development was deemed a serious threat to the nation’s long-term prosperity, stability, unity and sustainability. In recent years, the movement of “developing the Great West” indicates a regional development policy change that is supposed to keep the China’s prosperity in a long period of time. The current literature on China’s regional development pays less attention to the less developed regions and provinces. As aforementioned, most studies on China’s development focus their geographical scope on either the national level or those relatively developed regions, mainly the southeastern coastal provinces, such as Shanghai, Zhejiang, Jiangsu, Guangdong, Hong Kong, etc. This is due largely to scholars’ interests and government’s permission to the access to data in the early years of reform (Wei and Ye 2004). Even so, the full boom of the research on China’s national level regional development, especially the period after the reform, provides a relatively full picture about the general developing patterns of China’s regional development (see Wei 2000, Wei and Fan 2000, Wei and Kim 2002, Ying 2003, Yu and Wei 2003, Wei and Ye 2004). A set of post-reform Chinese regional development theories and mechanisms is established based on relatively accurate and complete data analyses. However, scholars
8 still recognize that better understanding demand multi-scalar (Wei 2002) and multimethodology analysis. Wei and Fan (2000) argue that sub-provincial data might provide richer information on regional development in China. In addition, the developing trajectories, the Chinese government’s step-by-step reform strategies and the great geographical differentiation result very different provincial reality, not only from the national patterns, but also from one another. As a result, the researching focus turns to individual provinces. Nevertheless, the regional focus is mainly on the southeastern coastal areas. For example, Wei (1999; 2000; 2002 with Kim; 2004 with Ye) has a series of cases studies on Jiangsu and Zhejiang province. Fan’s (1995, 1997) research on Jiangsu and Hong Kong; and recent boom of studies on Shanghai (Wu, 2000, Walcott 2002, to name but a few) make the majority of the literature on China’s regional studies at sub-provincial level. The studies on northern and inland China remain quite limited, where the economic and ownership structures, reform policies and focuses are quite different from their southern peers. For instance, the southern provinces are previously less developed and less favored by the central government during the Mao’s period, but benefit the most from the reform. In addition, the industrial structure in the northern provinces is dominated by state-owned heavy industries, while their southern peers are relatively light in state-owned industries (Wei and Fan 2000; Wei and Ye 2004). It hence merits further investigation (with comparison) on these northern provinces to see how development patterns differ from or agree with their southern peers. 1.4 Regional Development Theories: A Critical Review From a theoretical perspective of regional development, there has long existed the dichotomy between regional divergence and convergence. Scholars studying China’s
9 regional development observe a mixed and often changed view of practices of those theories in China (Wei 1999; Ying 2003). In particular, convergence school based on neo-classical theories (Borts and Stein 1964; Barro and Sala-I-Martin 1991, 1995), divergence school based on cumulative causation (Myrdal 1957), and the grow pole and product life circle theories are of important influence on regional development in China. 1.4.1 Convergence: the neo-classical theories and inverted-U models Convergence theories claim that regional development in the first stage of national economic growth generally exhibits an imbalanced pattern. Such regional inequality stems from the lack of coordination between the local system (regional economic growth) and the national system (national economic growth). Over a period of time, however, with the increase in free movement of production factors (labor and capital) between regions, regional inequality will be minimized or possibly eliminated (Lipshitz 1992). Two major types of convergence are identified and studied intensively in the literature (Barro and Sala-i-Martin 1991; Wei 1996, 2000). First, the σ-convergence, indicates the common version of convergence, while the dispersion among regions decline over time. This is an absolute convergence. Second, the β-convergence, indicates the trend that poorer regions grow faster than richer ones, while the absolute difference between the two may not necessarily decline among a certain period of time. In the convergence school, the typical representatives are the neo-classical economic theory and inverted-U models. The neo-classical theory views regional development in a three-stage path, driven by the process of resources relocation, i.e., the mobility of capital and labor (Lipshitz 1992; Wei 1999, 2000; Ying 2003). At the initial stage, there is very little mobility of labor and capital among different regions. The
10 difference among regions, i.e., the difference between the labor-intensive less developed regions and the capital-abundant developed regions, triggers these two resources to relocate themselves for higher profit or higher wages. The following stage hence sees high level of resources mobility. Consequently in the third stage the relocation of labor causes labor to be abundant in the developed regions while decrease in the less developed regions. Capital, on the other hand, tends to move to labor-intensive (thus less expensive) and more profitable sectors in less developed regions, also creates demand for labor and increases wages. Both processes trigger the equalization of wage levels between the developed and less developed regions and lead the regional development into an equilibrium state (Borts and Stein 1964). Neo-classical theory emphasizes equilibrium conditions, free mobility of production factors (labor and capital) and the importance of market in allocating resources. It holds that regional inequality is an inevitable stage for the final equilibrium state. The neo-classical theory has been very influential in the initial reform policies in China, and provides politically operable theoretical understanding of regional inequalities. However, the neo-classical economic convergence theory is established based on the long-term development of western industrial countries. The vast difference between the contemporary China, even after the economic reform, and the free market western economies, leads to serious considerations of the applicability of such theory in China’s practice (Wei 1999, 2000, 2002). Several attacks can be addressed to the neo-classical explanation of regional development in China: First, the neo-classical theory assumes that the perfect competition exists in the market system, and production factors move freely. However, even in the most advanced
11 market economy countries like the United States, these assumptions are hardly “perfectly” satisfied. In reality, market competition is always conducted under the circumstance of incomplete information, which results “imperfect competition”. In addition, production factors (labor and capital) are rather stickier than the theory would expect. Labor may not be willing to move to distant places for “pure” high wages. For instance, neo-Keynesian economists also argue that the allocation of the capital stock largely depends on its initial distribution and the structural parameters such as saving and capital-output ratios (Reynolds 1987). Second, the essential logic of neo-classical theory may not necessarily hold true. The flows of production factors do not necessarily cause equalization among regions as the theory predicts. It is true that there is a large amount of migration from the peripheral regions to the core regions during the second stage of economic development. However, this migration is very selective, composed of the young, skilled, and relatively more educated labor that is also most needed by the peripheral regions for further development (Brown and Stetzer 1984; Brown and Lawson 1989; Lipshitz 1992, Yu and Mao 1999). Such migration therefore detracts from the growth potential of peripheral regions, instead of diminishing the regional inequality. The early stage after the reform in China witnessed such a fact (Lu 1997; Yu 1999). Third, the theory largely ignores geographical, historical, cultural, behavioral, institutional, and political factors in regional development. Economic incentives are considered the most important ones in examining regional development. However, as pointed out by Myint (1971), Porter (1990) and Krugman (1991), regional development is a mixed result of the combined influence of all those factors. Among those factors,
12 geographical space is very important in influencing regional trade flows and regional development. Yet neo-classical school of development tends to assume a homogeneous space. Following the neo-classical theory, the inverted-U model illustrates how the regional development goes through the path of an “inverted-U” pattern. It holds that though regional development is not necessarily driven by factor mobility, the eventual status of a regional economic system will achieve equilibrium following an “inverted-U” path. Mydal (1957) argued that development begins in a few regions with location advantages and began a polarization or cumulative causation process. The process advanced and tended to equalize the regional inequality among regions via economic diffusion, improvement in transportation and communication (factor mobility), and the government intervention. Hirschman (1958) suggests that early polarization and cumulative causation process will end via the “trickle-down” effect, which he believes can be achieved via strong government intervention and aid. Williamson (1965), apart from emphasizing on the “labor migration, capital migration” (as the neo-classical theory suggests), states that central government policies will help to form “interregional linkages” while regional development reaches certain stage. This will diminish the regional inequality in the long run (for example, Vining and Strauss 1977; Mera 1978; Tabuchi 1988). However, like the neo-classical theory, the inverted-U model received many critiques (for example, Friedmann 1973; Gilbert and Goodman 1976; Pred 1977; Friedmann and Weaver 1979; Stohr and Todlting 1979; Krebs 1982; Fisch 1984). First, history indicates that the interregional gap is not a phenomenon of an ephemeral existence, and most of the countries in the world suffered and are suffering from weak
13 diffusion or even no diffusion at all from the rich regions. Second, the effects of government intervention differ dramatically all over the world. Though there are some successful examples, most government interventions are not as effective as expected. Third, the spatial economic interactions that bring about regional growth, such as the trickle-down effect, exist primarily within those cores and rich regions themselves. They have little to do with the peripheral regions’ development. 1.4.2 Divergence: cumulative causation and the backwash argument In divergence school of regional development, the spatial flow of production factors, contrary to the convergence school, actually increases inter-regional gaps. This school of thoughts comprises two differing theories: the planned economics or government intervention theory and the radical theory, by their attitudes towards the government intervention – whether the intervention is effective or not. The former one postulates that government intervention is liable to reduce the interregional gap (as some inverted-U scholars, such as Williamson 1965, suggest). The radical school holds that the government’s activities tend to perpetuate regional inequality. Among the planned economic theories, the most widely known representative is that of Myrdal’s cumulative causation explanation (Myrdal 1957). The cumulative causation refers to the spiral build-up of advantages that occurs in specific geographic settings as a result of the development of agglomeration effects, external economies, and localization economies (Knox and Agnew 1998). Once those regions get started with advantages, they will keep getting better. This is based on the contention that in a market economy changes in the location of economic activities produce cumulative advantages
14 for one region rather than a straightforward equalization of growth across all regions. In his book Economic Theory and Under-developed Regions, Myrdal observes that: “… in the normal case there is no … tendency towards automatic selfstabilization in the social system. The system is by itself not moving towards any sort of balance between forces, but is constantly on the move away from such a situation. In the normal case a change does not call forth countervailing changes but, instead, supporting changes, which move the system in the same direction as the first change but much further. Because of such circular causation a social process tends to become cumulative and often to gather speed at an accelerating rate.” (Myrdal 1957, p13). By this observation, Myrdal (1957) points out the very problem that confronts the neo-classical theory: the spiral of local growth involved in cumulative causation tends to attract those enterprising young people and investment funds (capital) from the less developed ones. This creates the “backwash” effect, and it is the reason why regional inequality persists or even intensify. Observing that Myrdal put little attention on the role of technological innovation on regional development, Friedmann (1973) built a wider theoretical framework that places on the role of technological innovation in regional development. In his broad framework, four factors emerge as the major forces for regional development: migration, flow of capital investment, spatial diffusion of technological innovation, and spatial organization of political power. Their simultaneous influences cause instability in the spatial system and tend to intensify the inequality between the core and the periphery (Friedmann 1973; Friedmann and Douglas 1978).
15 Among the radical theories, the dependency and structural schools are among the most widely studied ones. They hold that regional inequality under the capitalist society and the market economic system is inevitable and irreversible (Peet 1975; Walker 1978; Soja 1980; Harvey 1982; Lipshitz 1992). Regional inequality under capitalism has been seen as necessary for capital accumulation, and the capitalism mode of production actively creates, intensifies, and maintains regional inequality, as capital tends to move to core areas with higher profit rates. This is opposite to the neo-classical assumption about the core and periphery (Soja 1980; Harvey 1982). In their opinions, the spread of capital to the periphery creates a dependent core-peripheral structure, sustains labor outflow and unequal transfer of value, exacerbates the stagnation of the periphery, and intensifies regional inequality (Franks 1967; Emmanuel 1972; Slater 1975). Although distinction between the two groups of theory is discernible, the divergence models essentially allow for increasing returns to scale. The developed regions tend to outgrow the poor regions due to economies of scale and agglomeration. Capital and labor tend to flow in the same direction, which, contrary to the equilibrium theories, indicates that factor mobility stimulates regional divergence other than equalizes it. In today’s China, after more than two decades of economic reform, the regional development reality finds its similarity as the divergence school might suggest. 1.4.3 Growth pole and product life cycle theories Apart from the above two schools of regional development, there are also other theories trying to generalize the patterns and mechanisms of regional development. The growth pole theory and production cycle models in regional development have important impacts on China’s regional development. These two theories deem that divergence and
16 convergence are not a unique trend for regional development. Instead, they are seen as results from various causes, and may present in different regions at the same time, or different times in the same region. Perroux (1955) first introduces the concept of growth poles. Growth poles are those locations where entrepreneurial innovation and “propulsive industries” (those that attract other industries and stimulate new ones in the vicinity) are located, and they serve as the engines for regional development (Perroux 1955). Krugman (1991) suggests a more complex and more formal model to account for the core-periphery pattern of economic development and for the possibility of its transformation. In his book Geography and Trade, he argues that once the core locations are established through the conjoint operation of increasing returns to scale in plant operations, transportation costs, and demands, it tends to lock into certain geographical location and keeps growing. However, critical point can be reached in the periphery once the population here grows to a certain mass. The new market opportunities in the periphery areas stimulate the growth of production facilities, followed by a dramatic shift of regional fortunes. Regional inequality tends to be reversed (or at least even) at this point. Increasing returns, imperfect competition, and even historical accident contribute to the reversal. The identity of the favored region is not set for all time, convergence and divergence interweave through regional development. The product life cycle theory also argues for this interweavement of convergence and divergence in regional development. This theory holds that regional economic development exhibits a wave-like behavior (Amos, 1990; Berry, 1991). Prosperity and depression dominates the regional economic development in turn. The change from
17 prosperity to stagnancy, decline, depression and recovery synchronize and follow the technical innovation life-cycle process. Prosperity provides an impetus for concentration and divergence dominates the regional development process. However, during the stagnancy period, economic spread from the core to the periphery decreases the regional imbalance. This product life-cycle style of regional development has been usually used to explain the reemergence of regional inequality in major development countries in early 1980s, a time of economic booming right after the 1970s’ economic depression in the developed economies (Amos 1989; Maxwell and Hite 1992). Nonetheless, this explanation is also criticized in the literature. Scholars argue that the economic longwave theory does not necessarily apply to the regional development process (Wei 1999). In addition, the core regions with well-established innovation-support facilities are more competitive and more capable in leading another cycle of innovation. That means even during the stagnancy period, economic spread from the core to the periphery does not necessarily happen. Furthermore, regional development might not be the same as economic development, and technological innovation (Friedmann 1973) is only one of the driving forces of regional development. 1.5 Organization of The Research This research contains six chapters. Following this introductory chapter, chapter 2 sets the GIS and spatial analysis environment in the regional development studies in China through an analysis of regional development at provincial level. In this chapter, I will demonstrate the trends, patterns and general mechanisms of China’s regional development using coefficient of variation, Gini’s Index, exploratory spatial data analysis (mainly Moran’s I and its decomposed components) and regression models. The basic
18 spatial analytical techniques are employed and framework of utilizing spatial data analysis techniques and GIS is established in the chapter. Chapter 3 begins the investigation of GIS/spatial data analysis at the county-level in Greater Beijing. Choosing per capita GDP as an index for regional development, this chapter follows the GIS/ESDA framework established in the previous chapter. Methodology of ESDA and the integration within the framework of GIS will be discussed in detail. Intensive attention will be paid at the local pattern analysis. Chapter 4 turns the attention of GIS/spatial data analysis to multivariate aspects, in an attempt to understand regional development from various driving forces. In this chapter, the regression model for examining regional development mechanisms is established. The methodology of spatial regression model is introduced. The comparison between non-spatial regression and spatial regression model sets the viewpoint of examining regional development mechanisms from a spatial econometric perspective. Chapter 5 extends the understanding of development mechanisms. Based on the previous chapter’s analyzing results, my fieldwork and past working experience in this region, I hypothesize that regional development in this area contains different spatial regimes. That is spatial heterogeneity in the underlying forces of regional development. A geographically weighted regression (GWR) model is developed to test and investigate the hypothesis. The geographically weighted regression (GWR) model and the analytical results are discussed intensively, in a hope to identify spatial regimes in regional development mechanisms. Chapter 6 concludes the study by discussion of the findings of the study and future research foci.
19
Chapter 2 A GIS/Spatial Analysis Environment for Regional Development in China During the post-1978 reform period, China implemented radical economic reforms and opened up to the outside world. Influenced by the Inverted-U model and the “ladder-step theory”1, the government argues that concentration and coastal development will accelerate national development, and that as the economy matures diffusion will eventually stimulate interior development and reduce regional inequality (Fan 1995; Wei 2000). Since the early 1990s, China has accelerated the process of globalization and economic liberalization. With rising inequality and the resentment of interior China, the government has put more focus on poverty reduction and interior development. The central government has announced a strategy of western development to promote the development of poorer regions, and reduce regional inequality in China. Research on post-Mao China has revealed the intensification of China’s coastalinterior divide, although interprovincial inequality declined in the 1980s (Fan 1995; Wei 2000; Lu and Wang 2002). Scholars have emphasized the importance of multi-scale and multi-mechanisms in understanding changing patterns of regional inequality in China (Wei 2000; 2002). They have also analyzed regional inequalities within provinces of China, especially in Jiangsu and Guangdong (e.g., Wei and Fan 2000; Gu et al. 2001; Wei and Ye 2004), and trajectories of local/regional development (e.g., Ma and Cui 2002). 1
The Inverted-U model regards that regional development will follow an inverted-U trajectory: Rising regional inequalities and dualism are typical of early development stages, whereas regional convergence and a disappearance of severe dualism are typical of the more mature stages of regional growth and development (Williamson, 1965). “Ladder-step theory” follows the same logic, and argues that during the process of development, the wealth will diffuse from richer regions to poorer ones, like down from a ladder.
20 Despite the advance in the study of regional development in China during the last few years, further research is needed in several areas: There is, first, a need for more rigorous data analysis. Much new ground has been broken in GIS and spatial data analysis research (Goodchild and Hanning 2004), but sophisticated GIS and spatial data analysis techniques are rarely used in the study of China’s regional development. Second, most of the publications on interregional and interprovincial developments in China have study periods that ended in the early 1990s, such as Fan (1995) and Wei and Ma (1996). It has become necessary to update previous studies to understand patterns of change in the 1990s, a decade during which China experienced more rapid reform. Through the analysis of recently published data (up till 2000) within a GIS and spatial data analysis framework, this chapter intends to embark the task. In particular, this chapter attempts to achieve three objectives. First, recent development in GIS in studying regional development is incorporated and an analytical environment and framework for GIS and spatial data analysis. Using GIS, especially exploratory spatial data analysis (ESDA) techniques, this chapter analyzes clusters of regional development for a better understanding of changing regional development patterns in China. Second, since most of the existing studies cover the period up to the early 1990s, this chapter attempts to provide a more complete picture of dynamic regional development pattern in post-Mao China by extending the study period to 2000. In particular, this chapter will examine whether inter-provincial inequality continued to decline in the 1990s, a trend detected for the 1980s. Finally, through regression analysis, this chapter intends to explain the changing patterns of regional development by considering the context of China and incorporating recent development in economic geography. The following section discuss
21 the developing status of the 29 provincial spatial units (Not include Taiwan and the newly established municipality Chongqing, which is included in Sichuang province, for temporal data consideration). Cluster analysis in conjunction with empirical evidence tries to group them according to the fortune transform process during the reform. Section 3 investigates the spatial clustering process of the Chinese provinces during the reform with ESDA. Specifically, global and local Moran’s I, the Moran’s scatterplot are employed for detailed analysis. The next section establishes a regression model to further understand the underpinning forces of regional development in China during the reform period. Foreign investment, State Owned Enterprises (SOEs) reform, population factors, β-convergence and the location factor are elected as development mechanisms. The last section concludes this chapter. 2.1 Multi-Scalar Patterns of Regional Development While national income is frequently used in analyzing regional development under Mao’s period, for the post-Mao period, regional per capita GDP has become the indicator most often used. In addition, since regional inequality is a particularly burning issue in current China’s regional development (Wei 2000), this chapter intends to investigate regional development via the study of regional inequality instead of depicting regional development via per capita GDP per se. Several indexes are commonly employed in the literature for measuring regional inequality. The most often used is the coefficient of variation (CV) (see Wei 2000 for a review). For temporal comparison, the real per capita GDP is used instead of that in current prices.
22 Research on regional inequality in post-Mao China documented a trend of declining interprovincial inequality during the 1980s, but the changing patterns in the 1990s remain less clear. Figure 2.12 shows geographical regions and provincial level administrative units (hereafter simplified as provinces) in China for reference. As evident in Figure 2.2, CVs for interprovincial inequality exhibit a U-shaped pattern: it declined in the 1980s, as revealed by many others (Fan 1995; Wei and Ma 1996; Zhao 1996; Wei 2000), but has risen substantially since 1990. Specifically, interprovincial CV decreased from 0.96 in 1978 to 0.83 in 1990 but increased to 0.94 in 2000. An analysis of the other commonly used inequality index, such as Gini coefficient and Theil’s index, also shows a U-shaped pattern, albeit less smooth than CV. This extended data series analysis indicates that interprovincial inequality in China during the past two decades does not follow convergence, divergence, or inverted-U patterns. For scalar comparison purpose, interregional inequality across the eastern, central, and western regions is added. Figure 2.2 clearly shows that at larger scale, China’s regional inequality has been rising consistently during the study period, without any sign of convergence.
2
Chongqing was divided from Sichuan province in 1997 and set as a provincial-level municipality. However, for the consideration of data continuation, in our analysis, we included Chongqing into Sichuan province.
23
Figure 2.1 China and its regions Development trajectories of individual provinces have a tremendous impact on the change of regional inequality (Wei and Fan 2000; Wei and Kim 2002). To understand individual provinces’ development trajectories and their influence on China’s regional development, the location quotient (LQ) of per capita GDP for each province is calculated here to depict the changing fortunes of the provinces. Based on the provinces’ geographical location and changing patterns of LQs, as well as a cluster analysis and other classifications by scholars working on China (e.g., Wei and Ma 1996), I have
24 classified the provinces into six groups (Table 2.1). The three municipalities of Beijing, Tianjin and Shanghai are often grouped together (Group I), although cluster analysis identified substantial differences among their changing patterns of LQs. In general, their LQs fluctuate at high values, with Beijing staying almost the same during the reform era, while Tianjin’s LQs declining (Figure 2.3). LQs for Shanghai, however, declined from 1978 to 1990, but have been rising since then. The changing LQs of Shanghai resembles that of the CV, indicating the significant contribution of Shanghai to changing interprovincial inequality in China.
Coefficient of Variation
1.20 1.00 0.80 Interprovincial
0.60
Interregional
0.40 0.20
Year
0.00 2000 1999 1998 1997 1996 1995 1994 1993 1992 1991 1990 1989 1988 1987 1986 1985 1984 1983 1982 1981 1980 1979 1978 Figure 2.2 Interprovincial and interregional CV in China from 1978-2000 Table 2.1 Grouping of China's provinces based on change of LQs Group I Municipalities II Coastal III Central IV Industrial V Northwestern VI Southwestern
Provinces Shanghai, Beijing, Tianjin Jiangsu, Zhejiang, Guangdong, Fujian, Shandong Hubei, Hunan, Anhui, Jiangxi, Henan, Heibei, Hainan Liaoning, Jilin, Heilongjiang, Neimenggu, Shanxi Shaanxi, Gansu, Qinghai, Ningxia, Xinjiang Sichuan, Guizhou, Yunnan, Xizang, Guangxi
25
LQ
8 7 6 5 4 3 2 1 0
Beijing Shi
Tianjin
Shanghai
Year
2000 1999 1998 1997 1996 1995 1994 1993 1992 1991 1990 1989 1988 1987 1986 1985 1984 1983 1982 1981 1980 1979 1978 Figure 2.3 Changes of LQs of the three municipalities Group II includes five coastal provinces: Jiangsu, Guangdong, Zhejiang, Fujian and Shandong. This group benefits the most from China’s reforms, and their status in the national economy rises dramatically (Figure 2.4). Guangdong, Zhejiang, and Jiangsu were among the fastest growing provinces during the reform in China, and are identified by cluster analysis as a coherent group. Fujian, while traditionally poorer, has also recorded rapid growth, and emerges to join its southern coastal peers. Shandong, another coastal province, but in north China, grows slower than the other four provinces in this group; but the changing pattern of its LQs resembles that of its southern neighbors.
LQ
26
2 1.8 1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 0
Jiangsu Shandong
Zhejiang Guangdong
Fujian
Year
2000 1999 1998 1997 1996 1995 1994 1993 1992 1991 1990 1989 1988 1987 1986 1985 1984 1983 1982 1981 1980 1979 1978 Figure 2.4 Changes of LQs of the five coastal provinces The LQs of Hebei, Hainan and the five central provinces of Hubei, Hunan, Anhui, Jiangxi and Henan (Group III) had relatively slight changes during the reform. Indeed, except for Hainan, which was separated from Guangdong province in 1988 and recorded rapid growth from 1988 to 1993, their status was relatively stable during the 1990s (Figure 2.5). The status of Hebei declined during the 1980s but rose in the 1990s. The status of Anhui and Hubei also rose during the 1990s, while Hunan declined the most. Even Hainan’s status, with the shift of preferential policies to the Yangtze Delta in 1992, declined and stabilized at the end of the research period.
LQ
27 1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 0
Hubei Henan
Hunan Hebei
Anhui Hainan
Jiangxi
Year
2000 1999 1998 1997 1996 1995 1994 1993 1992 1991 1990 1989 1988 1987 1986 1985 1984 1983 1982 1981 1980 1979 1978 Figure 2.5 Changes of LQs of the five central provinces and Hebei, Hainan The status of the three northeastern provinces and the two northern central provinces (Group IV), which were either favored by Mao’s industrialization policy or industrialized based on the natural resources (such as coal), eroded during the reform period (Figure 2.6). The status of Heilongjiang declined quite substantially, as evidenced by the decrease of LQ from 1.5 in 1978 to 0.87 in 2000. Their economies, dominated by SOEs and resource-consuming heavy industry, have been slow in restructuring and challenged by non-state enterprises in other coastal regions. They have also fallen behind their coastal peers in opening up to the outside world.
2
LQ
1.5 1 0.5 0
Jilin Liaoning Shanxi
Heilongjiang Neimenggu
Year
2000 1999 1998 1997 1996 1995 1994 1993 1992 1991 1990 1989 1988 1987 1986 1985 1984 1983 1982 1981 1980 1979 1978 Figure 2.6 Changes of LQs of the three northeastern and two central provinces
28 Groups V and VI include all nine western provinces and one coastal province, Guangxi. Their status declined substantially during the reform. Guizhou, Qinghai, and Ningxia are the biggest losers during the reform period (Figures 2.7 and 2.8). Their struggle and declining status during the 1980s caught the attention of the central government, and more efforts have been made since the mid-1990s to improve their economic conditions. However, the LQs declined further in the 1990s, indicating that central government policies have not been able to stop their eroding status. Xizang (Tibet) has been the recipient of special policies from the central government since the early 1950s. With the rapid growth of the coastal provinces, however, its relative status has been declining as well.
1.4 1.2
LQ
1 0.8 0.6 0.4 0.2 0
Shaanxi Ningxia
Gansu Xinjiang
Qinghai
Year
2000 1999 1998 1997 1996 1995 1994 1993 1992 1991 1990 1989 1988 1987 1986 1985 1984 1983 1982 1981 1980 1979 1978 Figure 2.7 Changes of LQs of the five northwestern provinces
29 1.4 1.2
LQ
1 0.8 0.6 0.4 0.2 0
Sichuan Guangxi
Guizhou Xizang
Yunnan
Year
2000 1999 1998 1997 1996 1995 1994 1993 1992 1991 1990 1989 1988 1987 1986 1985 1984 1983 1982 1981 1980 1979 1978 Figure 2.8 Changes of LQs of the four southwestern provinces and Guangxi To shed more light on growth side of regional development in China, the growth rates of provincial per capita GDP is calculated. Specifically, in 1978, the municipalities (Group I) and some of the industrial provinces (Group II) had higher per capita GDP, followed by selected coastal and interior provinces (most are resource-based or politicaloriented, such as Qinghai, Gansu and Xizang), while provinces in the southwest were the poorest (Figure 2.9)3. However, as shown in Table 2.2, the growth rates of Shanghai and Tianjin (Group I) during 1978-1990 were among the lowest, and that of Beijing was below the average, while the southeastern coastal provinces (Group III) led in growth. The slower growth of the municipalities is attributed to their slower economic reforms. During the first decade of the reform, Shanghai and Tianjin in particular were heavily burdened by their problematic state-owned enterprises. The eroding status of traditionally leading provinces and the emergence of a group of coastal provinces led to the decline of interprovincial inequality during the period from 1978-1990. Overall, the coastal region
3
We used natural breaks (Jenks) in ArcGIS as classification method to divide the 30 provinces/municipalities into five classes according to their GDP per capita (in current prices) in Figures 9 and 10.
30 recorded faster growth than the interior region, leading to the rise of the coastal-interior divide.
Central Region Western Region
Eastern Region
±
Yuan 175 - 291 292 - 430 431 - 680 681 - 1290 1291 - 2498
0 0
250 500
500 1,000
1,000
Miles 2,000
Figure 2.9 Per capita GDP in 1978
Kilometers
31
Central Region Western Region
Eastern Region
±
Yuan 869 - 1913 1914 - 3162 3163 - 4532 4533 - 8411 8412 - 15593
0 0
250
500
500
1,000
1,000
Miles 2,000
Kilometers
Figure 2.10 Per capita GDP in 2000
Since 1990, not only did the coastal provinces keep outgrowing the interior provinces, but also the municipalities began to speed up. With the efforts to revitalize Shanghai, Shanghai’s growth accelerated. It became one of the fastest growing provinces in 1990-2000 (Table 2.2). Consequently, due to the rising status of Shanghai and the continued growth of the coastal provinces, interprovincial inequality in China increased during the 1990s. By 2000, the provinces with the highest per capita GDP were all located in the coastal areas, while the poorest provinces were located in the interior region (Figure 2.10). The gap between the coastal and interior regions increased dramatically during the period from 1978-2000.
32 Table 2.2 Growth rates of the provinces in China, 1978 – 2000 Per capita GDP Growth Rate (%) Growth Rate (%) (Yuan, in real prices) 1978 1990 2000 (1978-1990) (1990-2000)
Label Eastern (Coastal) Region Liaoning LN 680 Hebei HeB 364 Beijing BJ 1290 Tianjin TJ 1160 Shandong SD 316 Jiangsu JS 430 Shanghai SH 2498 Zhejiang ZJ 331 Fujian FJ 273 Guangdong GD 369 Hainan HaN 314 Guangxi GX 225 Central Region Heilongjiang HLJ 564 Jilin JL 381 Neimenggu NMG 317 Shanxi SX 363 Henan HeN 232 Hubei HuB 330 Anhui AH 244 Hunan HuN 286 Jiangxi JX 276 Western Region Sichuan SC 262 Guizhou GZ 175 Yunnan YN 226 Xizang XZ 375 Shaanxi SSX 291 Gansu GS 348 Qinghai QH 428 Ningxia NX 370 Xinjiang XJ 313 Data Source: SSB (1999; 2001).
1491 792 3038 2345 848 1300 5052 1105 829 1276 1009 412
3530 2474 8411 6629 2927 4532 15593 4253 3162 4340 2898 1111
6.8 6.7 7.4 6.0 8.6 9.7 6.0 10.6 9.7 10.9 10.2 5.2
9.0 12.1 10.7 11.0 13.2 13.3 11.9 14.4 14.3 13.0 11.1 10.4
1092 941 822 784 598 820 593 592 647
2295 2244 1913 1755 1636 2316 1703 1498 1777
5.7 7.8 8.3 6.6 8.2 7.9 7.7 6.2 7.4
7.7 9.1 8.8 8.4 10.6 10.9 11.1 9.7 10.6
617 419 572 725 711 757 737 843 880
1515 869 1260 1704 1605 1638 1442 1568 1894
7.4 7.5 8.0 5.6 7.7 6.7 4.6 7.1 9.0
9.4 7.6 8.2 8.9 8.5 8.0 6.9 6.4 8.0
33 2.2 Investigating Regional Divergence with Global Moran’s I
Popular regional inequality indexes such as CV and the Gini coefficient can only reveal overall inequality while the relationships among neighboring regions are ignored. While location quotients are useful in depicting the changing status of regions, both types of indexes have limited utility in unfolding spatial agglomeration and interregional relations. Recent development in GIS and spatial data analysis techniques has provided effective tools to analyze spatial association, agglomeration and clustering, which can shed more light on the understanding of regional development in China. Moran’s Index (or Moran’s I) has been used to detect spatial autocorrelation, and to analyze spatial relationships among regions (Upton and Fingleton 1985; Aneslin 1995; 1996). The Global Moran’s I is given as: n
I=
n n
•
n
∑∑ w i =1 j =1
n
∑∑ w i =1 j =1
ij
( xi − x )( x j − x ) (2.1)
n
∑ (x
ij
i =1
i
− x)
2
Here, wij is the binary weight matrix of the general cross-product statistic, such that wij=1 if locations i and j (two spatial units) are adjacent (share boundary) and zero for all non-adjacent pairs, and by convention wii =0 (a cell or region is not adjacent to itself). xi and x are the summary measure (in our case, the per capita GDP) in the ith province and the average of them, respectively. Without losing generality, the weights could be manipulated so that
n
n
∑∑ w i =1 j =1
ij
= n . The most often used manipulation is the so-called
row-standardization (Anselin 1988). After the manipulation, the values of Moran’s I fall within the range from -1 to +1, with positive values indicate spatial clustering process or
34 there existing strong positive regional spillover; near the expected value (1/n-1) means a random pattern (i.e., there is no obvious spatial pattern), and with negative values negative spatial autocorrelation (i.e., geographic units with dissimilar values cluster on space, which is vary rare in the context of regional development studies). In this analysis, the study is concerned with the positive value only, since the calculation of China’s Global Moran’s I indicates all the values are positive and significant. As an indication of spatial concentration, global Moran’s I can be used to identify the spatial clustering process. Increasing global Moran’s I indicates geographic units become more and more alike their neighbors. Decreasing global Moran’s I indicates that the clusters are disappearing and regional development is randomly distributed over space. To calculate the Moran’s I, the most important step is to determine the spatial neighbor weight matrix4. In this study, the weight matrix is derived according to each province’s spatial contiguity in ArcGIS® and R (R Development Core Team 2004). I then calculate global Moran’s I for every year from 1978 to 2000 in R (R Development Core Team 2004). Figure 2.11 depicts global Moran’s I of per capita GDP for each year from 1978 to 2000. The analysis confirms that the index in each year is significant (at the 10% confident level in 1978 and 1979, and at least 5% confident level in the rest of the years – significant test is carried out based on saddlepoint approximation – Tiefelsdorf 2000,
4
There are different ways of determining how to assign values to wij instead of simply considering the spatial units’ spatial adjacency. Anselin (1988) has an extensive discussion on this issue. Distance between two spatial units is also commonly used, but debates still exist (Anselin, 1988; Bao, et al, 1995). In this chapter, since our purpose is to set the GIS and spatial data analysis environment, detailed discussion of specifying different spatial weight matrixes will be delayed until the next chapter. Here, for simplification, I only consider spatial adjacency.
35 2002) and ranged positively from 0.11 to 0.22, indicating that provincial China exhibits positive spatial association during the reform era. Unlike the U-shaped pattern of CVs, however, global Moran’s I increased over time, although there was a little drop in 1989 and 1990 due to the Tianamen incident and economic slowdown. This indicates provinces in China become more similar to their spatial neighbors.
Global Moran's I
0.25 0.2 0.15 0.1 0.05
Year
0 2000 1999 1998 1997 1996 1995 1994 1993 1992 1991 1990 1989 1988 1987 1986 1985 1984 1983 1982 1981 1980 1979 1978
Figure 2.11 Change of global Moran’s I from 1978 – 2000 In the 1980s, CVs declined, which indicates interprovincial convergence. However, the convergence revealed by CV only indicates a convergence in development (as represented by per capita GDP). Global Moran’s I, on the other hand, reveals a trend of spatial concentration. This indicates that the development convergence in the 1980s has a spatial clustering characteristic. However, in the 1990s, when CV reveals a divergence trend in development, global Moran’s I keeps increasing. Moran’s I increased the most during the mid-1980s and early 1990s when China implemented more radical economic reforms. This concurrent increase of CV and global Moran’s I reveals a similar spatial process that yields different development consequences than in the 1980s. The analysis of global Moran’s I and CV suggests that spatial concentration was already in process during the 1980s. CV, however, fails to identify such a trend, suggesting that CV,
36 as an index measuring overall dispersion, does have limitations in analyzing regional inequality as it ignores spatial interaction among geographic units. With the rise of the coastal-interior gap, the Chinese government in the mid-1990s began to stress the reduction of poverty and regional inequality, as evidenced by the Ninth Five-Year Plan (1995-2000) and the China Agenda 21st (Yu and Mao 1999). It is also noticeable from Figure 2.11 that after 1995, global Moran’s I increased only slightly, indicating that although spatial concentration continues, its speed has been declining. 2.3 Analyzing Spatial Association with the Moran Scatterplot
Global Moran’s I reveals the overall spatial associations, but does not provide information about the spatial association of individual spatial units (provinces). It may therefore mask pockets of non-stationarity among provinces that deviate from the overall pattern. In addition, it’s also hard for GIS to visualize the global index and provide a more detailed analysis. For these reasons, this section utilizes the Moran’s scatterplot to undertake a disaggregated analysis of regional development in China. The Moran scatterplot was first developed by Anselin (1996) as an exploratory spatial data analysis (ESDA) tool to assess local instability in spatial association. The central idea of the Moran scatterplot is to treat Moran’s I as a regression coefficient of the spatial lag wy5 against the observed value (in the form of departure from the mean) y. In the Moran scatterplot, the four different quadrants divide two types of spatial association (i.e., the positive and negative associations) into four detailed types of local spatial association between individual geographic units. Among the positive association,
5
A spatial unit’s spatial lag is a weighted average of the value of its neighbors. The weights are obtained from the same spatial contiguity matrix that was used to calculate the Moran’s Index.
37 quadrant I, HH, indicates high values surrounded by high values, and quadrant III, LL, indicates low values surrounded by low values; while among the negative associations, quadrant II, LH, and quadrant IV, HL, indicate low values surrounded by high values, and high values surrounded by low values, respectively. In addition, since the scatterplot is derived from a linear regression equation, the extent of the fitness of the scatterplot indicates the degree of local instability. Such an interpretation of the index provides an effective way of summarizing the overall spatial pattern and detecting the individual spatial instability in the sense that a lack of fit would indicate important local pockets of non-stationarity (Anselin 1996). Figures 2.12 and 2.13 give the Moran scatterplots in 1978 and 2000, with a linear smoother superimposed. Figure 2.14 is the corresponding GIS visualization of the 2000’s scatterplot. The provincial labels are listed in Table 2.2.
Spatial lag of GDP per capita (standardized)
38
3.0 JS
TJ BJ
ZJ
2.0
HeB 1.0 JL
NMG XJ HaN
0.0
SH
FJ QH SX HeN AH SGS D SSX NX JX SC XZ GD HuN GX YNHuB GZ
HLJLN
-1.0 -1
0
1
2
3
4
5
GDP per capita (standardized)
I: High values surrounded by high values II: Low values surrounded by high values III: Low values surrounded by low values IV: High values surrounded by low values Figure 2.12 Moran scatterplot of per capita GDP in 1978 The two scatterplots and map indicate clearly that most provinces fall in quadrant III, the LL type (20 out of 30 in 1978, and 18 out of 30 in 2000), indicating that at the beginning year of reform and the end year of the study period, spatial patterns of China’s regional development were dominated by geographical clustering of relatively underdeveloped provinces. It is also noticeable from the superimposed regression lines that in both years the scatterplots lack fitness, which suggests there is spatial instability of the general spatial concentration process in regional development in China.
Spatial lag of GDP per capita (standardized)
39
3 JS TJ
ZJ
2
BJ
HaN
SH
HeB
1
FJ AH JX JL SD SX NMG LN GD HeN GX HuNHLJ SSX NX QH GS HuB XJ SC XZ GZ YN
0
-1
-2 -1
0
1
2
3
4
5
GDP per capita (standardized)
I: High values surrounded by high values II: Low values surrounded by high values III: Low values surrounded by low values IV: High values surrounded by low values Figure 2.13 Moran scatterplot of per capita GDP in 2000 The changing positions of the five coastal provinces of Shanghai, Jiangsu, Zhejiang, Guangdong, and Fujian in the scatterplots in particular draw our attention. In 1978, Shanghai was in quadrant IV, i.e., high values surrounded by low values, indicating that Jiangsu and Zhejiang, the two provinces surrounding Shanghai, had low per capita GDP. Provinces surrounding Fujian (Guangdong, Zhejiang, and Anhui) fell into quadrant III, which means that, Fujian was an underdeveloped province in comparison to its neighbors at this time. However, the rapid growth of coastal China during the reform
40 period changes their positions in the Moran scatterplot in 2000. As we can see in Figure 2.13, except for Guangdong, all the other southeastern provinces (Jiangsu, Zhejiang and Fujian) moved to quadrant I, i.e., high values surrounded by high values. This indicates that a highly developed economy cluster has formed in southeastern China (Figure 2.14), while most of the other provinces remained less developed. This emerging cluster very much accounts for the increase of interprovincial inequality in the 1990s. The shift of Guangdong from quadrant III to quadrant IV reinforces the fact that there is an enlarging gap between coastal provinces (especially the southeastern ones) and their interior neighbors. In 1978, Guangdong was in quadrant III, grouped with its neighbors as a province with low per capita GDP. However, its faster growth dwarfed its neighbors, and Guangdong moved to the fourth quadrant; its high value is surrounded by low ones. 2.4 Identifying Geographical Clustering with Local Moran’s I
Another decomposition strategy of the global Moran’s I is to derive a so-called local index of spatial association (LISA, Anselin 1995) for further understanding of local instability. Scholars have documented that the weighted average of the local Moran’s I is equal to the global Moran’s I, up to a factor of proportionality (Anselin 1995; 1996). However, the value of the local Moran’s I is usually ignored for analysis in the literature. Since the global Moran’s I is an indicator for global spatial clustering, at the provincial level, I intend to use the value of local Moran’s I to detect how individual provinces contribute to the global index, and therefore understand how individual provinces contribute to the spatial clustering process.
41
Central Region Western Region
Eastern Region
±
High-High High-Low Low-High Low-Low
0 0
250 500
500 1,000
1,000
Miles 2,000
Kilometers
Figure 2.14 Moran scatterplot map in 2000 As shown in Figure 2.11, during the reform era, the global Moran’s I increased almost every year, indicating a spatial clustering trend among China’s provinces. When the global Moran’s I is decomposed into its local form, and comparing 1978 with 2000 (Figures 2.15 & 2.16)6, several interesting findings emerge. First, China’s coast-interior divide persisted, and interior provinces exhibit great geographical similarity in terms of per capita GDP and their spatial contributions to the global Moran’s I. In 1978, the local Moran’s I in most of the interior provinces fell within the range 0-0.3, except for Guizhou, the poorest province, which contributes slightly more than the interior average. Coastal 6
The maps in figures 15 and 16 do not include Hainan province. We did this intentionally for two reasons: first, Hainan province was separated from Guangdong Province in 1988; it would make no sense for us to separate it from Guangdong in the 1978’s map. Second, in our case, we use the topological contiguity as a standard to retrieve the spatial neighbor matrix; since Hainan spatially does not have any instant connection with the mainland, the calculation process will treat it as an island and ignore it. For these reasons, we included Hainan into Guangdong province, and did not explicitly show it on the maps.
42 provinces of Shanghai and Zhejiang are the two biggest negative contributors, while Heilongjiang, Jilin, Liaoning, Hebei and Jiangsu have a negative local contribution around -0.1. This indicates that at the beginning of reform, spatial clustering is not a major spatial pattern among China’s provinces. Indeed, Shanghai, the most developed province, was the largest negative contributor to the global Index. Considering Shanghai’s large Location Quotient, it is understandable that Shanghai’s large negative contribution to the global index greatly evened the regional imbalance at the beginning of the reform.
Central Region Western Region
Eastern Region
±
Local Moran's I < -0.3 -0.3 - 0 0 - 0.3 0.3 - 1
0
250
500
1,000
Miles
>1 0
500
1,000
2,000
Figure 2.15 Local Moran’s I in 1978
Kilometers
43
Central Region Western Region
Eastern Region
±
Local Moran's I -0.3 - 0 0 - 0.3 0.3 - 1
0
250
500
1,000
Miles
>1 0
500
1,000
2,000
Kilometers
Figure 2.16 Local Moran’s I in 2000 Second, we can observe from the two maps that Beijing and Tianjin are always the major sources of spatial concentration. Their large positive contribution to global Moran’s I in both 1978 and 2000 can be explained by their specific geopolitical location. As the capital of the nation, Beijing has been the recipient of more favorable policies and more investment from the central government than many other provinces, while Tianjin has traditionally been the gateway to Beijing, and one of the most important industrial bases of China. In addition, the large positive local Moran’s Iis of these two municipalities indicate that Beijing and Tianjin have formed a spatially integrated unit. This unit differentiates from Hebei, which spatially surrounds them but had negative values. However, the contributions of these two municipalities decrease during the study
44 period, in contrast to the increase of Shanghai’s contribution, which in 2000 became the single largest positive contributor to the global index. This, as well as the increasing contributions of other coastal provinces, indicates both an increase of regional inequality and a shift of cluster center from its traditional seat in north China to the Yangtze Delta. Third, Guangdong’s contribution to the global index corresponds with its changing spatial behavior depicted in the Moran scatterplot. In 1978, Guangdong was a spatial unit not significantly different from its neighboring provinces, and contributed a small positive part to the global index. In 2000, however, its fast growth dwarfed most of its neighbors (excluding Hainan) and there was little clustering with its neighbors. It became the biggest negative contributor to the global index. Fourth, while most of the interior provinces have similar patterns, coastal provinces vary greatly. Coastal provinces are located in all four quadrants in the Moran scatterplot in 2000. There are two high-high clusters: Beijing and Tianjin, ShanghaiZhejiang-Jiangsu. The latter group is newly emerged, and their advanced development status not only distinguishes them from their neighbors, but also becomes the primary factor in the rise of the coast-interior gap. Hebei and Hainan (included in Guangdong in calculating the local Moran’s I) are less developed provinces. Fifth, Shandong falls into the low-low quadrant, and contributes very little to the global index. In 2000, its local Moran was only 0.005, although Shandong is among the fastest growing provinces. It is found that Fujian, another fastest growing province, also contributes little to the global index, as in 2000, its local Moran was 0.007. This is due to the fact that fast growth diminishes its gap with the top provinces, and brings convergence among them, which is reflected by the small contribution to the global
45 index. Liaoning, an old industrial base of China, still has relatively high per capita GDP if compared with its neighbors other than Beijing and Tianjin. With slow growth, Liaoning becomes more similar to its neighbors, and its contribution to the global index is small (in 2000, its local Moran’s I is –0.05). Lastly, Guizhou and Yunnan, two provinces in southwest China, contributed relatively highly to the global index in 2000. As shown in Table 1 and 2, they are among the poorest and slowest growing provinces. From 1990 to 2000, per capita GDP growth rate ranked 24th for Yunnan and 28th for Guizhou (Table 2). The combination effect is that they lag far behind both their neighbors and the national average, reflected by their relatively high contribution to the global Moran. This forms a cluster with low per capita GDP in southwest China, which contrasts sharply with the east-coastal cluster, exacerbating regional inequality. 2.5 Understanding Regional Development in China
The above analyses reveal that the pattern of regional development in China changes dramatically during the reform era. The underpinning force driving the change is the reform of the Chinese economy. In addition, geographical factors also serve as an important component in the evolution of regional development in China (Bao et al 2002). To further understand regional development in China, specifically the understanding of development mechanisms in current China, A regression analysis is conducted. Regional growth rate is often used in evaluating the performance of regional development. This section hence uses growth rate of per capita GDP in 1990-2000 as the dependent variable (the reason of using 1990-2000 instead of 1978-2000 is that some of the identified mechanisms have no data prior to 1990). Based on recent literature in
46 economics and geography (e.g., Barro and Sala-I-Martin 1995; Wei and Fan 2000; Bao et al 2002), along with the consideration of the context of China, the following independent variables (mechanisms) are chosen for the regression. (1) Per capita foreign direct investment (FDI) is an important indicator of China’s open door policy and the process of globalization. Because some provinces of China opened in the late 1980s, foreign investment data of those provinces were not available until 1990, which is why we use the data set for 1990~2000 instead of 1978~2000 to do the regression analysis. It is expected that FDI per capita in 1990 will positively affect regional growth. (2) The share of State-Owned Enterprises (SOEs) in a region’s fixed asset investment (FAI) is usually deemed a proxy reflecting the infusion of market mechanisms and local endowments in the regional economy. In the model, the share of SOEs in 1990 is chosen for analysis. Given the poorer returns of SOE investment and declining status of the SOEs, a negative relationship between the share of SOEs and regional growth is expected. (3) Education level (percentage of college or above educated population in the total population) is used to represent labor quality, and a positive relationship with growth is expected. (4) Population growth rate represents the contribution of labor quantity to regional growth. (5) Per capita GDP in 1990, the beginning year of the research, is also used to test whether there is a β-convergence in regional inequality.
47 (6) Lastly, a dummy variable representing the influence of geographical location on regional growth is used, with the coastal provinces defined as 1 and the others 0. Given the advantage of the coastal regions in development, the dummy variable is expected to positively relate to the growth rate. The results of the regression analysis are reported in Table 2.3. The regression model is highly significant as revealed in the table (ANOVA test indicates the model is significant at 1% confident level). The tests for the multi-collinearity among the independent variables and spatial autocorrelation of the residuals suggest that the regression coefficients are reliable and the model is not misspecified7. Table 2.3 Result of regression analysis for the 1990 – 2000 period Unstandardized Standardized Coefficients Coefficients t-value p-value (Beta) B Std. Error Constant 15.543 1.067 14.566 0.000 Coast Dummy 1.569 0.550 0.361 2.852 0.009 Population Growth Rate -0.213 0.311 -0.073 -0.684 0.501 Per capita GDP in 1990 0.001 0.000 0.316 2.039 0.053 FDI per capita in 1990 0.097 0.046 0.305 2.089 0.048 Share of SOEs in FAI in 1990 -0.094 0.015 -0.616 -6.212 0.000 Education Level in 1990 -0.461 0.272 -0.259 -1.695 0.103 Dependent Variable: Average Annual Growth Rate of Per capita GDP from 1990 to 2000 R Square: 0.849 Adjusted R Square: 0.809 Number of Observations: 30 ANOVA test: F = 21.470 (p= 0.000) The two population-related independent variables (population growth rate and education level) are not significant. The variables significant at the 5% confident level 7
The largest variance inflation factor (VIF) for the 6 independent variables in the regression model is 3.65, indicating that collinearity in these two models does not explain more than 10 percent of any independent variable's variance (in which case, VIF is equal to or larger than 10), and that regression coefficients are reliable. The test of the regression residual’s autocorrelation is conducted through the calculation of the Lagarangian Multiplier statistics (Anselin 1988; Anselin and Rey 1991; Anselin 1992), which is distributed as χ2 with one degree of freedom, and is found not significant at 5% confident level.
48 include the coast dummy, share of FAI in SOEs, and per capita FDI in 1990. Per capita GDP in 1990 is not significant at the 5% confident level, although it is at 10%. The six variables together explain 84.9% of the variation of per capita GDP growth during 1990 to 2000. As expected, the growth rate of provinces’ per capita GDP is positively associated with per capita FDI in 1990 and with the coast dummy variable, and negatively associated with the share of FAI in SOEs in 1990. The t-values suggest that the share of FAI in SOEs in 1990 is the most important determinant of regional growth, which reinforces the common wisdom of China’s regional development that provinces depend more heavily on SOEs tend to grow slower (Wei and Fan 2000), and this factor is so far still a very important lagging effect in China’s provincial development. Furthermore, the result also suggests that provinces that receive more foreign investment and are located in the coastal region tend to grow faster. This finding corresponds to the Chinese Government’s coastal development strategy during the reform. The central government launched urban reforms and open door policies first in the coastal provinces, followed by other provinces. The coastal provinces, with their stronger economic bases and closer cultural and economic relationships with foreign investors, have attracted a great amount of foreign investment, especially from Taiwan, Hong Kong, Singapore, North American and West Europe. Consequently, the coastal provinces have recorded a great amount of FDI inflow, and experienced rapid growth of non-state enterprises, which boosts the regional development greatly. The relationship between the per capita GDP growth rate and the beginning year’s per capita GDP is not significant at the 5% confident level but is at the 10% level. The positive sign also suggests that β-convergence doesn’t hold true in China during the
49 studying period from 1978 to 2000. The two labor-related variables are not statistically significant in the model, indicating that orthodox notions of labor quality and quantity are less significant in determining uneven regional development in present-day China. For further verification of this conclusion, I also conducted a regression analysis between the growth rate and education level per se, and again found the regression model is not statistically significant. 2.6 Conclusion
Before the reform, due largely to political and defense considerations, Mao emphasized the development of regions with ideological and strategic significance, such as the leading municipalities of Shanghai and Beijing, and heavy industrial bases such as Liaoning. Consequently, while promoting reducing income and spatial inequalities, new forms of regional inequality were created, leading to the persistence of regional inequality during Mao’s tenure as China’s leader (Wei 2000). Post-Mao China has implemented economic reforms, and undergone profound changes. During the reforms of the late 1970s and 1980s, China opened up four special economic zones in Guangdong and Fujian, and provided favorable policies for the “growth out of plan” of non-state enterprises. These policies, sometimes simplified as the coastal development strategy, led to the emergence of a group of coastal provinces, including Guangdong, Fujian, Zhejiang, Jiangsu, and Shandong. Traditional municipalities and industrial bases, however, were troubled by problematic SOEs, rigid state control, and late opening up to the outside world. Using recently released data, this chapter has examined regional development through the investigation of regional inequality in China with the assistance of GIS and
50 spatial data analysis techniques. A few interesting conclusions can be drawn from the above analyses. First, it is found that regional inequality exhibited a U-shaped pattern during the reform period. As shown by the LQs and Moran’s Is, the status of China’s traditionally leading provinces in the national economy, including Shanghai, Beijing, Tianjin, and Liaoning, declined in the 1980s, which leads to the observation that overall China’s regional inequality declined. However, during the 1990s, especially when Deng Xiaoping toured southern China in 1992 and pressed larger scale market reforms, marketization was furthered and regional inequality increased. Also with the shift of the focus of reform to the Yangtze Delta and a strong industrial base, Shanghai embarks on a dramatic transformation, with radical reforms of SOEs and massive inflow of domestic and foreign investments. Consequently, Shanghai has recorded more rapid economic growth than during the first decade of the reform, while the coastal provinces again spearhead in reform and economic growth. As shown in Table 2.1, these coastal provinces have achieved dramatic economic growth during the 1990s, with Zhejiang ranks first, followed by Fujian, Jiangsu, Shandong, Guangdong, Hebei and Shanghai. Second, through an analysis of location quotient, it is found that the changing relative status of provinces in China is quite coastal-oriented. East and south coastal China have become the biggest winners of economic reforms, while the central provinces have remained relatively unchanged. The status of traditional industrial bases (such as Liaoning and Tianjin) and poorer western provinces declines. The gap between the emerging provinces and the poorest ones has been enlarging. Third, except for the general analysis, by utilizing GIS and spatial data analysis technique, this study also detectes a trend of spatial agglomeration among China’s
51 provinces, although interprovincial inequality declined first and then rose. Underdeveloped provinces retained their spatial similarity during the reform period, while more developed coastal provinces showed different spatial patterns. In particular, north coastal provinces declined, while a group of south and southeastern coastal provinces from Guangdong to Shandong has been rising as the wining cluster. Through an analysis of the values of the local Moran’s I, it is found that Shanghai has played a very important role in China’s regional inequality and spatial clustering process. The decline and rise of Shanghai alone contribute greatly to the declining or enlarging of China’s interprovincial inequality, and the formation of the southeastern China’s spatial cluster. The practice of employing GIS and spatial data analysis in this chapter exhibits promising potential of investigating regional development issues within the GIScience framework. The practice also sets the working environment for following chapters’ analyses. Fourth, through investigation of recent development of economic geography, studies on China’s regional development, a regression model is established to analyzing the impact of reforms on the change of regional development. Regression analysis reveals the significance of SOEs reform, location, and investment variables. This explains that since the reforms were furthered in the 1990s, interprovincial inequality in China has risen, reversing the declining trend in the early years of the reforms. Another finding through the regression analysis is that while theories of regional development emphasize labor and capital, in China, the share of investment commanded by SOEs is the most important factor in affecting regional development. Labor factors, as defined based on educational level in this study, is an insignificant variable. This suggests that country-
52 specific variables, not considered by established theories, are important in understanding China’s transition and spatial changes.
53
Chapter 3 Regional development in Greater Beijing 3.1 Introduction
The previous chapter discussed regional development in China from within a GIS/spatial data analysis framework. This chapter initiates the effort of applying the analysis framework to a specific locale, the Greater Beijing. To begin with, this chapter focuses on univariate analysis for dynamic pattern detection, and detailed spatial agglomeration understanding using data from counties instead of provinces as in the previous chapter. It has been noticed that there is a renewed multidisciplinary interests on regional development theories, primarily regional convergence and divergence (e.g., Barro and Sala-I-Martin 1995; Clark et al 2000). Development in China, although previously escaped western scholar’s attention, now has attracted scholarly interests during the past decades (see Wei 1999 for detailed review). This is largely due to the fact that China has experienced some spectacular success in its economic reform and transition after the end of the Cold War, when other transitional economies such as countries in Eastern Europe have experienced some quite chaotic reform. Professional journal articles on China’s regional development can date back as early as the 1970s, when scholars argued about the increase or decrease of regional inequality in China (for example, Lardy 1975, 1978, 1980; Donnithorne 1976; etc.). However, those pioneering works on China are often impeded by the lack of systematic and reliable data in their analyses. The economic reform in 1978 not only released the development potential of China, but also provided the scholars more amenable environment for further studies of the country. During the
54 1990s and early this century, China issued more and more complete provincial and county level datasets. Consequently, a large body of literature on regional development in China is emerging (Ma and Wei 1996; Zhao 1996; Wei 1999; Wei and Fan 2000; Wei 2000; Fujita and Hu 2001; Gu et al. 2001; Xu and Tan 2002; Wei and Kim 2002; Yu and Wei 2003; Wei and Ye 2004, to name but a few). As revealed in the previous chapter, most scholarly debates on regional divergence, convergence, and inverted-U model tend to agree upon the fact that China’s regional development has strong regional characteristics. None of a single development theory is sufficient in accounting for the fact. In addition, development trajectories of some spatial units, such as Shanghai, Beijing etc., seem to project significant influence on China’s regional development (Yu and Wei 2003). Due largely to data issue, most studies focus on China’s regional development at provincial level, less attention has been paid to studies on county-level. In more recent literature, scholars realize that the underlying reality of spatial transformation and regional development in China cannot be thoroughly understood without the studies on smaller scale spatial units, such as the counties (Wei and Fan 2000; Wei and Kim 2002). A series of publication emerges recently trying to address such concern (for example, Knight and Song 1993; Lyons 1997; Wei and Fan 2000; Huang and Leung 2002; Wei and Ye 2004). In addition, because of data issues and scholarly interests, the southern coastal China provinces, such as Jiangsu, Zhejiang, Guangdong, are usually the studying foci in the literature (for instance, Wei and Fan 2000; Gu et al. 2001; Wei and Kim 2002; Huang and Leung 2003; Wei and Ye 2004, etc.). Systematic research on other provinces or
55 regions of China remains limited. As argued by Wei and Ye (2004), given the massive scale of China (most provinces in China have more population and are larger than most European countries), geographical difference among different spatial units is as vast as the nation itself. A better understanding of regional development in China requires more studies in different locales. Such practices will not only provide better understanding of the regional development in China, but also bring the possibility of theoretical breakthrough in regional development. This chapter, based on the author’s personal working experiences, interviews and fieldwork in Greater Beijing, intends to initiate the steps to fill in this blank. Furthermore, as argued in chapter 2, most of the empirical methods used in the literature on China’s regional development are typically dominated by traditional nonspatial approaches. Spatial effects are more or less ignored. The reasons may be twofold. First, since most of the studies on China are at the provincial level, this higher-level data aggregation may mask subtle spatial effects and negate the usefulness of spatial analysis (Yu and Wei 2003). In fact, the regression analysis in the previous chapter fails to reveal significant residual spatial autocorrelation might be attributed to this scalar reason. Second, convenient and sophisticated research methods for handling spatial data, such as the exploratory spatial data analysis (ESDA), were not readily available. Although software implementation initiated in the early 1990s (Anselin 1992; Anselin and Hudak 1992), statistical and practical issues such as significant tests and large number of data input are only developed very recently (Tiefelsdorf 2000, 2002; Bivand 2004). Nonetheless, studies on county-level spatial units require taking into account spatial effects. As argued by Gallo and Ertur (2003), spatial effects become salient at finer scales,
56 since a number of factors lead to geographically dependent spatial units, such as increasing intensity of trade between spatial units, more frequent technology and knowledge diffusion at relatively small scale, etc. Those factors, if investigate at a larger scale (such as province), may be unobservable or negligible due to their local characteristics, but will turn to be important factors influencing regional development from the view of finer scale. Spatial effects are inherent features in regional development instead of exogenous inputs (Anselin 2001), and therefore should be integrated into the analysis particularly at the county-level. More specifically, two types of spatial effects need to be investigated closely, i.e., spatial autocorrelation and spatial heterogeneity. Spatial autocorrelation is defined as the “coincidence of value similarity with locational similarity” (Anselin 2001). Spatial heterogeneity means that developmental behavior is not stable across space, and may generate characteristic spatial patterns of development under the form of spatial regimes. For example, a cluster of rich regions (the core) is distinguished from a cluster of poor regions (the periphery). To this end, this chapter employs the rapidly growing GIS and recently developed methods of exploratory spatial data analysis (ESDA) to investigate regional development patterns and processes at the county-level in Greater Beijing. Specifically, this chapter investigates the spatial dependence and dynamic spatial process at both global and local level in Greater Beijing’s regional development. More attention is paid to the local patterns of regional development through GIS and ESDA analytical techniques. I contend that GIS and spatial analysis in regional development can reveal significant socioeconomic structures and help identify local development trajectories.
57 In the section that follows, the settings and regional characteristics of Greater Beijing are introduced. The followed section depicts the global spatial pattern and dynamic spatial processes in Greater Beijing at county-level through Moran’s I. Section 4 employs Moran scatterplot and local indicator of spatial association (LISA, Anselin 1995) to analyze the local patterns and spatial non-stationarity. Section 5 concludes the paper with a discussion. 3.2 Greater Beijing: Setting and Development Process
Greater Beijing is located in the northern coastal China, including three provincial level units: Beijing (with 11 county-level units8, or counties for simplification), Tianjin (with 8 counties), and Hebei (with 11 prefectures, 151 counties), totally 170 counties9 (Fig. 3.1). The region has an area of 216,420 square kilometers (approximately 2% of China’s territory), and a total population of 91.6 million in 2001. The region has also been called the Capital Region, or the (Bei)Jing-(Tian)Jin-Ji Region (Wei and Jia 2003), reflecting the close socioeconomic linkages within the region, and the efforts of local governments to promote regional growth and collaboration. Greater Beijing, as well as the Yangtze Delta and the Pearl River Delta, have become the focus of state efforts to transform them into China’s emerging global city regions, and dragon’s head of economic growth (e.g., Li 1999, Wei and Yu 2005).
8
8 small districts in Beijing’s city core, i.e., Haidian, Chaoyang, Fengtai, Shijingshan, Dongcheng, Xicheng, Xuanwu and Chongwen Districts, are considered as one county level spatial unit. 9 Two counties are spatially separated and are treated as four separated spatial units in the analysis. They are Handan city and Luannan county.
58
Neimenggu Chengde
Liaoning Zhangjiakou Qin huan gd
Beijing
Tangshan
Baoding
Langfang
Shanxi
Tianjin
ao
± Prefecture County
Cangzhou Shijiazhuang Hengshui Xingtai
Shandong Handan
0
50
100
Kilometers Figure 3.1 Location of Greater Beijing Beijing has been the most important ancient capital city of China, especially since the thirteenth century, making Greater Beijing China’s most important political center.
59 Shortly after the establishment of the new republic in 1949, massive state investment went to the city to transform Beijing from a semi-feudal, semi-colonial city to a socialist productive city, and people’s new capital (Wei and Yu 2005). Such efforts not only greatly enriched Beijing’s cultural and political assets, but also made Beijing the second most important industrial city in China, surpassing Tianjin, with tremendous buildup in heavy industry. However, despite economic growth under Mao, on the eve of economic reforms, Beijing and its surrounding regions faced severe structural problems fundamental to the socialist economy, even more serious than many other regions under state socialism. First, the product structure dominated by heavy, defense-oriented industry, and the out-of-dated technological equipment, mostly imported from the former Soviet Union in the 1950s, impeded industrial productivity. Second, industrialization in the region was a typical urban industrialization, while rural areas in Hebei remained less developed. Third, linkages among industries and regions in Greater Beijing were weak, leading to the waste of raw materials and duplication of similar industries (Lu 1997). Such an industrial structure, as well as the artificial barriers of the strict household registration system, created an imbalanced core-periphery structure. The core region, represented by Beijing, Tianjin and urban areas in Hebei province, are more developed than their rural peripheries. Such spatial structure, though diminished during the reform, still holds in 2001 (Fig. 3.2).
60
± Per capita GDP 1978
a
71 - 219 220 - 373 374 - 681 682 - 1095 1096 - 1780
0
Per capita GDP 2001
b 100
200
1877 - 4914 4915 - 8199 8200 - 11819 11820 - 16972 16973 - 33462
Kilometers
Figure 3.2 Per capita GDP in Greater Beijing, 1978 (a) and 2001 (b) China’s post-Mao reform is a gradual, experimental, and spatially uneven process. Unlike its southern peers, such as Jiangsu, Zhejiang, and Guangdong etc., the benefit of reforms to this area was limited during the early stages of the reform, and economic status of Greater Beijing was relatively stable (Yu and Wei 2003). This is due largely to the following two reasons. First, The spatially uneven reform policy was spearheaded by special economic zones in south China and later coastal cities. It allowed “specific policies and flexible management” in Guangdong and Fujian provinces in 1979 that are far away from Beijing. The central government treated the reform of the political center, Beijing, as well as Tianjin, with a high precaution. Even after the reform brought
61 dramatic success in Guangdong and Fujian, cautious reform was still a dogma for the government. Tianjin and Qinhuangdao, two of the fourteen open coastal cities (OCCs) designated in 1984, have been slow in reform and have lagged behind other OCCs. Moreover, their trickle down effects to other regions in Greater Beijing remain limited. Reform policy packages were not fully granted to Greater Beijing until after the early 1990s. Second, The dominance of state-owned enterprises (SOEs) and heavy industry in the region limited the speed of economic growth. The long-established industrial and ownership structure in Greater Beijing is quite different from southeastern coastal China, where rapid economic growth has been driven by foreign and non-state enterprises (Wei and Fan 2000). This also indicates the focus, process, and results of the reform are quite different from what was observed in the southeastern provinces. As a result, the reform did not benefit Greater Beijing like it did to the southern provinces. The overall economic status of the three provinces/municipalities in Greater Beijing decreased during the studying period, while their southern peers (represented by Guangdong and Fujian provinces) gained substantially from the reform (Fig. 3.3). Studies in Guangdong and Zhejiang have shown the emergence of new growth centers, such as Wenzhou and Dongguang, stimulated by the development of foreign and private enterprises, are primary local economic development engines (Lin 1997; Wei and Ye 2004). The slow process of reform and the dominance of SOEs in Greater Beijing are very different from its southern peers. Hence a study on this area complements the research on southeastern coastal provinces, and improves our understanding of the process and effects of China’s economic reforms.
62 4 3.5 3 2.5 2 1.5 1 0.5
Beijing
0
Tianjin
Hebei
Fujian
GuangdongYear
2001 2000 1999 1998 1997 1996 1995 1994 1993 1992 1991 1990 1989 1988 1987 1986 1985 1984 1983 1982 1981 1980 1979 1978
Figure 3.3 Changing LQs of provinces in Greater Beijing and southeastern China 3.3 Global Spatial Patterns and Dynamic Spatial Processes in Greater Beijing
In the literature, there are a few statistics employed in detecting global spatial process, especially the spatial autocorrelation, such as the Geary’s C, the Moran’s I (e.g., Cliff and Ord 1973; Anselin 1988; Getis and Ord 1992). Although appeared differently, these two indexes generally represent similar information. Following the analytical framework set in the previous chapter, the study in this chapter is based on the analysis using Moran’s I. The numerical and statistic characteristics of global Moran’s I were intensively discussed in the literature (for instance, see Anselin 1988; Tiefelsdorf 2000, 2002 for extended discussion). In addition, some of Moran’s I’s properties are discussed in the previous chapter. Nonetheless, during the previous chapter, since the primary purpose is
63 to set a GIS and spatial data analysis environment, I adopt a spatial adjacency strategy in determining the spatial linkage instead of discussing in detail the steps and alternatives. For now, however, a more detailed discussion on the spatial weights linkage matrix is called for. In general, two strategies in determining spatial linkage are commonly employed in the literature. In particular, spatial linkages based on border sharing and distance. Either strategy will determine which geographic units will be neighbors of one another, and it will assign 1s to neighbors and 0s otherwise to numerically label the linkage. The border-sharing strategy is actually the simple spatial adjacency example used in the previous chapter. Under such strategy, spatial units are considered as neighbors when they share borders. The relationship can be extracted from a topologically cleaned shapefile from a GIS. This strategy is the most straightforward one, but it bears the shortcoming that it cannot deal with islands. For instance, in the previous chapter, to generate numerically effective spatial linkage matrix, I have to merge Hainan into Guangdong province. The linkage matrix based on distance is more abstract. In a GIS framework, geographical units can be epitomized by their geographical center (for instance, such as a centroid of a polygon) in space. Hence geographical units that fall within certain distances (usually Euclidean distance) from one another are deemed neighbors. This strategy can yield effective spatial linkage matrix as long as the distance chosen is reasonable. However, how “reasonable” is reasonable still remains a question for further investigation. In addition, spatial linkage matrix produced from distance-based strategy can become dense as the distance increases, hence might lost subtle local information.
64 Indeed, due to the complexity of interactions among geographic units (Anselin 1988), it is appropriate to examine alternative spatial weight matrixes. This chapter hence constructs spatial weight matrixes based on both border sharing and distance strategies. For distance strategies, 4 different alternatives, namely, geographic units within 80 km, 160 km, 240 km, and 320 km, are considered as spatially linked neighbors for each alternative. The choice of 80 km is based on the field experience that this is the distance for approximately an hour drive on the expressways. The characteristics of the five spatial weight matrixes are reported in table 3.1. Table 3.1Characteristics of alternative spatial weight matrixes Characteristics BDSH* Dist80 Dist160 Dist240 Dist320 Number of regions 170 170 170 170 170 Number of non-zero links 922 3028 9518 16066 21340 Percentage of non-zero weights 3.19 10.48 32.93 55.59 73.84 Average number of links 5.42 17.81 55.99 94.51 125.53 *: BDSH stands for border-sharing; Dist80 stands for the 80 km weighting strategy, and so forth. The computation in this study still employs the SPDEP package (Bivand 2002, 2004) in R. While using different weighting strategies, it is found that in Greater Beijing the spatial autocorrelation decreases as the distance increases. However, except for the 240km and 320km strategies before 1990, all other strategies yield significant Moran’s I at (at least) 95% confidence level by saddlepoint approximation (Table 3.2). From Table 3.2, it is also noticeable that the significance of Moran’s I decreases as the average number of neighbors in the spatial weight matrix increases (from column 2 to column 5 in Table 3.2). This indicates that development of counties is spatially autocorrelated, and moreover, the autocorrelation has strong local characteristics.
65 Table 3.2 Significance test (p-values) of Moran’s I for different weighting strategies Year BDSH Dist80 Dist160 Dist240 Dist320 1978 0.003* 0.015 0.006 0.080 0.201 1980 0.000 0.001 0.001 0.030 0.137 1985 0.001 0.002 0.004 0.543 0.978 1990 0.000 0.001 0.000 0.019 0.041 1995 0.000 0.000 0.000 0.000 0.001 1996 0.000 0.000 0.000 0.001 0.003 1997 0.000 0.000 0.000 0.003 0.006 1998 0.000 0.000 0.000 0.001 0.003 1999 0.000 0.000 0.000 0.001 0.004 2000 0.000 0.000 0.000 0.001 0.005 2001 0.000 0.000 0.000 0.000 0.003 * p-values are obtained through saddlepoint approximation (Tiefelsdorf 2000). For dynamic spatial autocorrelation, all strategies yield very similar changing patterns from 1978 – 2001. Here I only depict the patterns of Moran’s I under the bordersharing strategy since it gives the strongest autocorrelation calculation (Fig. 3.4). Fig. 3.4 displays the dynamic changing process of the global Moran’s I statistic of per capita GDP over the period 1978-2001 (data for the year 1979, 1981-1984, 1986-1989, and 19911994 at county-level are not available) for the 170 county-level spatial units in Greater Beijing. The significant test of the statistics is again conducted through the saddlepoint approximation (Tiefelsdorf 2000, 2002). It appears that per capita GDP among the counties during the studying period are positively spatially autocorrelated since all the pvalues are significant at 1% confident level. In addition, the dramatic increase of the global statistics suggests two findings: first, the distribution of per capita GDP is by nature clustered over the counties in Greater Beijing during the studying period instead of purely random. Similar values of per capital GDP are clustered spatially during the studying period in Greater Beijing. However, which type of similarity, i.e., high values or
66 low values, cannot be told from the global index. Second, such clustering process is strengthening as China’s economic reform proceeded. Such strengthening of spatial clustering process contains two possibilities. On one hand, the strengthening could be due to the counties in each cluster becoming more similar in per capita GDP. On the other hand, it could also be due to the emerging of newly formed clusters. Global Moran’s I, which is by nature a global measure of spatial dependence, does not allow us to distinguish between these possibilities. For that I turn to a more disaggregated view of the
Global Moran's I
structure of spatial dependence in Greater Beijing’s per capita GDP. 0.45 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0
Year 2001 2000 1999 1998 1997 1996 1995 1994 1993 1992 1991 1990 1989 1988 1987 1986 1985 1984 1983 1982 1981 1980 1979 1978
Figure 3.4 Changing global Moran’s I in Greater Beijing, 1978-2001 3.4 Local Spatial Patterns and Non-stationarity 3.4.1 Moran scatterplot and statistical characteristics of local Moran’s Ii
As discussed and addressed above, the global Moran’s I does not allow us to assess the specific spatial structure of spatial autocorrelation. In regional studies, it is quite common to ask whether there are local spatial clusters of high or low values, which regions contribute how much to the global spatial process, and to what extent the global
67 evaluation of spatial autocorrelation masks atypical localizations or “pockets of local non-stationarity” (Gallo and Ertur 2003). In addition, the types of clustering and the strengthening of the global spatial process can be better answered by decomposed local analysis. In this respect, two tools are employed to address the issues, i.e., the Moran scatterplot and local Moran’s I. The Moran scatterplot is a graphic tool derived from the global Moran’s I. As argued by Anselin (1996), since the global Moran’s I is formally equivalent to a regression coefficient in a regression of the spatial lag on the measuring variable, a scatterplot between these two would yield useful information on local non-stationarity. In the scatterplot, the outliers and/or leverages will be represented as extreme points with respect to the central tendency reflected by the regression slope (the global Moran’s I). The characteristics of the four quadrants of the scatterplot and the visualization properties are discussed in chapter 2, and will not be repeated here. Apart from the Moran’s scatterplot and maps, the local Moran’s Ii can also be used to assess the local non-stationarity under significant global spatial processes. In the pervious chapter, I have discussed in detail the analysis of the individual values of the local Moran’s I, which is used as an indication of specific spatial units’ contribution to the global spatial process. However, the values of local Moran’s Iis cannot tell whether their contributions are significant or not. This chapter is more interested in the statistical attributes of the local Moran’s I instead of its values for detecting local spatial structure and regimes. Significant test in the local case can also be conducted through the saddlepoint approximation, though more complicated than in the global Moran’s I’s case (detailed technique discussion can be found at Tiefelsdorf 2000).
68 Similar to the global Moran’s I, local Moran’s Ii does not follow a normal distribution (Tiefelsdorf 2002). In addition, the distribution of the local Moran’s Ii is conditional to the distribution of the global Moran’s I. Fortunately, since both the global and local Moran’s I can be expressed as a ratio of quadratic forms (Tiefelsdorf 2000, 2002; Leung et al 2003), the exact distribution can be obtained by applying the Imhof (1961) formula (detailed discussion can be found at Lieberman 1994; Tiefelsdorf 2000, chapter 5 and 6; Leung et al 2003). However, obtaining the exact distribution of local Moran’s Ii is usually very computationally intense, especially when the data sets are large. The approximation methods are developed in the literature for computational benefit (Tiefelsdorf 2000, 2002; Leung et al 2003). Again it is found that the saddlepoint approximation (Lieberman 1994; Tiefelsdorf 2000, 2002) generates relatively accurate approximation for the local Moran’s Ii’s distribution (technique discussions can be obtained from Lieberman 1994; Tiefelsdorf 2000, 2002). Moreover, saddlepoint approximation can deal with the local Moran’s I’s null distribution conditioning on the underlying global spatial process (Tiefelsdorf 2000). The methods and computation are fully implemented in the SPDEP package in R (Bivand 2004). 3.4.2 Empirical analysis of Greater Beijing’s local pattern, 1978 - 2001
For consideration of conciseness and generality, this paper only presents the Moran scatterplot, Moran scatterplot maps and the result of local Moran’s Ii in Greater Beijing for the year 1978 and 2001, the initial and final years of this study. The Moran scatterplots (Fig. 3.5 and 3.6), corresponding Moran scatterplot maps (Fig. 3.7) and the local spatial pattern maps (Fig. 3.8) are from Fig. 3.5 to 3.8. The scatterplots are again created using R and its spatial package SPDEP. The superimposed LOcally Weighted
69 regrESSion (LOWESS) curve lines use the standardized per capita GDP and its spatial lags (standardized as well) as inputting coordinates. Two thirds of the spatial units are used as the span for the locally smoothing process (the trend can be detected using span from 20% to 80% of the spatial units, see Cleveland 1979 for a detailed discussion). To clearly show the leverage and influential points, a vertical line in each scatterplot at x = 2 (standardized per capital GDP = 2) is added as well. Points beyond this vertical line represent high-leverage spatial units and are automatically labeled by the SPDEP package. In addition, some of the low-leverage points also impose relatively high influence on the overall spatial pattern, which is labeled as well. These influential spatial units are potential locations that usually determine the spatial structure of the studying region. All the Moran scatterplot maps and the related local spatial pattern maps are created in ArcGIS 9.0® in conjunction with R. In the local spatial pattern maps, the hot spots are spatial units that have a p-value10 that is less than 0.05 conditioning on the corresponding year’s global spatial process. It represents significant local non-stationarity. The spatial units are significant local heterogeneities in the global spatial process. Clusters, on the other hand, are spatial units that have a p-value larger than 0.95, which represents positive local autocorrelation conditioning on the global spatial process (Tiefelsdorf 2000, p134).
10
A spatial unit’s p-value is calculated through the saddlepoint approximation employing BarndorffNielsen formula. p = Pr ( Ii < Ii0 | Ω), where Ii0 is the observed local Moran’s I of spatial unit i, Ω is the level of global spatial process, it can be obtained through a slope-only spatial regression.
70
Moran Scatterplot, 1978 Tanggu Changping
Shunyi
Beijing Sanhe Qing Xian
0.0
0.5
Tianjin
Qinhuangdao Shi Tangshan Shi Shijiazhuang Shi Xingtai Shi Handan Cangzhou Handan Shi1 Shi2 Shi Zhangjiakou Shi Baoding Shi
-0.5
Spatial Lags (Standardized)
1.0
Wuqing
-1
0
1
2
3
GDP per capita in 1978 (Standardized)
Figure 3.5 Moran scatterplot of Greater Beijing, 1978
4
71
Moran Scatterplot, 2001 2.0
Tongzhou
Fengnan
1.0
1.5
Wuqing Tanggu Fengrun Beijing
Shijiazhuang Shi
0.5
Tangshan Shi
Tianjin
0.0
Luquan
-0.5
Qinhuangdao Shi
Chengde Shi Zhangjiakou Shi
-1.0
Spatial Lags (Standardized)
Shunyi Changping
-1
0
1
2
3
GDP per capita in 2001 (Standardized)
Figure 3.6 Moran scatterplot in Greater Beijing 2001
4
5
72
± Moran scatterplot 2001
Moran scatterplot 1978
a
High-High (I-21) Low-High (II-32) Low-Low (III-102) High-Low (IV-15)
b
0
100
High-High (I-48) Low-High (II-14) Low-Low (III-85) High-Low (IV-23)
200
Kilometers
Figure 3.7 Moran scatterplot maps in 1978 (a), 2001 (b)
± Local pocket map 2001
Local pocket map 1978
a
Hot Spots Non-significant Clusters 0
b 100
200
Hot Spots Non-significant Clusters
Kilometers
Figure 3.8 Local spatial pattern maps, 1978 (a), 2001 (b)
73 An investigation of the figures and maps leads to a series of findings. Firstly, local patterns of spatial association reinforce the global trend of positive spatial association reported earlier. More specifically, around three quarters of the counties in the two years fall within either the first or the third quadrant in the Moran scatterplot and the number continues to increase (1978 – 72.4%, 2001 – 78.2%, Fig. 3.5-3.7). Moreover, low value clusters seem to dominate the process, although high value clusters are increasing. In 1978, among the counties representing negative spatial autocorrelation, spatial units in the second quadrant were more than twice that in the fourth quadrant (Fig. 3.7a). This suggests that at the beginning of the reform, although counties with similar per capita GDP tend to cluster together, the gap between rich and poor counties was large and widely distributed (Fig. 3.7a), especially around the Beijing-Tianjin core. With reform, however, the counties surrounding the core regions seemed to catch up, since in 2001, counties in the second quadrant decreased to 14 while that in the fourth quadrant increased to 23 (Fig. 3.7b). A comparison of the two Moran scatterplot maps (Fig. 3.7) reveals that a spreading trend from the core stands out. The HH region extended from the traditional Beijing-Tianjin to Beijing-Tianjin-Tangshan-Langfang, indicating that the strengthening of the global spatial dependence is largely due to the newly formed cluster. Indeed, except for the extending of the core region, a newly formed HH cluster around the Hebei’s capital city, Shijiazhuang, also stands out in the Moran scatterplot map in 2001 (Fig. 3.7b). Secondly, the superimposed LOWESS curve line became closer to the regression line from 1978 to 200111. A sudden dip in the LOWESS curve in the 1978 Moran
11
A series of Moran scatterplots with superimposed LOWESS curve were created but not reported here for the purpose of conciseness, and a trend of smoothing LOWESS curve was observed as well.
74 scatterplot indicates a shift from positive to negative autocorrelation (Fig. 3.6), which contributes to the lower value of the global Moran’s I (Fig. 3.4). This also indicates that though globally Greater Beijing showed a positive spatial association, negative local spatial autocorrelation existed at the beginning of the reform. This is more apparent in the local pocket map (Fig. 3.8a). In Fig. 3.7a, all the counties in the fourth quadrant of the scatterplot (Fig. 3.5) are significantly different from their neighbors, and are marked as “hot spots”. However, such local regimes disappeared in 2001, and the LOWESS curve line is much smoother (Fig. 3.6). The number of potential leverage and influential units in the fourth quadrant decreased. Indeed, some of the 1978 leverage “fourth quadrant” spatial units moved to the first quadrant, such as Shijiazhuang and Tangshan. In addition, the 2001 local pocket map shows less “hot spots” but more “clusters” (Fig. 3.8b), indicating the decrease of local non-stationarity during the study period, which reinforces the above finding that the strengthening of the global spatial process is likely due to the newly formed/expanded clusters, and overall the spatial structure of Greater Beijing becomes more uniform. Thirdly, in 1978, all the leverage points in the fourth quadrant, and also the “hot spots” in the local pocket map, i.e., Shijiazhuang, Baoding, Cangzhou, Handan, Tangshan, Xintai, Zhangjiakou and Qinhuangdao, are prefecture capitals in Hebei (Fig. 3.1). Although classified as counties, they are more urbanized than their county-peers. Such local heterogeneities reveal a significant divide between urban and rural areas in the region existing in the pre-reform period. Mao’s industrialization was urban-centered, while rural areas focused on agriculture with little industry and were resource suppliers for the urban areas. Such a division of labor created a core-periphery structure between
75 the more developed urban areas and less developed rural areas. Local heterogeneities of both the scatterplot and local pocket map identified such a two-tier structure. The two-tier structure was shattered due to problematic SOEs in the cities, and the development of rural areas through the development of non-state enterprises, removing urban-rural barriers of trade, and encouraging rural migration to urban areas. In 2001’s Moran scatterplot and the local pocket map, the significant differences between the prefecture capitals and their inland peers lessened in most regions (Fig. 3.6, 3.8b). Shijiazhuang even turned from a “hot spot” to a “cluster” center (Fig. 3.8b). Lastly, while urban-rural divide lessens, a north-south divide is emerging. In 1978, the significant urban-rural divide was distributed all around Greater Beijing. In 2001, the divide in the southern regions (south of Beijing-Tianjin) became spatially insignificant, while such divide remained significant in the northern regions (Fig. 3.8b). In addition, the two Moran scatterplot maps in 1978 and 2001 (Fig. 3.7) show that counties around Shijiazhuang formed a new HH cluster in 2001, which indicates as the provincial city of Hebei, Shijiazhuang and its vicinity caught up with the traditional Beijing-Tianjin core. Some southern counties even moved from the initial LL cluster to the HL cluster, while the spatial structure of the northern region remained relatively stable. Moreover, the extending of the core is solely eastwards and southwards, excluding northern counties. Such a north-south divide can be attributed to both natural conditions and development trajectories of the counties. North Hebei, which includes the prefectures of Zhangjiakou and Chengde, is part of the Bashang plateau, with mountainous topography, which constrained the development of economy and transportation, especially compared with their southern peers that belong to the North China Plain. Although some military
76 industries were moved to Zhangjiakou during Mao’s era, policy failure and geographical constraints limited the effectiveness of industrialization. In addition, historically Zhangjiakou and Chengde were deemed the northern gates of Beijing, both militarily and ecologically, which discourages drastic industrialization. Meanwhile, Beijing took advantage of such a position by enjoying the external economy of its gates without much positive trickle down effect, further preventing the development of the northern regions. 3.5 Conclusion and Discussion
Using the newly developed GIS and ESDA tools to analyze regional development in Greater Beijing, this chapter highlights the importance of spatial effects in regional studies. This research stands at the forefront of trying to integrate GIS and ESDA in China’s regional development research at county level. It appears that GIS and ESDA can finely reveal the spatial characteristics of regional development that other tools are incapable of. Through the study, an overall picture of spatial pattern and dynamic process in Greater Beijing at county-level is present. The analyses reveal significant positive global spatial autocorrelation in Greater Beijing, which is dominated by low value clusters, but high value clusters are enlarging. In the decomposed local analysis, the study finds that the strengthening of spatial autocorrelation is likely a result of the extension of original cluster and new cluster formation. Moreover, the spatial units’ changing trajectory indicates that the economic reform projects positive influence on Greater Beijing’s regional development, since the extended and new-formed clusters are primarily high value clusters. In addition, the findings from the study support the hypothesis that spatial effects take stronger role in finer geographical scale regional development. This conclusion
77 corroborates the findings proposed by Wei and Fan (2000) that finer geographical scale (county-level in the China’s study context) tends to include richer information than larger scales (such as provincial or regional). For example, in the previous chapter, the calculated global Moran’s I based on the provincial aggregated dataset are much smaller than the global Moran’s I obtain in this chapter using county-level spatial dataset. The result here also agrees with the study of Wei and Ye’s (2004) on the county-level study in Zhejiang province. Moreover, the spatial structure exploration at the county-level reinforces the common wisdom that there was a two-tier spatial structure (Eryuan Jiegou) between the urban and rural areas at the beginning of the study period (Lu 1997). Dynamic pattern observation reveals, however, such two-tier structure is disappearing when China’s economic reform deepens. Apart from the urban-rural two-tier structure, the local spatial patterns also reveal a north-south divide in Greater Beijing during the studying period. Unlike the disappearing urban-rural two-tier spatial structure, the north-south divide emerges during the reform. That is, due to natural conditions but more importantly the development trajectories and their specific geopolitical position, northern regions in Greater Beijing lagged behind their southern peers in development. This finding indicates a very unique regional development pattern in Greater Beijing. The specific geopolitical position of Beijing as the nation’s capital enables it to enjoy the external economy of its two northern gates – two prefectures of Hebei province, Zhangjiakou and Chengde – as both safeguards and ecological screen. With insufficient trickle down effects, these two prefectures are artificially put in an unfavorable position in development.
78 The empirical strategy of this study is based on the newly proposed ESDA approaches in conjunction of GIScience, especially the Moran scatterplot and saddlepoint approximation for the Moran’s I’s distribution under the null hypothesis (no spatial autocorrelation). The conjunction of Moran scatterplot and saddlepoint approximationidentified significant “hot-spots” and/or “clusters” in this study, which successfully detects significant local regimes conditioning on the underlying global spatial process. However, as pointed by Tiefelsdorf (2000, p136), the significance of the single local Moran’s I cannot be interpreted independently from each other because they are mutually correlated. In the literature, a multi-comparison Bonferroni correction was usually applied (for instance, see Anselin 1995; Gallo and Ertur 2003). It is also noticed that the Bonferroni correction is easily overly conservative, especially for spatial units with many neighbors. In this study any correction is avoided and the correlated p-values for local regime detection are applied with caution. That means that interpretation of significant spatial unit is based on the spatial units and their neighbors. For instance, a significant “hot spot” is the spatial unit that significantly different from its neighbors, while at the same time, its neighbors are also significantly different from it, although no significant pvalues for those neighbors will be obtained (this is also due to the mutual correlation with other neighbors). The similar procedure applies to the interpretation of significant “clusters”. This study made an attempt to integrate the ESDA tools into the greater framework of GIScience. Through this study, I intend to express such a message that GIScience concerns effectively manipulating geographical/spatial information through various statistical, graphical, analytical and/or numerical approaches. Nonetheless, this
79 chapter does not explore further into the mechanisms of regional development at county level since the analysis is univariate by nature. Multivariate analysis such as regression and regional econometric models can be constructed to fulfill such task. Furthermore, as this study has already revealed significant spatial effects in regional development, ordinary econometric models, such as the least squares regression models, should be avoided and appropriate spatial specification (such as spatial lag or error models) is needed for proper analysis of regional development mechanisms. To delve further into the underpinning driving forces behind the local regimes, a geographically weighted regression model (Fotheringham et al. 2002) can be employed as well. In the context of China and specifically Greater Beijing, these aspects have not yet been fully examined in the empirical literature and will be the research foci for the next two chapters.
80
Chapter 4 Development Mechanisms in Greater Beijing – A Spatial Econometric Perspective 4.1 Introduction
During the past decades, the world has witnessed a fast growing China. Numerous studies using data from a variety of sources have engaged in the understanding of the underlying forces of China’s rapid growth (Fan 1995, 1997; Wei and Ma 1996; Lin 1997; Wei 1999, 2000, 2002; Wei and Fan 2000; Wei and Kim 2002; Ying 2003; Yu and Wei 2003; Wei and Ye 2004). These studies of development mechanisms result in a series of insightful yet controversial conclusions (Woo 1999; Rawski 1999; Wei 2000, 2002). In general, Wei (2000, 2002) summarizes that three forces are driving the fast growth in China during the reform period, namely, globalization, marketization and decentralization. Other factors, from a neoclassical standpoint, such as labor mobility, increase in capital stock, are also identified in the literature (Ying 2003). The controversial nature of the conclusions might be expected, as the angles of view, sources of data, geographic scales, and study regions are very diversified in the literature. Largely due to data availability, most of the studies intending to understand the mechanisms of China’s regional development focus either on the provincial level (such as Lu and Wang 2002; Ying 2003; Yu and Wei 2003), or on coastal provinces in the southeastern China (such as Lin 1997; Wei 2000; Wei and Fan 2000; Gu et al. 2001; Wei and Kim 2002; Wei and Ye 2004). Mechanism studies addressing sub-provincial scale or geographic locales in the northern and inland China remain limited (Wei and Fan 2000).
81 Furthermore, studies on development mechanisms follow a traditional non-spatial route. Ordinary least squares (OLS) regression technique is employed widely in the literature (for instance, see Barro and Sala-I-Martin 1990, 1995; Wei 2000; Wei and Fan 2000; Wei and Kim 2002; Yu and Wei 2003). Nonetheless, as argued by Anselin (1988), when OLS estimator is used for cross-sectional data on geographic units, the existence of spatial autocorrelation among these geographic units can pose serious problems of model misspecification. Recent efforts of employing spatial econometric methodology in understanding China’s development mechanisms are arising (Ying 2003). However, the study is still limited at the provincial level. As argued by Wei and Fan (2000), the underlying reality of spatial transformation and regional development in China cannot be thoroughly understood without the studies on smaller scale spatial units, such as the counties. This is especially true when spatial effects need to be taken into account. Since disaggregated spatial units tend to reveal more spatial interactions, and the discrepancies between the boundaries of the developing process and the spatial units under which the cross-sectional data is collected tend to be large. It is to this respect that this chapter intends to initiate an effort to employ the spatial econometric methodology in understanding China’s regional development mechanisms at sub-provincial level. Again, data from Greater Beijing are used in this study. In so doing, I hope that more localized developing characteristics that are different from the often-studied southern provinces might present. The rest of the chapter is organized as follows. Section 2 will discuss the possible developing mechanisms in Greater Beijing based on fieldwork, working experiences and local interviews. This is followed by an introduction and description of the spatial
82 econometrical methodology in section 3. Results and discussion based on both the OLS estimator and the spatial regression models are detailed in section 4. Section 5 concludes the paper with final remarks. 4.2 Development in the Greater Beijing: State, Globalization, and Marketization
Scholars argue that there are general laws governing the process of economic development (Chaudhuri 1989; Barro and Sala-I-Martin 1995). For instance, the neoclassical growth theory emphasizes capital and labor mobility (Solow 1956; Lipshitz 1992). In the case of regional development, however, these universal laws are usually subject to regional constraints. In particular, in China, scholars often argue that the orthodox neoclassical growth theory does not apply to the current development reality well (Wei 2000, 2002; Yu and Wei 2003). One of the major reasons is that the orthodox neoclassical theory is developed based on experiences from the well-marketized western countries, where production factors (labor and capital) are free to move, and market is the primary force in allocating resources. However, even after the socialism-market economy system has been established in China for more than ten years, the state still plays a significant role in China’s regional development (Wei 2000; Wei and Fan 2000). As one of the nation’s heavy industrial bases in the planned economy period (before and during the early stage of the reform), regional development in Greater Beijing was particularly influenced by the state’s macropolicies. In addition, also because of Greater Beijing’s long-term concentration on heavy industrial development, some often-studied marketization factors, such as the reform on the state-owned enterprises (Wei 2000; Yu and Wei 2003), might present different relationship with development in this area. In order to understand the underlying development mechanisms in Greater Beijing, I
83 undertook two summer field trips to the area (2002 and 2003, respectively, for 6 weeks), and managed to interview a few local officials and scholars. Based on these field trips in conjunction with the review of recent economic geography literature and development studies on China (Barro and Sala-I-Martin 1995; Fan 1995; Lin 1997; Yu 1999; Yu and Mao 1999; Clark et al. 2000; Wei 2000; Fujita and Hu 2001; Hill 2002; Wei 2004), I choose the following possible development factors to construct a model for mechanism study. (1) Per capita fixed asset investment (FIXINV), as often used by many scholars, is elected as the primary factor for input of regional development. Regional allocation of fixed investment is considered a key instrument in China’s industrialization and regional development policy, and fixed asset is considered as a major factor of growth in the literature and in the Chinese economy (Ma and Wei 1997; Wei and Kim 2002). FIXINV is expected to positively contribute to regional development. (2) An important aspect of China’s economic reform is the open-door policy, or strategies for developing external-oriented economies (Yu 1999; Yu and Mao 1999), which has resulted in the infusion of foreign direct investment (FDI) and the opening up of the domestic economy. Overwhelming evidence has shown the significant contribution of FDI in China’s regional development, especially in coastal China (Wei 2000; Fujita and Hu 2001). Although Greater Beijing lagged behind other coastal regions in implementing economic reforms, the infusion of FDI to its open coastal cities in Tianjin and Qinhuangdao suggests the potential significance of FDI on the development of Greater Beijing. Per capita FDI (FDIPC) is elected as a proxy for the open door policy
84 and the effects of globalization, and is expected to be positively related with regional development. (3) My fieldwork in Greater Beijing revealed that SOEs still play somewhat important role in regional development in this area, which is very different from southern provinces like Zhejiang (Wei and Ye 2004), where SOEs are not playing important roles in local development. There is a consensus among scholars that SOEs generate negative influences on development (Wei and Kim 2002; Yu and Wei 2003). The percentage SOE investment in total investment (SOEPCT) is utilized to examine the effects of SOEs on regional development. (4) Decentralization has provided local governments increased financial power and more resources to support local development. However, the role of local government spending has been less studied. I include per capita local budgetary expenditure (FINEXP) in the hope to capture the role of decentralization as well as local governments in regional development, which is also expected to be positively related to regional development. (5) In addition, as noticed by scholars (Yu 1999; Yu and Mao 1999; Wei and Fan 2000, Yu and Wei 2003), and also suggested by the pattern analysis in the previous chapter, spatial agglomeration, particularly the urban-rural divide, may have important influence on regional development in Greater Beijing. Urban areas tend to be more developed than their rural peers, and urbanization stimulates regional development. Urbanization (URB), represented by the percentage of urban population in the total population, is employed to investigate the effect of urbanization on regional development.
85 4.3 Model Specification and Spatial Econometric Methodology 4.3.1 Model specification
Growth rate of per capita GDP is the most often used proxy for regional development in understanding development mechanisms in China and other locales (see for instance, Barro and Sala-I-Martin 1991, 1995; Rey and Montouri 1999; Wei and Fan 2000; Wei and Kim 2002; Ying 2003; Yu and Wei 2003). A model establishing relationships between the growth rate and other identified mechanisms usually takes a linear form. Such model specification has the straightforward appealing that model coefficients and the t-values can be interpreted as the relative importance of the mechanism in question. Unfortunately, although through the interview and field trips, I successfully obtained two sets of county-level data for 1995 and 2001 from the statistical bureaus, they were recorded in current prices instead of comparable prices. This makes calculation of growth rate highly risky since the price adjustment factor cannot be taken into account. To mend this problem, I attempt another model specification instead of the orthodox linear form. The general idea of a production function is borrowed to introduce this model specification. In short, a production function formally expresses the output of an economic system as the product of its input factors, namely, labor and capital: Y = A*Lα*Cβ
where Y is the economic system’s output, and L and C are labor and capital, respectively, and α and β are parameters need estimating. Following the idea of the production function, and considering the limitation of the data, I treat the individual year’s per capita GDP as akin to the output of Greater Beijing’s economic system, while
86 the identified mechanisms, namely, FIXINV, FDIPC, SOEPCT, FINEXP and URB, are treated as the inputs. Hence a production-function-like regional development mechanism model can be specified as: GDPPC = A*FIXINVβ1*FDIPCβ2*SOEPCTβ3*FINEXPβ4*URBβ5
(4.1)
Although (4.1) takes an exponential form, it can be transformed into linear form through logarithm transformation. This practice results in the familiar linear model: GDPPC* = β0+β1 ∗ FDIPC*+β2 ∗ FIXINV*+β3 ∗ SOEPCT*+β4 ∗ FINEXP*+β5 ∗ URB*
or in matrix form: y = βX + ε
(4.2)
where y is the logarithm transformed GDPPC (GDPPC*), X is the matrix containing the five identified development mechanisms in their logarithm transformed forms (FDIPC*, FIXINV*, SOEPCT*, FINEXP* and URB*) and a constant term, β is the vector of model coefficients and ε is the vector of unobservable noise. 4.3.2 Spatial econometric methodology
In the literature, usually model specified as (4.2) is calibrated using the ordinary least squares (OLS) estimator (Barro and Sala-I-Martin 1991; Wei and Fan 2000; Wei and Kim 2002; Yu and Wei 2003). Under the linearity, normality, homoskedasticity (independently identically distributed error terms) assumptions, the OLS is a BLUE (best, least-squares, unbiased and efficient) estimator. However, when dealing with crosssectional data on geographic units, existence of spatial dependence (either in the dependent variable or the error term) violates the basic assumptions for the OLS estimator. Hence employing OLS estimator in the analysis might lead to misleading
87 model interpretations when there is significant spatial dependence (Anselin 1988; Anselin and Rey 1991; Rey and Montouri 1999;Ying 2003). There are two types of alternatives that incorporate the spatial dependence in the model explicitly (see Anselin 1988; Anselin and Rey 1991; Anselin and Hudak 1992; Smirnov and Anselin 2001 for detailed discussions). They represent two closely related but different spatial effects. The first is the dependence in the spatially lagged dependent variable (akin to the time series autocorrelation), which is referred to by Anselin and Rey (1991) as “substantive dependence”. This type of dependence suggests spatial spillover is dominant in the development. The second is the dependence in the regression’s error term, or the “nuisance” dependence by Anselin and Rey (1991). It is more likely a result from the mismatch between the boundaries of the spatial process and data collection units. The substantive dependence model can be expressed as: y = ρWy + βX + ε
(3)
where W is a spatial weight matrix describing the spatial linkage among spatial units (this is the same as in calculating the Moran’s I, and following the common practice, the weight matrix is row-standardized so that elements in each row will sum up to 1); Wy is hence a so-called spatially lagged dependent variable (the same as in obtaining the Moran’s scatterplot); ρ is the coefficient of the spatially lagged dependent variable. The nuisance dependence model usually takes a spatially autoregressive error form12, which is: y = βX + ε A moving average process form such that the spatial lag pertains to the new error term μ instead of the original error term ε is also possible. However, as pointed out by Anselin (1988, 1992), the tests for both error models turn out to be the same. In this chapter, I therefore focus on the spatial autoregressive error model only. 12
88
ε = λWε + μ
(4)
where now the spatial dependence is in the error term (that’s why it is called nuisance), μ is a well-behaved (i.i.d.) error. λ is the spatial autoregressive coefficient of the error, and all other symbols are defined as above. To test the spatial dependence in the models, two Lagrange Multiplier (LM) robust diagnostics for both models can be calculated from the data and residuals of the OLS estimator (detailed technique discussion can be found in Anselin 1988, 1992; Anselin and Rey 1991). The robust LM tests are χ2 distributed with one degree of freedom (Anselin and Rey 1991; Anselin 1992). As pointed out by Anselin and Rey (1991), in addition to providing diagnostic for the existence of spatial dependence in the models, the comparison between the two LM diagnostics can also be used as a guidance to choose the better alternative model. In essence, the larger the significant LM statistic, the better the alternative. Because of the inclusion of spatially lagged dependent variable or error term, model (3) and (4) cannot be calibrated using OLS since the result is either biased (substantive model) or inefficient (nuisance model) (Anselin 1988). Instead, maximum likelihood estimator is suggested as an alternative (see Anselin 1988, Chapter 6 for an extended discussion). As a result, the OLS’s goodness-of-fit, R2, which is based on the decomposition of the total sum of squares, is no longer applicable. Likelihood function based goodness-of-fit statistics, mainly log-likelihood and Akaike Information Criterion (AIC), are used to measure the spatial model’s goodness-of-fit (Anselin 1992). Moreover, these statistics are directly comparable to their OLS estimator’s. The model with the highest log likelihood or lowest AIC is considered as the better model (Anselin 1992).
89 Test of the spatial autoregressive coefficients (either ρ or λ) can be done through the likelihood ratio test, which is computed as twice the difference between the log likelihood of the spatial autoregressive models and the linear model (in which the null hypothesis of no spatial dependence holds). The likelihood ratio is also χ2 distributed with one degree of freedom (Anselin 1988; 1996). 4.4 Results and Discussion
This section proceeds as follow. I first examine model (4.2) by means of the OLS estimator (for data from years 1995 and 2001, for temporal comparison purpose). Then based on the OLS estimation and the LM tests on spatial dependence, I present the results of the appropriate alternative spatial model. All model calibration and test statistics are carried out by means of the SPDEP package (Bivand 2004) in R (R Development Core Team 2004) or coded by the author according to the instructions in Anselin (1992). The OLS results using data from 1995 and 2001 are reported in Table 4.1. The corresponding LM tests for both substantive and nuisance spatial dependence under the different weighting strategies are reported in Table 4.2. Table 4.1 OLS results for 1995 and 2001 1995 Coefficients t-values Constant 5.291 14.128*** FDIPC 0.073 4.505*** FINEXP 0.301 3.801*** FIXINV 0.218 4.827*** SOEPCT 0.002 0.051 URB -0.079 -1.178 Model Adjust R2: 0.594 F statistics: 50.38 on 5 and 164 summary degrees of freedom, p = 0.00 ***: significant at 1% level; *: significant at 10% level
2001 Coefficients t-values 6.015 11.293*** 0.066 5.139*** 0.099 1.335 0.285 4.374*** -0.090 -1.632 0.113 1.672* Adjust R2: 0.607 F-statistic: 53.14 on 5 and 164 degrees of freedom, p = 0.00
90 Table 4.2 LM tests for the OLS models under different weighting strategies BDSH Dist80 79.710 29.172 (2.19E-19)* (3.42E-08) LM_lag 1995 63.120 27.188 (9.87E-16) (9.55E-08) LM_err 2001 155.549 83.851 (5.35E-36) (2.70E-20) LM_lag 2001 103.964 58.284 (1.04E-24) (1.15E-14) *: Numbers in the parentheses are p-values. LM_err 1995
Dist160 16.175 (3.05E-05) 25.794 (1.97E-07) 40.995 (7.81E-11) 54.713 (7.10E-14)
Dist240 8.330 (2.15E-03) 5.805 (9.09E-03) 25.286 (2.56E-07) 12.264 (2.47E-04)
Dist320 0.793 (0.301) 1.924 (0.110) 0.747 (0.318) 1.140 (0.211)
As the LM tests suggest, the spatial dependence in our case study is primarily in the error term. According to Anselin and Rey (1991) and Rey and Montouri (1999), this primarily results from a mismatch between the actual spatial process’s boundaries and the county’s boundaries, where the data is organized. Table 4.1 also suggests that spatial dependence farther than 240 km (3 hours of driving on the expressways) is not significant, but it is strongest when using the border-sharing spatial weighting strategy (also agrees with the results found in the pattern analysis in the previous chapter). Hence I will focus the discussion on the two spatial error autoregressive models based on the border-sharing spatial weight matrix for both 1995 and 200113. The results of the spatial error models for 1995 and 2001 are reported in Table 4.3. From the results in Table 4.1 to 4.3, three interesting findings emerge. First, agreeing with the LM tests, the λs in both years are highly significant. As a result, and also reflected by the log likelihood and AIC, in both years the spatial error models indicate significant improvement from their OLS counterparts (Table 4.3). This result indicates that there exists strong spatial dependence in the regional development of
13
An extensive set of model results based on other weighting strategies is done, but not reported here to keep the paper concise.
91 Greater Beijing in both years. Furthermore, such spatial dependence mainly appears in the error terms. This indicates a mismatch between the boundaries of the spatial process, i.e., regional development, and the administrative boundaries within which the data are collected, i.e., the counties. This might be due to the way that how the basic geographic unit (county) is defined in the study. In order to obtain consistent data for the study, I treate the prefecture capitals in Hebei province and city cores of Beijing and Tianjin at the same geographic level as other counties. However, particularly in Greater Beijing, regional development processes in these different geographical units vary dramatically. Specifically, the city cores of Beijing and Tianjin (actually, even their surrounding counties) and the prefecture capitals in Hebei province represent the urban areas, which are usually more developed (Yu and Mao 1999), and much more attractive to FDI than their rural peers. On the other hand, the financial inputs from the governments, especially the local governments, might be more important to local development in the periphery counties than in these core areas. Hence as I treate these potentially very different geographic units as the same, the models result strong nuisance spatial dependence. Table 4.3 Spatial autoregressive error models for 1995 and 2001 1995 2001 Coefficients z-values Coefficients z-values Constant 5.507 16.343*** 5.799 12.570*** FDIPC 0.049 3.402*** 0.040 3.635*** FINEXP 0.365 5.047*** 0.247 3.683*** FIXINV 0.122 3.068*** 0.190 3.597*** SOEPCT 0.007 0.212 -0.029 -0.691 URB -0.039 -0.687 0.051 0.987 0.665 43.138***a 0.709 61.332***a λ Model Log likelihood: -21.556 for error Log likelihood: -28.728 for error summary model (OLS: -43.13) model (OLS: -59.39) AIC: 59.112, (OLS: 100.25) AIC: 73.455, (OLS: 132.79) ***: significant at 1% confident level; a: for λ, this value is the likelihood ratio.
92 Second, for the spatial error models in both years, comparing with their OLS counterparts, very interesting changes in different mechanisms emerge. In particular, although remaining significant in both OLS and spatial error models, the t/z values of the coefficients for FDIPC and FIXINV decrease from OLS to the spatial models (Table 4.1 and 4.3). On the other hand, the t/z value of the coefficient for FINEXP increases. More importantly, in 2001, the coefficient of FINEXP in the OLS model is not significant, while once the nuisance spatial dependence is incorporated, it becomes highly significant. These findings might also be the results from the choice of the specific analyzing geographic units. FDIPC and FIXINV are set as proxies of globalization and the role of input, whereas FINEXP is a proxy for decentralization and local governments. As aforementioned, in Greater Beijing, the inequality of the globalization’s influence on regional development is quite salient. Actually, in both 1995 and 2001, more than 80% of the FDI concentrates on the small amount of “counties” that are actually city cores of Beijing and Tianjin and their surrounding counties and the prefecture capitals in Hebei province (Fig 4.1). FIXINV has similar pattern, though less extreme as FDI. Hence for FDI, Beijing and Tianjin are the two most attractive locales in Greater Beijing, as one is the national capital (Wei and Yu 2005), and the other is one of the open coastal cities during the early reform (Wei and Jia 2003). In addition, comparing with their rural peers, prefecture capitals in Hebei provinces tend to have better investment environment and hence more chances to attract FDI (Lu 1997; Yu and Mao 1999). For FIXINV, it tends to concentrate on the more urbanized areas due to the fact that most of the industries are in the cities instead of countryside. In the OLS models, due to the existence of significant nuisance spatial dependence, their effects tend to be exaggerated. Once the spatial
93 dependence is incorporated in the spatial error model, however, the models seem to obtain more appropriate values for their effects. The similar rationale applies to FINEXP. Actually, according to the t-values in the OLS models, FINEXP is the least important mechanism among the three significant ones. However, once the nuisance spatial dependence is incorporated, it emerges to be the most important mechanism in both years (Table 4.3). This result implies that although FDI and FIXINV might have more influence on regional development in the core regions (Beijing-Tianjin and Hebei’s prefecture capitals), decentralization tends to play more important roles across the entire Greater Beijing.
± a
FDIPC 1995
FDIPC 2001
0 - 10 11 - 32 33 - 89 90 - 275 276 - 711
0 - 14 15 - 48 49 - 103 104 - 234 235 - 793
0
100
200
b
Kilometers
Figure 4.1 Distribution of per capita FDI in 1995 (a) and 2001 (b) Third, not like its southern peers, SOEs in Greater Beijing seem not to be significant at all. In both OLS and spatial error models, the coefficients of SOE are not
94 significant. Similar results are found for the urbanization factor (URB). For Greater Beijing, this result might suggest two findings. For the SOE, agreeing with the general trend in China’s regional development (Wei 2000; Wei and Fan 2000; Wei and Kim 2002; Yu and Wei 2003), state-owned enterprises become less important in local economic development. However, a significant negative relationship between SOE and regional development is not observed, either. This is often suggested by most region development studies on China (see Wei and Fan 2000; Wei and Kim 2002; Yu and Wei 2003 for detailed discussions) due to the SOEs’ lag in efficiency. This might be attributed to Greater Beijing’s relative long-term of state-owned enterprises dominance before the reform. Actually, in 2001, the gross industrial output of SOEs in the Grater Beijing still accounts for 49.39% of the total recorded gross industrial outputs. While in the southeastern provinces (Shanghai, Jiangsu, Zhejiang, Fujian and Guangdong), the same indicator is 27.12%, and the national average is 44.43% (Table 4.4, CSB 2002). For urbanization, the development of the rural area during the reform might be the reason why a significant relationship between urbanization and regional development is not found by the model. This also agrees to the findings by other scholars (Wei and Fan 2000; Lu and Wang 2002; Wei and Kim 2002) that the reform boosted the development of Township and Village Enterprises (TVEs), which are usually the representatives of rural industrialization in China during the reform. It also agrees with the findings in the previous chapter that the urban-rural two-tier spatial structure is disappearing during the study period.
95 Table 4.4 State-owned enterprises in Greater Beijing
National average Beijing Tianjin Hebei Greater Beijing
Gross Industrial output in 2001 (Billion Yuan) All recorded* State-owned Percentage (%) 9544.90 4240.85 44.43 290.88 189.38 65.11 294.04 97.44 33.14 376.69 188.07 49.93 961.61 474.89 49.39
Southeastern China 4361.46 1182.67 27.12 Shanghai 700.39 340.97 48.68 Jiangsu 1174.78 311.16 26.49 Zhejiang 788.25 119.47 15.16 Fujian 294.50 87.39 29.68 Guangdong 1403.54 323.67 23.06 * All recorded enterprises include all state-owned enterprises and non-state-owned enterprises with annual sales revenue that is more than 5 million Yuan (CSB 2002) 4.5 Conclusion
While the economic development in China attracts much scholar’s attention, mechanism study tends to understand the underpinning forces of the development. Numerous literatures established sound theoretical foundations for us to embark on this mission. However, primarily due to data availability and limitation of analytical techniques, mechanism studies on China remain limited in both geographic scale and studying methodology. In particular, most of the studies focus on provincial level or southeastern China where data are abundant; and spatial effects are less studied. This chapter initiates an attempt to applying spatial econometric methodology in subprovincial level regional development studies in China. In particular, using data from 1995 and 2001 at county-level from Greater Beijing, a set of spatial regression models are constructed. Specification tests yield that spatial error (nuisance) model is the better
96 alternative and point to a mismatch between underlying spatial processes and the geographic units where data are organized. From the empirical analysis, three key conclusions can be drawn. First, not only does the spatial econometric methodology employed incorporate the spatial effects in the models, it also identifies a possible data collection unit (county) – process (development) mismatch. This mismatch is due to the treatment of economically very different geographic units at the same level. Second, the mismatch between the data collection unit and process tend to exaggerate some location-specific development mechanisms. In this study, the importance of per capital FDI and per capital fixed asset investment, which tend to concentrate more on the core regions, is exaggerated. While on the other hand, the role of the local government (decentralization), which seems to project more uniform influence on the entire Greater Beijing, seems to be masked. Third, development in the Greater Beijing has its own characteristics. As a traditional industrial base of China, not like its southern peers, state-owned enterprises seem to have very little influence on its development. While due to their lack in efficiency and hardness to transform, SOEs are usually witnessed as a lagging factor to regional development in the literature (Wei 2000; Wei and Fan 2000; Lu and Wang 2002; Wei and Kim 2002; Yu and Wei 2003). However, in this research using data from 1995 and 2001 of Greater Beijing, no such significant relationship is successfully established. This finding complements the current understanding of China’s regional development in the literature.
97
Chapter 5 Spatially Varying Mechanisms – A Further Investigation of Development Mechanisms via Geographically Weighted Regression Analysis 5.1 Introduction
Investigation of regional development mechanisms in the previous chapter takes the initial step of incorporating spatial dependence in the modeling process. Although the analysis results identify possible heterogeneous spatial structure in Greater Beijing’s development mechanisms, the model by itself is still a global model. That is, the model implicitly assumes that the same set of development mechanisms will invoke the same response everywhere in space. The interpretation of the model results is fairly similar to the non-spatial OLS approach except that spatial effects are taken care of. In fact, spatial regression model is developed due to the recognition that there are inherent difficulties in analyzing spatial data (Cliff and Ord 1981; Upton and Fingleton 1985; Griffith 1988; Haining 1990; Cressie 1991), especially the spatial dependence, i.e., “the coincidence of value similarity with locational similarity” (Anselin 2001). Apart from spatial dependence, there is another difficulty in analyzing spatial data that might be better addressed by other models instead of spatial regression. That is the spatial heterogeneity, i.e., the same set of stimuli may yield different responses in different part of the studying region (Bailey and Gatrell 1995, Fotheringham et al. 2002). In fact, as pointed out by Brunsdon et al. (1999), spatial heterogeneity and spatial dependence are closely related. They proved that if spatially varying relationships were
98 modeled within a global framework such as standard regression, then the error terms in the global regression model would exhibit autocorrelation (Brunsdon et al. 1999). That is to say, data analysis on the heterogeneous spatial structure might produce the same output as there is spatial dependence in a homogeneous spatial structure. However, if the data analysis model adopted is a global model, the heterogeneous spatial structure might be masked. Actually, as the results in the previous chapter suggest, in Greater Beijing, the relationship between some of the mechanisms and the development could be better understood if the assumption of a homogeneous spatial structure is relaxed. Based on such a relaxation of homogeneous spatial structure assumption, the geographically weighted regression (GWR) model is suggested by Brunsdon and colleagues (Brunsdon et al. 1996, 1999; Leung et al. 2000a, 2000b; Fotheringham et al. 2002) as an alternative of the global spatial models to address the issue. In GWR, any heterogeneous relationships being measured (the development mechanisms in this study) are accounted for by allowing the calibrated coefficients to vary spatially. In this sense, the spatial drift from average global relationships is measured directly instead of as a second-order effect through the spatial distribution of residuals, as in the spatial regression’s case (Brunsdon et al. 1999). Moreover, the GWR model’s hypothesis of spatial regimes is statistically test-able. The test results provide statistically confident support for or rejection to the hypothesis. This chapter hence intends to extend the investigation initiated in the previous chapter of regional development mechanisms in Greater Beijing via the GWR models. The same set of data and model specification in chapter 4 are used here. In the sections
99 that follow, a brief introduction of the GWR techniques is given. The next section presents the empirical analysis results. The last section concludes the chapter. 5.2 Spatially Varying Mechanisms: Techniques of GWR
The GWR technique has recently received intensive attention from scholars (Brunsdon et al. 1996, 1999; Fotheringham et al. 1997, 1998, 1999, 2002; Leung et al. 2000a, 2000b; Huang and Leung 2002, Paez et al. 2002a, 2002b; Yu and Wu 2004, to name but a few). In general, GWR is an expansion of the standard regression method particularly designed for using with spatial data. The technique is appealing in analyzing spatial datasets in that it allows the regression coefficients to vary across space (nonstationarity). For instance, with geographically weighted regression, the relationship expressed in equation (4.2) is then: GDPi* = βi0+βi1 ∗ FDIi*+βi2 ∗ FIXINVi*+βi3 ∗ SOEPCTi*+βi4 ∗ FINEXPi*+βi5 ∗ URBi* (5.1)
instead of fixed in the global model, the coefficients α*j (j = 0, 1, 2, 3, 4, 5) now vary in respect to the location i. Different from spatial regression, the calibration of GWR utilizes the weighted least squares techniques. This is because it incorporates the spatial heterogeneity in the model specification; it assumes the error term to be i.i.d. (Fotheringham et al. 2002). The calibration processes as below. First, for data location 1, the procedure determines a weighting scheme to assign weights to each and every other data locations in the studying region. The weights are determined through the general application of the First Law of Geography (Tobler 1970), i.e., the weights act to ensure that data locations near the location i have more influence than those further away. In practice, usually a Gussian or Gussian-like (such as bi-square
100 or tri-cube) distance decaying function is used to generate the weights. It is found through numerous experiments that the exact form of the distance decaying function exerts very little influence on the resulting spatial pattern (Fotheringham et al. 2002). In this chapter, a bi-square distance decaying function is utilized for its simplicity and computation effectiveness. Second, once the weights for each data locations are generated, they will be applied to corresponding data locations and a weighted least squares procedure takes place and produces the set of coefficients of data location 1. Third, the procedure moves to data location 2, and repeat the above steps until the coefficients of all the data locations in the studying region have been generated. Different weighting schemes are presented and discussed in the literature (Fotheringham et al 2002; Paez et al 2002a, 2002b; Yu and Wu 2004). In general, all the schemes fall within two categories, i.e., fixed or adaptive weighting schemes (Figure 5.1). In a fixed scheme, one optimum spatial kernel (represented by the spatial bandwidth, b) is determined and applied uniformly across the study area (5.1a). Such approach, however, as pointed out by Fotheringham et al. (2002) and Paez et al. (2002a, 2002b), may produce large local estimation variance in areas where data are sparse, which might exaggerate the degree of non-stationarity present. In areas where data are dense, subtle spatial nonstationarity might be masked. On the other hand, the adaptive scheme produces adaptive spatial kernels. The kernels are able to adapt themselves in size to variations in the density of the data so that the kernels have larger bandwidths where the data are sparse and have smaller ones where the data are plentiful (5.1b). For this consideration, this
101 study employs an adaptive spatial weighting scheme. To implement the function, a nearest neighbor approach is used to produce the adaptive spatial kernels.
w(b)
Fixed scheme
w(bi)
b
bi w(b)
w(bj) bj
b
a
Adaptive scheme
b
Figure 5.3 Fixed (a) and adaptive (b) weighting schemes in GWR An optimum number of the nearest neighbor (nnb) is obtained through an out-ofsample cross-validation (cv) procedure (Cleveland 1979; Fotheringham et al. 2002) or through minimizing the goodness-of-fit statistics, the Akaike Information Criterion (AIC, Hurvich et al. 1998; Fotheringham et al. 2002). In essence, it is to minimize either the cv score or the AIC statistics: n ) cv = ∑ [ y i − y ≠i (nnb)] 2 i =1
⎛ n + tr (S) ⎞ ) ⎟⎟ AIC = 2n ln(σ ) + n ln(2π ) + n⎜⎜ ⎝ n − 2 − tr (S) ⎠
(5.2)
) where yi is the observed dependent variable on location i, y ≠i (nnb) is the GWR fitted value of yi using the nearest neighbor nnb, with the observation for location i omitted ) from the calibration process. n is the total number of observations, σ is the maximum
102 likelihood estimated standard deviation of the error term, and tr(S) is the trace of the hat matrix S of the GWR, which is defined as (Fotheringham et al 2002):
yˆ = Sy
(5.3)
where y and yˆ are the vector of the dependent variable and the GWR estimated values using the nearest neighbor nnb. During the experiments, it is found the optimal nearest neighbors produced through both the cv score or the AIC statistic do not differ from each other much. However, the AIC has the general appealing that it could be used to assess whether GWR provides a better fit than a global model taking into account the different degrees of freedom in the two models. Hence in this chapter, the procedure of using AIC is adopted. The minimization is implemented in the statistical language environment R (R Development Core Team 2004) through a combination of golden section search and successive parabolic interpolation (Brent 1973). All the codes of implementing the optimal procedure are produced by the author according to instructions in Fotheringham et al. (2002). Although the techniques of GWR seem to be very appealing in analyzing spatial heterogeneity, from the practical viewpoint, two critical questions need to be addressed. The first one is a goodness-of-fit vs. simplicity question. As pointed by Leung et al. (2000a) and Fotheringham et al. (2002), most of the time, the GWR model will fit a given data set better than a global OLS model since there are much more parameters (effective parameters) included in the calibration process. However, in practice, a simpler model is usually preferred over a more complex one if there is no significant improvement from the latter. To justify the employment of GWR instead of OLS in this study, an ANOVA
103 test (Brunsdon et al. 1999; Fotheringham et al. 2002) and two F tests (F1 and F2 test in Leung et al. 2000a) are implemented in the R environment to address this concern. The second question concerns the core concept of employing a GWR model, namely, whether there is significant spatial heterogeneity among the relationships modeled. To answer this question, another F test (the F3 test in Leung et al. 2000a), based on the sample variance of the estimated model coefficients (Brunsdon et al. 1996; Leung et al. 2000a), is implemented in the R environment as well (for detailed technique discussions, see Leung et al. 2000a, p 21-23; Fotheringham et al 2002). If any independent variable (including the constant term) shows spatial heterogeneity by this test, a mixed GWR model may be more appropriate (Fotheringham et al. 2002). While in a mixed GWR model, except for controlling the non-varying factors’ coefficient to be constant, other calibration procedures are quite the same as the normal GWR calibration. For detailed technique discussion, one is recommended to Fotheringham et al. (2002, p 65-68). 5.3 Spatially varying development mechanisms in Greater Beijing
For the purpose of brevity and also due to data availability, this study presents the global OLS regression and GWR analytical results for two years: 1995 and 2001. Results from the OLS regression models for both years are reported in Table 5.1. Comparisons between the OLS and GWR models are in Table 5.2. Stationarity-test results show in Table 5.3. The global OLS models indicate that the relationships between per capita GDP and the five selected independent variables are significant in both years. Further more, the adjusted R2 indicates that in both years around 60% of the per capita GDP’s variation
104 is explained by the five elected independent variables. Among them, SOEPCT and URB do not show significant relationships with per capita GDP in 1995 at 95% confidential level, though marginally (90% confidential level) significant in 2001. FINEXP was significant in 1995, turns to be an insignificant variable in 2001 (Table 5.1). Table 5.1 Global OLS regression results for 1995 and 2001 in Greater Beijing Global OLS regression 1995 Coefficients Estimate Std. Std. Error t-value Estimate Intercept 5.29 0.37 14.13 FDI 0.07 0.27 0.02 4.51 FINEXP 0.30 0.39 0.08 3.80 FIXINV 0.22 0.35 0.05 4.83 SOEPCT 0.002 0.003 0.04 0.05 URB -0.08 -0.12 0.07 -1.18 Dependent variable: per capita GDP Residual standard error: 0.3175 on 164 degrees of freedom Multiple R-Squared: 0.6057, Adjusted R-squared: 0.5936 F-statistic: 50.38 on 5 and 164 DF, p-value: < 2.2e-16 Global OLS regression 2001 Intercept 6.01 0.53 11.29 FDI 0.07 0.31 0.01 5.14 FINEXP 0.10 0.13 0.07 1.34 FIXINV 0.28 0.33 0.06 4.37 SOEPCT -0.09 -0.08 0.05 -1.63 URB 0.11 0.15 0.07 1.67 Dependent variable: per capita GDP Residual standard error: 0.3494 on 164 degrees of freedom Multiple R-Squared: 0.6184, Adjusted R-squared: 0.6067 F-statistic: 53.14 on 5 and 164 DF, p-value: < 2.2e-16
Pr(>|t|) 0.00 0.00 0.00 0.00 0.96 0.24
0.00 0.00 0.18 0.00 0.10 0.10
Table 5.2, however, indicates that from all the tests carried out, a GWR model performs significantly better than an OLS model in both years (actually, comparing the AIC statistics of GWR models to the spatial regression models in the previous chapter, Table 4.3, we can see that the GWR models fit the data at the same goodness level as the spatial regression models). The result justifies the employment of GWR model instead of
105 OLS analysis in investigating development mechanisms. Table 3, on the other hand, reveals that the variables FINEXP and URB in 1995 did not show significant spatial nonstationarity at the 95% level, though all the variables are significantly non-stationary in 2001 at 95% level. This result hence requires a mixed (Fotheringham et al. 2002) instead of a pure GWR model to apply for the 1995’s dataset. Table 5.2 Comparison between OLS and GWR models in 1995 and 2001 Goodness-of-fit vs. simplicity test in 1995 Fotheringham et al. 2002 ANOVA table test SS* DF* MS* F OLS residuals 16.53 164 0.10 GWR Improvement 7.78 38.11 0.20 GWR Residuals 8.75 125.89 0.07 2.94 Leung et al. 2000a F1 test SS DF MS F OLS Residuals 16.53 164 0.10 GWR Residuals 8.75 125.89 0.07 0.69 Leung et al. 2000a F2 test SS DF MS F OLS Residuals 16.53 164 0.10 GWR Improvement 7.78 38.11 0.20 2.02
Pr (>F) 0.00 Pr (F) 0.00
Goodness-of-fit vs. simplicity test in 2001 Fotheringham et al. 2002 ANOVA table test SS DF MS F Pr (>F) OLS residuals 20.02 164 0.12 GWR Improvement 11.13 36.75 0.30 GWR Residuals 8.89 127.25 0.07 4.34 0.00 Leung et al. 2000a F1 test SS DF MS F Pr (F) OLS Residuals 20.02 164 0.12 GWR Improvement 11.13 36.75 0.30 2.48 0.00 *SS stands for Sum of Squares, DF stands for Degrees of Freedom, and MS stands for Mean of Squares.
106 The GWR calibration results for the 1995 and 2001 datasets are produced through a combination of R and ArcGIS®. The stationary part of the mixed GWR model in 1995 is listed in Table 5.4. The non-stationary part of the 1995 GWR model and results for the 2001 model are presented in Fig. 5.2 and Fig. 5.3. Table 5.3 Spatial stationarity tests for individual variables in 1995 and 2001 Stationarity test in 1995 F statistic Intercept 1.62 FDI 1.83 FINEXP 1.30 FIXINV 1.50 SOEPCT 2.54 URB 1.24
Numerator d.f. 62.99 51.44 60.28 52.45 56.64 55.48
Denominator d.f. 163.12 163.12 163.12 163.12 163.12 163.12
Pr(|t|) 0.00 0.58
107
±
±
FIXINV 1995 -.12 - .03 .04 - .11 .12 - .18 .19 - .25 .26 - .33 .34 - .42 .43 - .57 Prefectures
FDI 1995 -.02 - .00 .01 - .03 .04 - .06 .07 - .10 .11 - .14 .15 - .20 .21 - .28 Prefectures
0
50
100
0
200
50
100
200
Kilometers
Kilometers
b
a
± Figure 5.4 Spatially varying mechanisms in 1995 for Greater Beijing: a. per capita foreign SOEPCT 1995 -.21 - -.14 -.13 - -.06 -.05 - -.01 .00 - .04 .05 - .12 .13 - .21 .22 - .31 Prefectures
0
50
100
200
direct investment; b. per capita fixed asset investment; c. percentage of fixed asset invested to the state-owned-
Kilometers
enterprises. c
108
±
±
FDI 2001 -.06 - -.03 -.02 - -.01 .00 - .01 .02 - .04 .05 - .07 .08 - .11 .12 - .14 Prefectures
0
50
100
FINEXP 2001 -.24 - -.10 -.09 - -.01 .00 - .06 .07 - .13 .14 - .23 .24 - .36 .37 - .51 Prefectures
200
0
Kilometers
50
100
a
b
±
±
FIXINV 2001 -.01 - .09 .10 - .19 .20 - .31 .32 - .48 .49 - .65 .66 - .79 .80 - .96 Prefectures
0
50
100
SOEPCT 2001 -.26 - -.17 -.16 - -.10 -.09 - -.04 -.03 - .01 .02 - .05 .06 - .10 .11 - .17 Prefectures
200
0
Kilometers
c
200
Kilometers
50
100
Kilometers
d
200
109
±
Figure 5.5 Spatially varying mechanisms in 2001 for Greater Beijing: a. per capita foreign direct investment; b. per capita local
URB 2001 -.18 - -.10 -.09 - -.03 -.02 - .01 .02 - .06 .07 - .13 .14 - .26 .27 - .41 Prefectures
0
50
100
200
financial expense; c. per capita fixed asset investment; d. percentage of fixed asset invested in the stateowned-enterprises; e. urbanization.
Kilometers
e From the results of the models, four impressions may be taken away. First, in 1995, as the non-stationarity test reveals, the variables FINEXP and URB are not significantly varying across the space. This result indicates that in 1995, the local government’s input was uniformly important for individual counties’ development in Greater Beijing. Furthermore, according to the result in table 5.4, URB is not significant in the mixed GWR model, which agrees again with our global model in the previous chapter. This indicates the relationship between urbanization and per capita GDP is not straightforward. One possible explanation for this insignificance between urbanization and per capita GDP is the policy of decentralization and development of Township and Village Enterprises (TVEs) during China’s reform. Although urban areas are previously industrialized areas, decentralization and development of TVEs soon shorten the gap
110 between urban and rural areas in Greater Beijing. The fact reflected in the model is that per capita GDP does not show significant difference in urban and rural areas in 1995. Second, the three spatially varying development mechanisms in 1995, i.e., FDIPC, FIXINV and SOEPCT, show very interesting spatial patterns. The FDIPC, which represents the force of globalization, has relatively high contribution to the per capita GDP in primarily two clusters. They include counties around Beijing-Tianjin and counties around the capital city of Hebei province, Shijiazhuang (Fig. 5.2). On the contrary, FIXINV, as an agent of the domestic input, seems to have very little contributions to the local development in these two clusters (Fig. 5.2a and 5.2b). This spatial pattern captures the fact that since after the early 1990s, Greater Beijing has accelerated its external-oriented economic development strategy (Waixiang Xing Jingji Fazhan Zhanlue, Lu 1997). As the nation’s capital and capital city of Hebei province, small wonder that counties around them took advantage of such economic development strategies. In the meantime, where foreign direct investment contributes most to the local development, domestic input seems to be less important for their development. SOEPCT, which is usually deemed the agent of infusion of marketization mechanisms and local endowments, usually shows negative relationship with local development in China (Yu and Wei 2003), as state-owned-enterprises (SOEs) usually suffered from a series of institutional problems and lagged behind in efficiency (Wei 2000). However, in 1995, a salient positive cluster of SOEPCT stood out in the counties around Tianjin-Tangshan in the east coastal regions of Greater Beijing (Fig. 5.2c). Unlike regions in the southeastern China, such as Jiangsu and Zhejiang, Tianjin and Tangshan are old industrial bases in Greater Beijing. SOEs were dominant economic forces in local development for a
111 relatively long time. For instance, the gross output value of SOEs in Tangshan in 1995 accounted for 61.5% of its total industrial gross output values. This finding reveals that development mechanisms have strong local characteristics. Third, as the economic reform in China deepened during the later 1990s and early this century, the spatial heterogeneity of development mechanisms in Greater Beijing was strengthening. In 2001, all five identified development mechanisms varied across Greater Beijing significantly (Table 5.3). Although FDIPC (Fig. 5.3a) and FIXINV (Fig. 5.3c) seem to remain the approximate spatial patterns as in 1995, changes emerge during the development. The most salient change is in FDIPC, the high contribution to per capita GDP of which now primarily concentrates on Beijing-Tianjin-Tangshan-Qinhuangdao in the northeastern coastal region of Greater Beijing, primarily the surrounding areas of the traditional capital circle area (Fig 5.3a). The cluster previously around Shijiazhuang disappeared. This is to be expected. As the reform deepened, Beijing took full advantage of its specific geopolitical position and became one of the hot spots in China of the inflows of FDI. Hence, FDI began to play more important role in local development in its vicinity. On the other hand, as a regional center, Shijiazhuang is much less attractive to FDI in the process, importance of FDI in local development dropped. This causes the disappearing of the FDI’s cluster around Shijiazhuang. In the summer 2003 interview with a local official in the Shijiazhuang Municipal Department of Investment Administration, I was informed that “Shijiazhung, during the past several years, was actually under the shadow of the Beijing in attracting outside investments”. In addition, “the local investment environment lagged far behind the demands…the conservative administrative style of the local government, such as lack in administrative efficiency,
112 unwilling to build more entertaining facilities, etc., further impeded the inflow of outside investments, especially foreign investors” (interview dairy 2003). The changes in spatial pattern of SOEPCT are also fairly interesting (Fig. 5.3d). Reform on SOEs in Tianjin and Tangshan seems to be quite noticeable. By implementing the external-oriented economic development strategies and decentralization, some of the SOEs in Tianjin and Tangshan changed their ownership to collective-owned-enterprises or shareholding enterprises (Yu and Mao 1999). Indeed, the gross output value of SOEs in Tangshan only accounted for 12.4% of its total industrial gross output values in 2001. The reforms de-emphasize SOEs’ dominance in local development in this region. Hence in 2001, the significant SOE cluster disappeared. Nonetheless, as globalization become less important in Shijianzhuang and its vicinity, SOEs in this area seems to take over (Fig 5.3d). Fourth, the FINEXP and URB, which were not significantly varying across Greater Beijing in 1995, showed significant spatial non-stationarity in 2001. The high contribution of local governments’ input, as represented by FINEXP, now primarily concentrates on the region of Baoding-Cangzhou, including parts of Langfang, Tianjin, Tangshan, and Hengshui in the central eastern coastal regions of Greater Beijing (Fig. 5.3b). While in most of other regions, the local governments’ input seems to have negative (though not significant) to negligible effects on local development. Compare this result with the one in 1995, when FINEXP seemed to be uniformly important for local development in Greater Beijing, its positive supporting effect to local development shrank to the central eastern part in 2001. This finding suggests that decentralization in Greater Beijing is more salient in the peripheral regions, while in the two capitals, i.e., Beijing and Shijiazhuang, decentralization might not be as effective as in non-capital
113 regions in supporting local development. URB, however, seems to have a fairly clear spatial pattern in 2001. Relatively high contribution to per capita GDP of this mechanism concentrates on the region of northern counties in Beijing, Zhangjiakou and Chengde in the northern part of Greater Beijing (Fig. 5.3e). This finding indicates that as the development inequality between urban and rural areas are disappearing in most part of Greater Beijing, in regions of Beijing-Zhangjiakou-Chengde, urban areas are still more developed than their rural peers. The finding again agrees with the result from the previous chapter. In Beijing’s case, the differentiation between the metropolitan area and its surrounding counties may be because of the metropolitan area’s specific geopolitical position. Counties surrounding the nation’s “heart” can be easily shadowed by its influence. However, for Zhangjiakou and Chengde in the northern and northwestern part of Greater Beijing, the divide among urban and rural areas indicates that TVEs’ development in these two prefectures during the late 1990s was slower than their southern peers. One of the possible reasons is their relative geopolitical linkage to the nation’s capital, Beijing. As aforementioned, historically, Zhangjiakou and Chengde were deemed the northern “gates” for Beijing, both militarily and eco-environmentally. Although the military meaning in “gates” is weakening, the eco-environmental part becomes more important in recent years (Hebei Development and Planning Commission 2002). Rapid development of TVEs as their southern peers is prohibited, which prevent these two prefectures from catching up with their southern peers. The summer interview in 2003 with a local official in the Department of Planned Economy Development in Zhangjiakou indicates that it is relatively hard for Zhangjiakou to get approved on developing or introducing some industries. These include bio-chemistry, coal refinery or
114 steel industries, which might have the potential to boost local economy based on the available resources (interview diary 2003). 5.4 Conclusion and Discussion
As China makes great progress in regional development in recent years, investigation of the development mechanisms attracts scholars’ interests (Wei 2000, 2002; Wei and Kim 2002). It is also noticed that in China, spatial variation is an inherent characteristic of regional development instead of an exogenous factor (Huang and Leung 2003; Wei and Ye 2004). A global model might be able to capture an overall view of the development mechanisms, for specific locales, the mechanisms expressed in a global model might be inaccurate, or even incorrect. This chapter, on the other hand, hypothesizes spatial non-stationarity in regional development mechanisms. It is important in that it promotes the concept of spatially varying mechanisms in China’s regional development. By relaxing the assumption that development mechanisms are the same across the space, interesting and significant spatial patterns are identified through a case study in Greater Beijing. In summary, three key conclusions from this chapter can be obtained: First, the analyses indicate that spatial non-stationarity in development mechanisms plays important role in understanding China’s regional development. Significant spatial non-stationarity in some key mechanisms is identified in Greater Beijing. The findings therefore provide solid evidence for location-specified policy implementation and strategic planning. Second, spatial patterns of the regional development mechanisms in Greater Beijing are of interest. The study finds that the force of globalization promotes local
115 development better in counties around the nation’s capital, Beijing, while the domestic support seems to be less important in this area. For SOEs, although the global models indicate that SOEs did not have significant impact on local development, salient positive contribution of SOEs to local development was identified in 1995 in the areas surrounding the traditional industrial base, Tianjin-Tangshan, though such contribution disappeared in 2001. The finding captures the fact that as the nation’s traditional old industrial bases, SOEs still play important role in local development in the 1990s, though such importance is weakening. This result also reveals that development in Greater Beijing is quite different from the development in the southeastern coastal provinces, as intensively studied by scholars in China’s regional development (Wei and Fan 2000; Huang and Leung 2002; Wei and Kim 2002; Wei and Ye 2004). Third, the GWR models reveal a possible remaining of urban-rural divide in the northern part of Greater Beijing. As aforementioned, such urban-rural divide is a result of the pre-reform’s urban-industrialization in China’s regional development policies. The remaining of such divide has its roots in the development trajectory and geopolitical position of the related spatial units. Moreover, it indicates an inequality in not only regional development but also the reform’s impacts on regional development in Greater Beijing. It seems now to be quite clear that regional development mechanisms have strong local characteristics. Geographically weighted regression, by relaxing the assumption of spatial homogeneity of development mechanisms in regional studies, allows such local characteristics to be usefully included in the analysis and understanding. This chapter presents an attempt to employ this technique in understanding China’s regional
116 development. It is hoped that such attempt will be of value to the research community of regional development studies.
117
Chapter 6 Discussion and Conclusion As China’s regional development during the past decades attracts more and more scholars’ attention, studies become abundant in understanding the patterns, processes and mechanisms of China’s development. During the process, a body of literature emerges to embark on the mission for better understanding China’s regional development via a series of theoretical and methodological breakthroughs (Ying 2003; Yu and Wei 2003; We and Ye 2004). However, due largely to the limitation of research methods development and implementation environment, data availability, willingness of local governments’ collaboration, and scholars’ personal experiences, a few more studying areas stand open. First, although most geographers studying China’s regional development realizes the importance of GIS and spatial data analysis in their work (for instance, see Wei and Fan 2000; Yu and Wei 2003), the inclusion of GIS and spatial data analysis in their study remains limited. One possible reason lies in the fact that sophisticated and statistically operable methods in analyzing spatial data are not quite developed and readily available until very recently. In addition, GIS has long been regarded as more of a cartographic tool instead of an analytical medium. Consequently, non-spatial methods dominate most of the studies on China’s regional development. For instance, coefficient of variation is often employed as an index for measuring regional inequality in China (Wei 2000). Although it gives important information concerning the degree of regional inequality in the development, ignoring spatial effects impedes it from recognizing spatial agglomeration processes that might present. Furthermore, spatial dependence and heterogeneity are usually present in spatial data analysis, which poses difficulties in employing traditional statistical procedures and models in understanding the processes
118 that generate the data. As GIS “as an intelligent assistant is giving way to a new view of GIScience as a medium of communication, in which spatial data analysis is one of several ways of enhancing the message”, the past decades see the gradual infusion of GIS and spatial data analysis (Goodchild and Hanning 2004). In such process, GIS’s analytical potential is largely exploited and improved. Although direct spatial data analysis techniques within specific GIS systems are still in the progress of developing, spatial data analysis frameworks or standalone environments based on information provided by GIS are booming in recent years (Anselin 2004, GeoDa; Bivand, 2004, SPDEP; Yu and Bivand 2004, SPGWR). Second, data availability is usually an inevitable obstacle in studying China’s regional development. Most studies on China focus on the understanding at provincial administrative level and sub-provincial geographic units in provinces of the southeastern coastal China (Wei and Fan 2000; Huang and Leung 2002; Wei and Kim 2002;Ying 2003; Yu and Wei 2003; Wei and Ye 2004). The reason is apparent, since these data are readily available through official published yearbooks. In addition, most of the local governments in the southeastern China are much more collaborative than their peers in the northern part of the country. However, in order to further understand the regional development processes in China, it might be more appropriate to turn the attention to finer geographical scale and locales other than the often studied southeastern China. For one thing, as mentioned in chapter 3, interesting spatial processes and effects might be masked by the aggregation of data from finer to larger geographic units. For another, for a country that has a territory as vast as China, better understanding of its regional
119 development processes demands studies on various parts of the country instead of focusing on a few spots (Wei and Fan 2000; Wei and Ye 2004). To this end, this research initiates the task of studying regional development in northern China, Greater Beijing, within a GIS and spatial data analysis framework. A few key conclusions could be drawn from the analyses presented in the previous chapters. First, the inclusion of GIS and spatial data analysis techniques in the study of China’s regional development patterns yield quite interesting findings. The results reveal a significant spatial agglomeration process in Greater Beijing during the studying period (1978-2001). This finding indicates counties with high or low values per capita GDP (as an indicator for regional development in this study) tend to cluster together in Greater Beijing. The dynamic process analysis reveals a strengthening trend of such spatial agglomeration, which is likely a result from the newly formed clusters in this region. In addition, from a spatial statistical standpoint, the local spatial analysis successfully identified the significant urban-rural divide in this region at the beginning of the studying period, which has long been recognized as a result of Mao’s China’s urbanindustrialization policies (Lu 1997; Wei 2000). Moreover, with temporal comparison between the local spatial patterns of the beginning and ending years of the studying period, a north-south regional divide emerges to replace the gradually disappearing urban-rural divide. This agrees with the policy change in China’s reform that promotion of Township and Village Enterprises (TVEs, Wei and Fan 2000; Lu and Wang 2002) start to take over the urban-industrialization process. It might be argued that similar results can be drawn from analyzing the raw data of per capita GDP along with geographic information. However, the employment of spatial data analysis technique,
120 specifically the Moran’s I and its decomposed components, establishes a statistically verifiable result. The analytical conclusion hence might provide stronger argument supporting the patterns observed. These results exhibit promising aspects of employing GIS and spatial data analysis techniques in understanding regional development processes. Second, the application of spatial econometric techniques, in particular, the spatial regression analysis, in further understanding the underlying forces of regional development enables us to incorporate spatial dependence in analyzing cross-sectional data on geographic units. During the development mechanism analysis, traditional methods tend to establish regression models between development and its mechanisms, and estimate such models through the ordinary least squares (OLS) estimator (Barro and Sala-I-Martin 1991, 1995). However, the existence of spatial dependence in crosssectional data on geographic units violates the OLS estimator’s i.i.d. error assumption. The violation leads to unreliable or even incorrect inferences from the OLS estimator. Spatial regression, on the other hand, relaxes the i.i.d. error assumption, and estimates the model through the maximum likelihood estimator. Furthermore, specification tests can finely point to the specific spatial processes that might generate the observed pattern. The analyses find out a mismatch of boundaries between spatial process and the data collection units in Greater Beijing. This is to be expected. As in the analysis, for data collection purpose, some of the core regions, such as the city cores of Beijing and Tianjin and urban regions in Hebei province, are aggregated as the same level as other counties. Since the development process, industrial bases, economic infrastructure and environments for these geographic units are very different. When they are treated at the
121 same analytical level, the boundary mismatch is only a natural result. However, such mismatch, when ignored, maskes the importance of a significant development mechanism, the agent of decentralization and local governments. Meanwhile, the importance of globalization and capital inputs, which are more in the core-regions than their peripheral peers, is exaggerated. Third, although the boundary mismatch is the cause of the observed spatial dependence in the regression models, the analytical results present a possible heterogeneous spatial structure. Hence, a geographically weighted regression model is called for trying to address the heterogeneity directly instead of modeling it as a second order effect. The results provide solid evidence that there exists significant heterogeneous spatial structure in Greater Beijing’s regional development mechanisms. Specifically, globalization and the domestic input tend to form a complementary spatial pattern in Greater Beijing. Globalization (represented by per capita foreign direct investment) seems to be a more important factor in the core (urban) regions while domestic input supports local development better in the peripheral regions. While decentralization and local governments’ support were uniformly important for local development in the middle 1990s, as the reform deepened, its importance shrank to the eastern side of the region. The spatial pattern might suggest a difference in local governments’ efficiency and local officials’ “marketization mind”. In general terms, local governments in the eastern part of Greater Beijing tend to be more efficient and their local officials more marketization-oriented than their peers in other part of the region. In addition, although overall the reform on state-owned enterprises (SOEs) is not significant in Greater Beijing, a counter-intuitive relationship between development of SOEs and regional development
122 was found in Tianjin and Tangshan region in the middle 1990s. As pointed out by scholars (Wei 2000; Wei and Fan 2000; Yu and Wei 2003), due to the lack of efficiency, SOE usually is the lagging factor in local development. However, the GWR model finds a significant positive relationship between SOEs and local development in Tianjin and Tangshan regions in 1995, though such relationship disappeared in later development. The fact, however, still indicates regional development in China possesses strong local characteristics, as Tianjin and Tangshan regions were dominated by SOEs during Mao’s China and the early reform period (the 1980s), which is different from the cases in the often-studied southern and southeastern coastal provinces. As mentioned in the research objectives, this study is all about geography through the intensive inclusion of GIS and spatial data analytical techniques in understanding China’s regional development. The case study on Greater Beijing serves as an example for a more complete view of China’s regional development. The study results yield some findings that might both support and complement the common wisdom on China’s regional development. In a spatial data analysis standpoint, the methods employed in the study provide reasonably robust contextual and statistical meanings. It is hence to this point that I hope this study will be of value to the research community of China’s regional studies.
123
References Amos, O. M., Jr. (1989). An inquiry into the causes of increasing regional income inequality in the United States. Review of regional studies, 19(2), 1-13. Amos, O. M., Jr. (1990). Growth pole cycles. Review of regional studies, 20(1), 37-48. Anselin, L. (1988) Spatial econometrics: methods and models. Dordrecht: Kluwer Academic Publishers. Anselin, L. (1992) SpaceStat: A Program for the Analysis of Spatial Data: Tutorial. Santa Barbara, CA: National Center for Geographic Information and Analysis, University of Califoria. Anselin, L. (1995). Local indicators of spatial association-LISA. Geographical Analysis 27, 93-115. Anselin, L. (1996) The Moran scatterplot as an ESDA tool to assess local instability in spatial association. In M. Fisher, H. J. Scholten & D. Unwin (Eds.), Spatial analytical perspectives on GIS (pp 111-125). London: Talylor & Francis Anselin, L. (2001). Spatial econometrics. In B. Baltagi (Ed.), A Companion to Theoretical Econometrics (pp310-330). Oxford: Basil Blackwell. Anselin, L. (2004). GeoDa Analysis Software, available from http://sal.agecon.uiuc.edu/geoda_main.php Anselin, L., & Hudak, S. (1992). Spatial econometrics in practice: a review of software options. Regional Science and Urban Economics, 22, 509-536. Anselin, L., & Rey, S. (1991) Properties of Tests for spatial dependence in linear regression models. Geographical Analysis, 23(2), 112-131.
124 Bailey, T. C., & Gatrell, A. C. (1995). Interactive Spatial Data Analysis. Essex: Longman Scientific and Technical. Bao, S., Chang, G., Sachs, J. D., & Woo, W. T. (2002) Geographic Factors and China’s Regional Development under Market Reforms, 1978-1998. China Economic Review, 13, 89-111. Bao, S., Henry, M. S. & Barkley, D. (1995) RAS: A Regional Analysis System Integrated with Arc/Info. Computer, Environment and Urban Systems, 19(1), 37-56. Barro, R., & Sala-I-Martin, X. (1991). Convergence across states and regions. Brookings papers of economic activity, 1, 107-182. Barro, R., & Sala-I-Martin, X. (1995). Economic growth. New York: McGraw-Hill. Berry, B. L. J. (1991). Long-wave rhythms in economic development and political behavior. Baltimore: Johns Hopkins University Press. Bivand, R. S. (2002). Spatial econometrics functions in R: classes and methods. Journal of Geographical System 4, 405-421. Bivand, R. S. (2004). SPDEP: Spatial Dependence: Weighting Schemes, Statistics and Models. Available at http://cran.r-project.org/src/contrib/spdep_0.2-14.tar.gz. Bo, Y. (1993). Reflections of some important decisions and events. Beijing: Central School of the Chinese Communist Party Press. Borts, G. H., & Stein, J. L. (1964). Economic growth in a free market. New York: Columbia University Press. Bramall, C. (1993). In praise of Maoist economic planning. Oxford: Oxford University Press.
125 Brent, R. (1973). Algorithms for Minimization without Derivatives. Englewood Cliffs, New Jersey: Prentice-Hall. Brown, L. A. & Lawson, V. A. (1989). Polarization reversal, migration related shifts in human resource profiles and spatial growth policies: A Venezuelan study. International regional science review, 12, 185-188. Brown, L. A., & Stetzer, F. C. (1984). Development aspects of migration in Third World settings: A simulation with implications for urbanization. Environment and planning A, 16, 1583-1603. Brunsdon, C., Fotheringham, A. S., & Charlton, M. (1996). Geographically weighted regression: A method for exploring spatial nonstationarity. Geographical Analysis, 28, 281-298. Brunsdon, C., Fotheringham, A. S., & Charlton, M. (1999). Some notes on parametric significance tests for geographically weighted regression. Journal of Regional Science, 39(3), 497-524. Chaudhuri, P. (1989). Economic theory of growth. Ames, Iowa: The Iowa State University Press. Clark, G. L., Feldman, M. P., & Gertler, M. S. (Eds.). (2000) The Oxford Handbook of Economic Geography. Oxford: Oxford University Press. Cleveland, W. S. (1979). Robust locally weighted regression and smoothing scatterplots. Journal of American Statistical Association, 74, 829-836. Cliff, A. D., & Ord, J. K. (1973) Spatial Autocorrelation. London: Pion. Cliff, A. D., & Ord, J. K. (1981). Spatial Processes: Methods and Applications. London: Pion.
126 Cressie, N. (1991). Statistics for Spatial Data. New York: Wiley CSB (Chinese Statistic Bureau), (2002). Chinese Statistical Yearbook. Beijing: China Statistics Press. Donnithorne, A. (1972). China’s cellular economy. China Quarterly, 52, 605-619. Donnithorne, A. (1976). Centralization and decentralization in China’s fiscal management. China Quarterly, 66, 328-354. Emmanuel, A. (1972). Unequal exchange. New York: Monthly Review Press. Fan, C.C. (1995). Of belts and ladders: State policy and uneven regional development in post-Mao China. Annals of the Association of American Geographers, 85(3), 421449. Fan, C.C. (1997). Uneven development and beyond. International Journal of Urban and Regional Research, 21, 620-639. Fisch, O. (1984). Regional income inequality and economic development. Regional Science and Urban Economics, 14, 89-111. Fotheringham, A. S., Brunsdon, C., & Charlton, M. (2002). Geographically weighted regression: the analysis of spatially varying relationships. West Sussex: John Wiley & Sons, LTD. Fotheringham, A. S., & Brunsdon, C. (1999). Local forms of spatial analysis. Geographical Analysis, 31, 340-358. Fotheringham, A. S., Brunsdon, C., & Charlton, M. (1997). Two techniques for exploring non-stationarity in geographical data. Geographical Systems, 4, 59-82.
127 Fotheringham, A. S., Brunsdon, C., & Charlton, M. (1998). Geographically weighted regression: A natural evolution of the expansion method for spatial data analysis. Environment and Planning A, 30, 1905-1927. Franks, A. G. (1967). Capitalism and underdevelopment in Latin American. New York: Monthly Review Press. Friedmann, J. (1973). Urbanization, planning and national development. Beverly Hill, CA: Sage Friedmann, J., & Douglass, U. (1978). Agropolitan development toward a new strategy for regional planning in Asia. In F. Lo & K. Salih (Eds.), Growth poles and regional development policy: Asian experiences and alternative strategies. London: Pergamon. Friedmann, J., & Weaver, C. (1979). Territory and function. Berkeley: University of California Press. Fujita, M., & Hu, D. (2001). Regional disparity in China 1985-1994: the effects of globalization and economic liberalization. The Annals of Regional Science, 35, 337. Gallo, J. L., & Ertur, C. (2003). Exploratory spatial data analysis of the distribution of regional per capita GDP in Europe, 1980-1995. Papers in Regional Science, 82, 175-201. Getis, A., & Ord, J. K. (1992). The analysis of spatial association by use of distance statistics. Geographical Analysis, 24(3), 189-206.
128 Gilbert, A., & Goodman, D.F. (1976). Regional income disparities and economic development: A critique. In A. Gilbert (Ed.) Development planning and spatial structure. London: Wiley. Goodchild, M. F., & Haining, R.P. (2004). GIS and spatial data analysis: converging perspectives. Papers in Regional Science, 83, 363–385. Griffith, D. (1988). Advanced spatial statistics. Dordrecht: Kluwer. Gu, C. L., Shen, J. F., Wong, K. Y., & Zhen, F. (2001). Regional polarization under the socialist-market system since 1978: a case study of Guangdong province in south China. Environment and Planning A, 33, 97-119. Haining, R. (1990). Spatial data analysis in the social and environmental sciences. Cambridge: University Press Cambridge. Harvey, D. (1982). The limits to capital. Oxford: Basil Balckwell. Hebei Development and Planning Commission. (2002). Lianghuan Kaifang Daidong Zhanlue: Yanjiu Yu Shijian Huibian (Strategies in Driven by Two Circles and Open-up: Research and Practices). Shijiazhuang: Hebei Develepment and Planning Commission. Hill, H. (2002). Spatial disparities in developing East Asia: a survey Asian-Pacific Economic Literature, 16(1), 10-35. Hirshman, A. O. (1958). The strategy of economic development. New Haven: Yale University Press. Huang, Y., & Leung, Y. (2002). Analysing regional industrialization in Jiangsu province using geographically weighted regression. Journal of Geographical System 4, 233-249.
129 Hurvich, C. M. , Simonoff, J. S., & Tsai, C. L. (1998) Smoothing parameter selection in non-parametric regression using an improved Akaike information criterion. Journal of the Royal Statistic Society, Series B (Statistic Methodology), 60(2), 271-293. Imhof, J. P. (1961). Computing the distribution of quadratic forms in normal variables. Biometrika, 48, 419-426. Interview dairy. (2003) Summer Interview with local officials and scholars in Beijing, Shijiazhuang, Handan, Tangshan, and Zhangjiakou. Personal dairy. Knight, J., & Song, L. (1993). The spatial contribution to income inequality in rural China. Cambridge Journal of Economics, 17, 195-213. Knox, P., & Agnew, J. (Eds.) (1998). The Geography of the World Economy (3rd edition). New York: Oxford University Press. Knox, P., & Taylor, P. J. (Eds.) (1995). World cities in a world-system. Cambridge: Cambridge University Press. Krebs, G. (1982). Macroeconomic versus international business approach to foreign direct investment. Hitosubashi Journal of Economics, 23, 1-19. Krugman, P. (1991). Geography and Trade. Cambridge, MA: The MIT Press. Kueh, Y. Y. (1989). The Maoist legacy and China’s new industrialization strategy. China Quarterly, 119, 420-447. Lardy, N. R. (1975). Centralization and decentralization in China’s fiscal management. China Quarterly, 61, 23-60. Lardy, N. R. (1978). Economic growth and distribution in China. Cambridge University Press, Cambridge.
130 Lardy, N. R. (1980). Regional growth and income distribution in China. In: R. F. Dernberger (Ed.), China’s development experience in comparative perspective, (pp.153-190). Cambridge: Harvard University Press. Leung, Y., Mei ,C-L., & Zhang, W-X. (2000b). Testing for spatial autocorrelation among the residuals of the geographically weighted regression. Environment and Planning A, 32, 871-890. Leung, Y., Mei, C-L., & Zhang, W-X. (2000a). Statistical tests for spatial nonstationarity based on the geographically weighted regression model. Environment and Planning A, 32, 9-32. Leung, Y., Mei. C-L., & Zhang, W-X. (2003). Statistical test for local patterns of spatial association. Environment and Planning A, 35, 725-744. Li, Y. (1999). Quyu Fazhan Xin Silu (New Thoughts in Regional Development). Beijing: Economy Daily Press. Lieberman, O. (1994). Saddlepoint approximation for the distribution of a ratio of quadratic forms in normal variables. Journal of the American Statistical Association, 89, 924–928. Lin, G. C. S. (1997). Red Capitalism in South China. Vancouver: UBS Press. Lipshitz, G. (1992). Divergence versus convergence in regional development. Journal of planning literature, 7(2), 123-138. Long, G., & Ng, M. K. (2001). The political economy of intra-provincial disparities in post reform China: A case study of Jiangsu province. Geoforum, 32, 215-234.
131 Lu, D. (1997). Zhongguo Yanhai Diqu 21 Shiji Chixu Fazhan (Sustainable Development of China’s Coastal Regions in the 21st Century). Wuhan: Hubei Science and Technology Press. Lu, M., & Wang, E. (2002). Forging ahead and falling behind: changing regional inequalities in post-reform China. Growth and Change, 33, 42-71. Lyons, T. (1997). Intraprovincial disparities in post-Mao China: a multidimensional analysis of Fujian province. Journal of Developing Areas, 1, 1-27. Ma, L. J. C., & Cui, G. H. (2002). Economic Transition at the Local Level: Diverse Forms of Town development in China. Eurasian Geography and Economics, 43(2), 79- 103. Ma, L. J. C., & Wei, Y. H. [D.] (1997) Ideology, national defense, and economic pragmatism: determinants of state investment in China Tijdschrift voor Economische en Sociale Geografie, 88, 211-225. Maxwell, P. & Hite, J. C. (1992). The recent divergence of regional per capita income. Growth and change, 23(1), 37-53. Mera, K. (1978). Population concentration and regional income disparities: A comparative analysis of Japan and Korea. In M. Hansen (Ed.) Human settlement. Cambridge, MA: Ballinger. Myint, H. (1971). Economic theory and the underdeveloped countries. London: Oxford University Press. Myrdal, G. (1957). Economic theory and under-developed regions. London: Gerald Duckworth & Co.
132 Paez, A., Uchida, T., & Miyamoto, K. (2002a). A general framework for estimation and inference of geographically weighted regression models: 1. Location-specific kernel bandwidths and a test for local heterogeneity. Environment and Planning A, 34, 733-754. Paez, A., Uchida, T., & Miyamoto, K. (2002b). A general framework for estimation and inference of geographically weighted regression models: 2. Spatial association and model specification tests. Environment and Planning A, 34, 883-904. Peet, R. (1975). Inequality and poverty. Annals of the Association of American Geographers, 65(4), 564-571. Perroux, F. (1955). Note sur la notion de ‘pole de croissance’. In D. L. Mckee, R. d. Dean, & W. H. Leahy (Eds.), Regional economics (pp.93-104), New York: Free press. Porter, M. (1990). The competitive advantage of nations. New York: Free Press. Pred, A. (1977). City systems in advanced economics: Past growth, present processes, and future development options. New York: Wiley. R Development Core Team, (2004). R: A language and environment for statistical computing. Vienna, Austria R Foundation for Statistical Computing, ISBN 3900051-00-3, URL http://www.R-project.org Rawski, T. G. (1999). Reforming China’s economy: what have we learned? The China Journal, 41(Jan.),139–156. Rey, S. J., & Montouri, B. D. (1999). US regional income convergence: a spatial econometric perspective. Regional Studies, 33(2), 143-156. Reynolds, P. (1987). Political economy. Brighton: Wheatsheaf.
133 Slater, D. (1975). Underdevelopment and spatial inequality. Progress in Planning, 4, 97167. Smirnov, O., & Anselin, L. (2001) Fast maximum likelihood estimation of very large spatial autoregressive models: a characteristic polynomial approach. Computational Statistics and Data Analysis, 35, 301-319. Soja, E. (1980). The socio-spatial dialectic. Annals of association of American geographers, 70(2), 207-225. Solow, R. M. (1956). A contribution to the theory of economic growth. Quarterly Journal of Economics, 70(February), 65–94 SSB (State Statistic Bureau). (2001). 2000 China Statistical Yearbook. Beijing: China Statistic Press. SSB. (1999). Comprehensive Statistical Data and Materials on 50 Years of New China. Beijing: China Statistic Press. Stohr, W. B., & Todtling, F. (1977). Spatial equality. Papers of the regional science association, 38, 33-53. Tabuchi, T. (1988). Interregional income differentials and migration: Their interrelationships. Regional Studies, 22, 1-10. Tiefelsdorf, M. (2000). Modelling spatial processes: the identification and analysis of spatial relationships in regression residuals by means of Moran’s I. New York: Springer-Verlag. Tiefelsdorf, M. (2002). The saddlepoint approximation of Moran’s I’s and local Moran’s Ii’s reference distributions and their numerical evaluation. Geographical Analysis, 34, 187-206.
134 Tobler, W. R. (1970). A computer movie simulating urban growth in the Detroit region. Economic Geography, 46, 234-240. Tsui, K. Y. (1991). China’s regional inequality, 1952-1985. Journal of comparative economics, 15, 1-21. Upton, G., & Fingleton, B. (1985). Spatial data analysis by wxample, volume 1: point pattern and quantitative data. Chichester: Wiley. Vining, D. R., Jr., & Strauss, A. (1977). A demonstration that the current deconcentration of population in the United States is a clean break with the past. Environment and planning A, 9, 751-758. Walcott, S. (2002). Science parks as magnets for global capital: locating high-tech growth engines in metropolitan Shanghai. In B.I. Logan (Ed.), Globalization, the Third World State and Poverty-alleviation in the Twenty-first Century (pp.89102). Burlington: Ashgate. Walker, R. (1978). Two sources of uneven development under advanced capitalism. Review of radical political economics, 10(3), 28-37. Wei, Y. H. D. (1996). Patterns of Regional Inequality in China: Global Force, State Policy, and Local Context, unpublished dissertation proposal, UCLA, California. Wei, Y. H. D. (1999). Regional inequality in China. Progress in Human Geography, 23, 48-58. Wei, Y. H. D. (2000). Regional development in China: states, globalization, and inequality. London: Routledge.
135 Wei, Y. H. D. (2002). Multiscale and multimechanisms of regional inequality in China: Implications for regional policy. Journal of Contemporary China, 11(30), 109124. Wei, Y. H. D. (2004). Trajectories of ownership transformation in China: implications for uneven regional development. Eurasian Geography and Economics, 45(2), 90113. Wei, Y. H. D. & Fan, C. C. (2000). Regional inequality in China: A case study in Jiangsu province. Professional Geographer, 52(3), 455-469. Wei, Y. H. D., & Jia, Y. J. (2003). The geographical foundations of local state initiatives: globalizing Tianjin, China. Cities, 20(2), 101-114. Wei, Y. H. D., & Kim, S. (2002). Widening inter-county inequality in Jiangsu province, China, 1950-95. Journal of Development Studies, 38, 142-164. Wei, Y. H. D., & Ma, L. J. C. (1996) Changing patterns of spatial inequality in China, 1952-1990. Third World Planning Review, 18(2), 177-191. Wei, Y. H. D., & Ye, X. (2004). Regional inequality in China: a case study of Zhejiang province. Tijdschrift voor economische en sociale geografie (Journal of Economic and Social Geography), 95, 44-60. Wei, Y. H. D., & Yu, D. L. (2005) State policy and the globalization of Beijing: emerging themes. Habitat International in press. Weng, Q. H. (1998). Local impacts of the post-Mao development strategy: the case of Zhujiang Delta, southern China. International Journal of Urban and Regional Research, 22, 425-442.
136 Williamson, J. G. (1965). Regional inequality and the process of national development. Economic development and cultural change, 13(2), 3-83. Woo, W. T. (1999). The real reasons for China’s growth. The China Journal, 41(Jan.), 114–137. Wu, F. L. (2000). Place promotion in Shanghai, PRC. Cities, 17(5), 349-361. Xu ,W., & Tan, K. C. (2002) Impact of reform and economic restructuring on rural systems in China: a case study of Yuhang, Zhejiang. Journal of Rural Studies, 18(1), 65-81. Ying, L. G. (2003). Understanding China’s recent growth experience: a spatial econometric perspective. The Annals of Regional Science, 37, 613-628. Yu, D. L. (1999). New population crisis. In S. C. Dong, W. Z. Zhang, C. L. Fang, W. Wu, J. Wen (Eds.), 21 Shijie Zhongguo Kechixu Fazhan Xinlun (Discussion on China’s Sustainable Development in the 21st century). Xi’an: Shaanxi People’s Publishing House. Yu, D. L., & Bivand, R. S. (2004) SPGWR: Alpha Version of Geographically Weighted Regression Analysis. Available at http://agec144.agecon.uiuc.edu/csiss/Rgeo/ Yu, D. L., & Mao, H. Y. (1999). Economic development in China’s costal regions. Jingji Dili (Economic Geography), 19(4), 25-30. Yu, D. L., & Wei, Y. H. D. (2003). Analyzing regional inequality in post-Mao China in a GIS environment. Eurasian Geography and Economics, 44, 514-534. Yu, D., & Wu, C. (2004) Understanding population segregation from Landsat ETM+ imagery: a geographically weighted regression approach. GIScience and Remote Sensing, 41(3), 187-206.
137 Zhao, X. B. (1996) Spatial disparities and economic development in China, 1953-92: a comparative study. Development and Change, 27(1) 131-160.
138
CURRICULUM VITAE Education
May 2005, Ph.D. (expected to defend in March, 2005), Department of Geography, University of Wisconsin-Milwaukee. Dissertation title: “GIS and Spatial Modeling in Regional Development Studies: A Case of Greater Beijing” Advisor: Dr. Yehua Dennis Wei 1997
M.S., Department of Geography, Lanzhou University, Lanzhou, P.R. China Thesis title: “Studies on the Urban Traffic Efficiency: Case Study on Lanzhou”
1994
B.S. Department of Geography. Changsha Electric Power University. Changsha, P.R. China
Recent Publications
Peer-reviewed articles (in English) •
Yu, Danlin, Spatially Varying Development Mechanisms in the Greater Beijing Area: A Geographically Weighted Regression Investigation. Accepted for publication in Annals of Regional Science.
•
Yu, Danlin; Wu, Changshan, Incorporating Remote Sensing Information in Modeling House Values: A Regression Tree Approach. Accepted for publication in Photogrammetric Engineering & Remote Sensing.
•
Wei, Yehua, Dennis; Yu, Danlin, State Policy and the Globalization of Beijing: Emerging Themes. In press in Habitat International.
•
Yu, Danlin; Wu, Changshan, 2004: Understanding population segregation from Landsat ETM+ imagery: a geographically weighted regression approach. GIScience and Remote Sensing, 41 (3): 187-206.
•
Yu, Danlin; Wei, Yehua, Dennis, 2003: Analyzing regional inequality in postMao China in a GIS environment. Eurasian Geography and Economics. 44 (7): 514-534.
139 •
Peng, Zhong-Ren; Yu, Danlin; Edward A. Beimborn, 2002: Transit user’s perceptions of automatic vehicle location benefits. Journal of the Transportation Research Board: Transportation Research Record, No. 1791:127-132.
Proceeding paper (in English) •
Yu, Danlin, 2004: Modeling housing market dynamics in the city of Milwaukee: a geographically weighted regression approach. Accepted in UCGIS’s 2004 Summer Assembly electronic proceedings.
Papers in review (in English) •
Yu, Danlin, Wei Yehua, Dennis. GIS and Spatial Data Analysis of Regional Development in Greater Beijing, China. Submitted to Environment and Planning A for publication consideration.
Software packages authored and contributed •
Authored (co-author with Professor Roger Bivand): Statistical package for geographically weighted regression analysis, SPGWR, see http://sourceforge.net/project/showfiles.php?group_id=84357&package_id=1205 95 for the alpha version.
•
Contributed: Spatial dependence: weighting schemes, statistics and models, SPDEP, see http://cran.us.r-project.org/src/contrib/Descriptions/spdep.html for detail.
Conference Presentations •
Yu, Danlin. 2004. Spatially varying development mechanisms in the Greater Beijing Area: A GWR investigation. 51st Annual North American Meetings of the Regional Science Association International, 2004, November 11 – 13, 2004, Seattle, Washington.
•
Yu, Danlin. 2004. Modeling housing market dynamics in the city of Milwaukee: a geographically weighted regression approach. GIScience 2004/UCGIS Assembly 2004 Meeting, October 20 – 24, 2004, College Park, Maryland.
•
Yu, Danlin. 2004. GIS and Exploratory Spatial Data Analysis at County-Level: Regional Development in the Greater Beijing Area from 1978-2001. 56th Annual Meeting of the West Lakes Division of the Association of American Geographers and 58th Annual Meeting of the Wisconsin Geographical Society, October 7 – 9, 2004, Oshkosh, Wisconsin.
140 •
Yu, Danlin. 2004. GIS and Spatial Modeling in Regional Development: Case Study on the Greater Beijing Area. 2004 Centennial Conference of the Association of American Geographers, March 12- 18, 2004, Philadelphia, Pennsylvania.
•
Yu, Danlin. 2003. Regional development in Greater Beijing Area: A GIS and spatial perspective. 2003 Meeting of the East Lakes & West Lakes Divisions of the Association of American Geographers, October 16-18, 2003, Kalamazoo, Michigan. Yu, Danlin. 2003. Studies on quantifying regional carrying capacity: Case studies on Bohai Rim Area. The Wisconsin Geographical Society 57th Annual Meeting, September, 19-20, 2003, Eau Claire, Wisconsin.
•
•
Yu, Danlin, Wei, Yehua. 2003. Analyzing regional inequality in post- Mao China in a GIS environment. Chinese Economy After WTO: Opportunities and Challenges of Globalization, August 2-3, 2003, Ann Arbor, Michigan.
•
Yu, Danlin, Wei, Yehua. 2003. Globalizing Beijing. Annual Meeting of the Association of American Geographers. March 5-8, 2003, New Orleans, Louisiana.
•
Peng, Zhong-ren, Yu, Danlin. 2002. GIS on the Internet: A state of the art review. Urban and Regional Information System Association’s 40thAnnual Conference. October 26-30, 2002, Chicago, Illinois.
•
Yu, Danlin, Wei, Yehua. 2002. Modeling regional inequality in China in a GIS environment. Annual Meeting of the Association of American Geographers. March 20-24, 2003, Los Angeles, California.
•
Peng, Zhong-ren, Yu, Danlin, Edward A. Beimborn. 2002. Transit user’s perceptions of automatic vehicle location benefits. Transportation Research Board 2002 Annual Meeting, January 11-15, 2002. Washington, D.C.
Honors and Awards •
2004: Best graduate student paper, 2004 Joint Meeting of the West Lakes Division of Association of American Geographers and Wisconsin Geographical Society, Oshkosh, WI, October 7 – 9, 2004
•
2004: University Consortium for Geographical Information Science (UCGIS) 2004 Annual Assembly Travel Award.
•
2004: UWM Student GIS Project Competition 2004, second place.
•
2003 – 2004. Mary Jo Read Award, Department of Geography, UWM.
141 •
2002 – 2003. Graduate School Dissertation Fellow, UWM.
•
2002 – 2003. Mary Jo Read Award, Department of Geography, UWM.
•
2001 – 2002. Graduate School Fellow, UWM.
•
2000 – 2001. Graduate School Fellow, UWM.
Major Professor
Date
