Introduction
The acceleration of the economic reforms initiated at 1978, in China, has encouraged economic growth over the past three decades. China surpassed Japan as the world's second largest economy in 2010 . During 1979-2010, China's annual growth rate of gross domestic product (GDP) was 9.9 percent . Behind the success of economic growth, however, China must clearly realize that the country still face prominent problems due to country's development being unbalanced, uncoordinated and unsustainable, especially development between regions are poorly coordinated . Regional inequality may threaten national unity, social stability, residential welfare and economic healthy development has been demonstrated by numerous empirical studies , it also has become a burning issue in China attracting considerable attention from all levels of government and scholars . From the Ninth Five-Year Plan (1996-2000) to the Twelfth Five-Year Plan (2011-2015), narrowing the regional gap of development and promoting regional economic coordination always is a main objective of China's top-level Plan .
There is no doubt that regional inequality has been a core issue of study for economists, geographers and regional scientists in theoretical, empirical and policy perspective. Economic growth in most developed countries has generated a series of theories, which includes Neoclassical Exogenous Growth Theory , Endogenous Growth Theory , Rostow's Stages of Growth Theory, Kuznets's inverted U hypothesis , New Convergence Theory and New Economic Geography. Over the two decades, the new convergence theory (NCT) founded by Barro and Sala-i-Martin has been a mainstream economic theory of regional inequality. β-convergence, σ-convergence, and subsequent club-convergence derived by unconditional convergence hypothesis, are three characteristic concepts of NCT. The notion of β-convergence refers to the tendency for poor regions tend to grow faster than rich ones . The concept of σ-convergence assumes that due to the β-convergence, the overall level of divergence tends to decline in the long run. Existing convergence among similar types of economy or region, but little or no convergence between such clubs, is defined as club-convergence . At the same time, a growing body of literature is being created, in the framework of the NCT, to examine the existence and magnitude of convergence at the international , European union , national , state and county scale.
However, just as other regional development theory, the issue of spatial scale, spatial heterogeneity, spatial dependence and spatial-temporal dynamic was not considered by the NCT . Meanwhile, the new economic geography (NEG) theory is characterized by the importance of space and geography. So it can be provided solid foundation in study of regional development and regional inequality. Especially, there has been a renewed concern over regional development and regional inequality in geographic perspective, fuelled by the emergence of geographical information system (GIS) technique and spatial econometrics, together with the integration of NEG with GIS and spatial statistics analysis technique .
Naturally, any theories of regional inequality are designed mainly to clarify these issues: namely, how to measuring regional inequality, the influence of social and economic development resulting from regional inequality, what has caused regional inequality, and how to reducing regional inequality . However, most existing studies in many countries excessively focus on the existence and magnitude of regional inequality . Only in recent years, investigation of the mechanisms and determinants of regional development and regional inequality has attracted scholars' concern, related studies emphases in the importance of the (relative) geographical location; human capital ; financial policy ; agglomeration economies, economic integration and economic structures ; geographical spillovers of human capital ; trade openness ; decentralization ; innovation ; the integration of women into the labor market, the average number of patents, low-tech manufacturing ; trade patterns ; and physical geography .
Within these broad contexts, China's rapid economic growth and institutional transformation in the past thirty years have provided a good laboratory to deepen and extend our understanding of the mechanisms of regional inequality . Since the mid-1990s, the determinants of regional inequality in China attracted considerable scholarly interest. A host of literatures have addressed the mechanism of regional development and regional inequality in China. Most existing studies have examined the effects of socio-economic factors such as foreign direct investment (FDI) and international trade ; fixed-asset investment ; decentralization ; state-owned enterprises (SOEs) ; urbanization ; labor mobility ; education level ; public investment ; the ratio of heavy industry to gross output value ; physical capital intensity and total factor productivity ; human capital ; total factor productivity (TFP) ; and population growth rate . While some scholars highlight geographical factors include location and topography . In the meantime, the effects of regional preferential policies , institutional convergence , special economic zone (SEZ) , and structural break of policy also has attracted researchers' interest. Especially, Wei's multi-scale and multi-mechanism framework generalize the process of China's regional inequality as a triple transitions of decentralization, marketization, and globalization.
Given the fact that this paper's objective, the current studies on China's regional inequality, limited by space, scale, data, and framework, might not present better understanding of development mechanisms. Frist, scholar's investigation of regional inequality mechanisms in China seldom takes into consideration of spatial effects , including spatial heterogeneity, spatial dependence and spatial scale . The not-spatial method, mainly includes ordinary least squares (OLS) regression, is widely applied in the literature for understanding the determinants of regional inequality . In addition, the parameter direction and magnitude of the model coefficients of different independent variables are applied to identify the significance of determinants underlying regional inequality. However, ignoring of spatial effects might lead to biased result due to spatial effects are inherent in geographic process . Given all of that, more recent, most of economists use dummy variable to explain spatial effects, whereas utilization of this orthodox method might not manifest the real geographic information. More recently developed spatial data analysis methods and GIS technique provided a new platform for understanding of spatial effect of China's regional inequity landscape and to ameliorate these problems.
Second, the spatial scale is very important in research of geographic issue (e.g. regional inequality). Regional inequality in China is sensitive to geographical scale, demonstrated by a host of studies . However, most of studies focus on regional and provincial scale. For countries as vast as China, regional inequalities not only exist among regions or provinces but are even more evident among prefectures and counties . Interest in downscaled analysis is likely to increase with the tendency of micro research. Since the early 21st century, an increasing number of studies have examined the regional inequality mechanisms at the prefectural (city-level) and county level in China. For example, Jones's city-level research and Liu's prefectural research provide fresh empirical evidence on the determinants of regional development at this neglected level of analysis. Most of studies at the county level mainly focus on the intra-province level, especially the eastern and southeastern coastal "winning" provinces, such as Jiangsu , Zhejiang , Guangdong , and Greater Beijing Area . Studies on other provinces, central and western areas, and national county-level remain limited.
Third, literatures that have appeared over the last decade or two have extended the period of study to the early 2000s. The utilization of data for most of recent studies, however, was update to the year 2005 or 2006 . In addition, time-series and cross-sectional data are used by most scholars to explore regional development in China. The application of panel data remains limited. Panel data (also known as longitudinal or cross sectional time-series data) are a dataset in which the features of entities are observed across time, are generally more informative, and they contain more variation and less collinearity among the observed variables , and help to achieve dynamic monitoring.
Fourth, although the wending gap of regional inequality in China shares some common determinants with other transitional countries and regions, the mechanism underlying the uneven landscape of regions in China are complicated and dynamic, which can hardly be explained by some of the specific indicator, such as FDI, international trade, decentralization, and SOEs. The most of studies focus on a onefold perspective of regional development, for instance, socioeconomic, geographical, or policy factors, which might mislead scholar into forgetting the complexity of regional inequality. Thus the development of related research was constrained by the lack of a comprehensive analytical framework .
Based on the front elaboration, the objectives of this paper are as follows: (1) to establish a multi-mechanism framework and model system that can model the determinants of regional inequality in China at the county level during the 1992-2010 period; (2) to measure the degree, magnitude and spatiotemporal characteristics of China's regional inequality at the county level using spatial panel data, in the the late 20th century and early 21st century; and (3) to discuss some of the major policy implications for achieving balanced and coordinated regional development in the future. This paper is organized as follows. The next section presents an introduction of data source, the multi-mechanism framework and preliminary analysis. The third section specifies the model system including the measures of regional inequality, ESDA, and spatial panel data model. This is followed by a detailed analysis of the spatiotemporal dynamic and determinants of regional inequality in China among over 2000 counties with ESDA, orthodox OLS method and spatial panel data models. The paper summarizes the major findings and policy implications.
