Urbanization’s effects on the urban-rural income gap in China_ A meta-regression analysis

The considerable gap between urban and rural areas in China has been one of those social problems during the urbanization process. Since the early 2000s, an increasing number of theoretical and empirical studies have discussed the association between urbanization and urban-rural income gap (URIG) in China. However, a very limited consensus has been reached so far, which makes it challenging to support formulating well-informed policies. To identify factors contributing to different conclusions of the effects of urbanization on URIG in China, we conducted a systematic literature review of 29 empirical studies and stepwise meta-regression analysis from 94 direct effect-size estimates. Our findings reveal that while urbanization is associated with larger URIG when URIG is measured via urban-rural income/consumption, urbanization is associated with smaller URIG when URIG is measured with inequality index (e.g., Theil index and/or Gini coefficient). Additionally, financial development is correlated with larger URIG. By contrast, human capital level, agricultural support policy, and farmland scale contribute to narrowing URIG. Finally, we did not find a significant publication bias from the primary studies. This work suggests that it is worth to conduct more in-depth analysis to examine the heterogeneous effects of different indicators of URIG and their associations with other potential driving factors. Future work is suggested to investigate the effects of financial development level, human capital level, agricultural support policy, farmland scale, and urban land scale on the relationship between urbanization and URIG. In the urbanization process, policymakers need to pay attention to the practice of remedying income-based urban-rural


Introduction
Urban expansion in the developing world has been dramatic (Jedwab et al., 2017). From 1950 to 2018, urban population in less developed regions has increased more than tenfold, from approximately 0.3 billion to 3.23 billion; the proportion of urban population has been more than doubled, from 18 % to 51 % (United Nations, 2019). Developing countries commonly face the problem of a large urban-rural divide during the urbanization process, which has been recognized as one of those major differences between developing countries and developed ones (Fields, 1974;Sahn and Stifel, 2004). The urban-rural income gap (URIG) is defined as the gap between urban and rural household incomes, which is frequently used as an important economic indicator delineating urban-rural divide in many studies, such as Selden (1988); Carter (1997), and Tian (2001).
implications for research and policy . Furthermore, an analysis of URIG in the context of China could be potentially helpful for other developing countries around the world.
Since 2004, there has been an increasing number of empirical studies on the association between urbanization and URIG in China. However, empirical results were oftentimes contradicting, and there has been no consensus regarding those driving factors yet. On the one hand, urbanization may be helpful to narrow URIG because the decrease of agricultural share and the expected income gap between urban and rural areas help to transfer the labor force from agricultural to non-agricultural industries. If the surplus rural labor force can be substantially included, the average income of the urban labor force will decrease due to a more competitive labor market. Therefore, the income level of the rural labor force will increase due to the decline of the rural surplus labor force and enhanced overall productivity. Consequently, the URIG will be narrowed, and the living standard of urban and rural residents will tend to be equal (Stark, 1988). Such an argument is already supported by some empirical studies (Chen and Jiang, 2014;Yang et al., 2013).
On the other hand, studies have revealed that the discriminatory institutional arrangements during the process of urbanization contribute to enlarging the URIG in China. These institutions have hindered the development of the labor market, for example, the household registration system has separated population and labor between rural and urban areas; the social security system has excluded rural residents from its welfare policy (Cai, 2007). The widening URIG is also considered to be caused by the urban-biased development policy, which extracts more than it invests in rural areas (Carter, 1997;Tian, 2001).
Despite some reviews summarizing the driving factors enlarging URIG in China (e.g., Cheung, 2012;Wang et al., 2013;Cai et al., 2019), there is a lack of comprehensive review of the relationship between urbanization and URIG. Importantly, substantial heterogeneity amongst empirical works makes it difficult to formulate appropriate policy recommendations. For example, empirical studies have adopted different data sources, indicators of urbanization and URIG, control variables, estimation methods, temporal span, etc. Against this backdrop, it is necessary to synthesize existing evidence. Meta-regression analysis (MRA) is explicitly designed to integrate econometric estimates and provide objective and comprehensive summaries of econometric findings (Stanley and Jarrell, 2005). To the best of our knowledge, this work is the first attempt to use meta-analysis in understanding the relationship between urbanization and URIG in China. Specifically, this paper employed MRA to identify the sources and levels of heterogeneity in empirical studies on the association between urbanization and URIG in China.
The rest of this paper is structured as follows. Section 2 presents a brief overview of the MRA method along with the estimation strategies and the procedure of literature selection. The findings of MRA are presented in Section 3. Then, the implications of the findings for further research are discussed in Section 4. Finally, Section 5 concludes this paper.

Meta-regression Analysis (MRA) method
Meta-analysis is a series of systematic methods for integrating quantitative research (Sterne, 2009). MRA is one of the meta-analysis approaches that is used to integrate econometric estimates, and it can provide an objective and comprehensive summary of quantitative research. To avoid potential "abuse" of the MRA in economics (Nelson and Kennedy, 2009) and ensure the objectivity of the MRA results, we followed the widely recognized reporting guidelines for MRA in economics proposed by Stanley et al. (2013). The workflow of the MRA (Fig. 2) included three steps: (i) selection and review of the primary studies, (ii) integration of the meta-dataset that connects measures of study heterogeneity sources to the meta-regression model, and (iii) analysis of variables with significant effects.

Literature search and review
A concrete definition of urbanization is needed before the literature search. In this study, we have adopted a principal-agent definition of urbanization suggested by OECD: "urbanization means an increase in the proportion of the population living in urban areas; the process by which a large number of people become permanently concentrated in relatively small areas, forming cities" (OECD, 2008).
By focusing on the dimension of population, studies without such are excluded in this meta analysis. By using this principal-agent definition of urbanization, we finally excluded 3.8 % of articles.
Literature published both in Chinese and English were searched from five academic research databases. We used the keywords "城市化" or "城镇化" (urbanization), "城乡收入差距" (urban-rural income gap), and "中国" (China) for Chinese literature via the database of China National Knowledge Infrastructure (CNKI), which is the most resourceful and authoritative academic sharing platform in China.Additionally, we used Baidu Scholar as a supplementary source for Chinese literature. For English literature, "urbanization", "urbanrural income gap/disparity," and "China" were used as keywords to search the databases of JSTOR, Taylor& Francis, and Web of Science. Given that the number of empirical studies on the association between urbanization and URIG in China began to increase since 2000, the publication date of journal articles, working papers, book chapters, and thesis with empirical analysis was set between 2000 and 2019.
We screened the title and abstract of 511 studies on urbanization and URIG in China. Then we examined full texts of these studies to exclude: 1) studies without estimates of direct effect-size; 2) empirical studies without statistical indicators (t-statistics or standard errors) necessary for the calculation of partial correlation coefficients (PCCs); 3) studies using measures for urbanization such as industrial structure and urban land scale instead of the population; 4) studies using regional instead of nationwide datasets.
The criteria left us with 29 studies (see the full list in Appendix A) and 94 estimates for the following MRA. Among these 29 studies, four are degree theses, and 25 are journal articles. Only two studies were published in English, and the rest were in Chinese 1 .

Estimators and modeling
Partial correlation coefficients (PCCs) are used widely in MRA (Churchill et al., 2017). In this study, they were calculated to measure the association between urbanization and URIG in China. PCCs are independent of the metrics with which the independent and dependent variables are measured.Therefore, PCCs allow those effect-size estimates reported in primary studies to be comparable. The PCC(r i ) of each selected effect-size estimate can be calculated via Eq. 1: Where t i and df i , respectively, are the test statistics and degrees of freedom reported by primary studies. The standard error (se ri ) of the PCC(r i ), as specified in Eq. 1, represents variations due to sampling error (Eq. 2).
One study may report multiple effect-size estimatesB; therefore, fixed-effect weighted means (X FEW ) of PCCs is calculated (Eq. 3): Where ( se 1/ ri ) represents the precision of each estimate, which can be used as a weight to calculate fixed-effect weighted means at study level (Ugur, 2014). Consequently, a higher weight will be assigned to more precise estimates, which will also reduce the with-in study variations. Appendix B summarizes those fixed-effect weighted means of PCCs in this study.
If primary studies suffer from publication bias (i.e., significant results are more likely to be published), fixed-effect weighted means calculated in Eq. 3 are no longer robust measures. Therefore, we apply the funnel plot and the jointed estimations of the precision-effect test (PET), funnel asymmetry test (FAT), and precision-effect estimation with standard errors (PEESE) to test if publication bias is a significant issue regarding the included primary studies.
The funnel plot is the most commonly used method to visually examine if publication bias exists (Sutton et al., 2000), which is a scatter plot of estimated effect (such as PCCs) against their precision ( se 1/ ri ). PET-FAT (Eq. 4) is used to test if there is a significant publication bias issue Stanley, 2009, 2013): Where t i and se ri are the t -value and standard error of PCCs specified in Eq. 1 and Eq. 2; i is the error term. In the PET-FAT test, while FAT tests . The existence of publication bias can be 1 One possible reason for the low proportion of English literature is that researchers who conduct empirical studies on this issue are mostly Chinese. In fact, the (first) authors of the two English articles included in this study are also Chinese researchers. These studies were mainly published in Chinese journals in the 2000s. Since 2010, there has been a growing number of English publications, but the total number is still very small.
confirmed if it is statistically significant that 0 0 (Stanley, 2008). 0 is the coefficient of precision, indicating the overall effect for a specific research (i.e., urbanization's effect on URIG in this study). A significant PET result indicates that there is a non-linear relationship between the reported effect-size estimates and their standard errors. Therefore, a precision-effect estimation with standard errors (PEESE) proposed by Stanley and Doucouliagos (2012) should be used to obtain a corrected estimate of 0 . Eq. 5 presents the estimation model of PEESE, Furthermore, the multivariate meta-regression (MMR) model enables us to identify heterogeneity sources and the exact effect of each variable contributing to the variation in results of empirical studies (Churchill et al., 2017), which can be estimated in Eq. 6: Where Z ji is a vector of covariates which may account for variation in the evidence base and be considered as potential sources of heterogeneity; i is the error term. We applied stepwise regression as the modeling approach. The PET-FAT, PEESE, and MMR models will be estimated based on the ordinary least squares (OLS). Primary studies regarding one topic usually applied data from the same or similar sources, for which the estimated magnitude of the effect may not be independent (Gerrish and Watkins, 2018). Therefore, we report the heteroskedasticity-robust standard error to partly mitigate such a potential issue (Ugur, 2014).

Variables of multivariate meta-regression (MMR) model
In total, 23 independent variables were included by the MMR model (Eq. 6) to test their effects with t-statistics of 94 effect-size estimates reported by primary studies (Table 1). The dependent variable is the tstatistics reported by reviewed articles.
Different methods are used to construct indicators of urbanization and URIG in reviewed articles. There are two primary methods for constructing urbanization indicators: 1) urbanization based on the ratio between the registered urban population who have urban hukou 2 and the total population; 2) urbanization based on the ratio between the working population in the secondary and tertiary sectors and the total population. There are also two types of URIG indicators adopted by primary studies, namely urban-rural income/consumption ratio (referring to the ratio of per capita income between urban and rural areas and/or the ratio of per capita consumption between urban and rural areas) and inequality index (e.g., Theil index and Gini coefficient).
Urbanization is a comprehensive process, including the transfer of industrial structure from agriculture to the secondary and tertiary sectors, urban expansion, and economic development. Therefore, reviewed articles often include different control variables, such as industrial structure, urban land scale, farmland scale, economic development level, investment in urban fixed assets, fiscal expenditure level, agricultural support policy, openness to trade, financial development level, marketization degree, and human capital level. Additionally, variations in estimation methods, data characteristics, and publication information may also lead to systematically different effect-size estimates. Therefore, we included those as control variables in the MMR model.

Characteristics of the evidence base
An overview of the reviewed articles included in the meta-analysis are presented in Fig. 3 and Appendix B. One prominent finding is that 63 % of studies indicate a negative relationship between urbanization and URIG. In other words, a higher level of urbanization is conducive to reducing the URIG in China. Regarding indicators of urbanization and URIG, the hukou urban population and urban-rural income/consumption ratio were used more frequently. About 79 % of articles used urban 2 Hukou is the household registration system in China. Urban residents with urban hukou have the right to access public welfare provided through employment in state-owned enterprises, while rural residents only have the right of land usage and a share in the collective village economy for a minimum subsistence.
hukou population to measure urbanization, and about 77 % of studies used the urban-rural income/consumption ratio to measure URIG. Furthermore, about 59 % of studies applied the OLS method; 15 % of them used spatial econometric models.Also, a relatively smaller number of studies applied Two-Stage Least Squares (2SLS), Three-Stage Least Squares (3SLS), and Generalized Method of Moments (GMM). Finally, more than half of the articles were published between 2010 and 2015. Control variables included by primary articles (Fig. 4) suggests that more than half of the articles included the effect of industrial structure, openness to trade, economic development level, and fiscal expenditure level as control variables. On the contrary, it is less common to include the effect of variables like the investment in urban fixed assets, financial development level, human capital level, agricultural support policy, marketization degree, urban land scale, and farmland scale.

Publication bias test
If no publication bias exists, the funnel plot (Fig. 5) should be symmetric and reflect more variations in PCCs with low precision and vice versa (Stanley, 2008). Following this criterion, some degree of publication bias can be observed in Fig. 5. Therefore, we have further employed the PET-FAT and PEESE. Table 2 presents the results of both PET-FAT and PEESE. The constant term ( 0 ) is not statistically significant, which indicates that the effect of publication bias is not significant. Importantly, the coefficients of precision ( 0 ) are statistically significant in both PET-FAT and PEESE, which indicates the overall effects of urbanization on URIG is significant.

Multivariate meta-regression (MMR) model results
The MMR model was applied to estimate the exact effect of covariates contributing to the variations in empirical results. Tables 3 and 4 presents our MMR results. While positive coefficients indicate the effect of urbanization on enlarging URIG (i.e., an adverse effect), negative values suggest the effect of urbanization on narrowing URIG (i.e., a beneficial effect).
In Table 3, the first column reports the unconditional effect size of    Notes: **p < 0.05, *p < 0.10; Heteroskedasticity-robust standard errors in parentheses. Positive coefficients indicate the effect of urbanization on enlarging urban-rural income gap; negative coefficients indicate the effect of urbanization on narrowing the urban-rural income gap.
precision ( se 1/ ri ) on URIG, which represents the average effect of urbanization on URIG from primary studies. The unconditional coefficient on precision is -0.050 and is statistically significant (p < 0.1), which echoes our finding that the majority of primary studies indicate a negative relationship between urbanization and URIG. Columns 2-5 examine how different urbanization and URIG indicators would affect their relationship. We tested four different indicators (i.e., two different urbanization measures and two different URIG indices) separately because primary studies oftentimes used only one indicator to measure urbanization and one to represent URIG. Results show that precision is statistically significant in all columns except Column 3, which suggests that the marginal effect of urbanization on URIG is condition on how urbanization is defined and measured.
Columns 2 and 3 include urbanization indicators used by primary studies, namely working (measured by the ratio between the working population in the secondary and tertiary sectors and the total population) and resident (measured by the ratio between hukou population and total population). By adding these control variables, magnititude of precision decreases from -0.050 to -0.047 and from -0.050 to -0.046, respectively. Nevertheless, the coefficients of the two urbanization indicators are insignificant by themselves.
In Columns 4 and 5, we tested the effects of different URIG indicators. Results show that studies have reported the opposite finding regarding the effect of urbanization on URIG. Specifically, when measuring URIG via the urban-rural income/consumption ratio (i.e., the ratio of per capita income between urban and rural areas or the ratio of per capita consumption between urban and rural areas), urbanization tends to enlarge URIG (p < 0.01). Conversely, when measuring URIG via inequality index (e.g., Theil index or Gini coefficient), urbanization tends to narrow the URIG (p < 0.01). The contradictory results suggest that how URIG indicators are constructed plays a critical role in the effect of urbanization on URIG. Table 4 presents the MMR results regarding different control variables, estimation strategies, data characteristics, and publication date. The coefficients of precision are negative in all three columns and those coefficietns of precision are significant except in Column 8. Notably, the magnititude of urbanization on URIG is conditioning on the set of control variables and the choice of estimation method.
Column 6 includes eleven control variables applied in primary studies. Among all those variables with a positive effect, while Financial significantly influences the relationship between urbanization and URIG (p < 0.05), the rest covariates (i.e., industry, economy, investment, and fiscal expenditure level) are statistically insignificant. Such findings may indicate that primary studies controlled for these variables tend to report the adverse effect of urbanization on URIG. The coefficients of the rest control variables are negative, where the coefficients of farmland, agricultural support, and human are statistically significant. Column 7 includes five different estimation methods adopted by primary studies. None of the coefficients regarding estimation methods is statistically significant, suggesting that the choice of the estimation strategy may not be an essential factor in understanding the relationship Notes: ***p < 0.01, **p < 0.05, *p < 0.10; Heteroskedasticity-robust standard errors in parentheses. Positive coefficients indicate the effect of urbanization on enlarging urban-rural income gap; negative coefficients indicate the effect of urbanization on narrowing the urban-rural income gap. Notes: ***p < 0.01, **p < 0.05, *p < 0.10; Heteroskedasticity-robust standard errors in parentheses. Positive coefficients indicate the effect of urbanization on enlarging urban-rural income gap; negative coefficients indicate the effect of urbanization on narrowing the urban-rural income gap.
between urbanization and URIG. Finally, Column 8 examines the roles of the range of datasets and the year of publication. As none of the coefficients associated with these variables is statistically significant, the temporal span of the empirical data and publication information seems not so critical in understanding the relationship between urbanization and URIG.

Discussion
This study used MRA to understand urbanization's effect on the URIG in China by synergizing studies quantitatively, which is different from the previous qualitative review articles, where general viewpoints and research contents of reviewed studies were summarized qualitatively (cf., Zeng and Hu, 2007;Tan et al., 2016;Colsaeta et al., 2018). With the tools of MRA, we found supportive evidence that urbanization is, on average, associated with narrowing URIG, based on the evidence that all the eight models of our meta-regression show the effect of urbanization on narrowing the URIG, and six coefficients associated with precision are significant, including the unconditional model for precision (i.e., Column 1, 2, 4, 5 in Table 3 and Column 6, 7 in Table 4). While China has its own characters in the urbanization process, there are many similarities between China and other developing countries when it comes to the rural-urban divide (Dong and Putterman, 2000;Cai, 2007;Tan et al., 2016). Therefore, studies of China's urbanization and URIG would have positive implications for other developing countries.
Our analysis shows that different measures of URIG have significant influence on the relationship between urbanization and URIG. Studies using urban-rural income/consumption ratio and inequality indices to measure the URIG tend to report contradictory findings regarding such relationship. Specifically, the coefficients of ratio reveal that urbanization is positively associated with URIG. Conversely, the coefficients of inequality indices illustrate that urbanization is negatively associated with URIG. One possible explanation is how these URIG indicators are calculated. The urban-rural income/consumption ratio is widely used because it is easy to calculate. However, compared to the inequality index based on the Theil index and/or Gini coefficient, it is inadequate to reflect urban-rural income inequality because the ratio itself does not reflect the distribution of urban and rural populations (Wang and Ouyang, 2008). We found that about 77 % of primary studies have used the urban-rural income/consumption ratio to measure URIG, while the inequality index was only chosen by 23 % of studies. Our analysis reveals that measures of URIG are critical in understanding the relationship between urbanization and URIG. Therefore, future research can make a more elaborated comparison of the results generated by urban-rural income/consumption ratio and inequality index.
Previous studies have shown that in developing countries, urbanbias in policy contributes to the widening rural-urban gap, through which urban industrial prosperity is created at the expense of rural sectors and rural dwellers (Lofchie, 1997;Fesselmeyer and Le, 2010). On the one hand, our analysis has confirmed that major national policies (e.g., control variables of financial development, agricultural support policy, human capital level, and farmland scale in regressions) have a significant effect on the results reported by primary studies. One the other hand, we did not find China's development policy is always urban-biased. Specifically, we found a significant positive effectof financial development on URIG, where five out of seven primary studies controlled for financial development have revealed a positive relationship between urbanization and URIG. As public policy is often biased toward urban areas, financial development which often supports urban development and facilitates the urbanization process (Kim, 1997;Cho and Boggess, 2003) may only benefit urban areas, therefore, enlarges the income inequality (Lou, 2008;Zhang and Zhu, 2010). Nonetheless, empirical evidence also finds that financial development has narrowed URIG in China (Clarke et al., 2003). Since 2004, the Chinese government has increased financial and fiscal investment in rural villages, rural enterprises, and rural farmers (nong cun, nong ye, nong min, or san nong), including tax-exemption policies to increase the income of farmers (Li, 2017). It is consistent with our finding that there is a significant negative effect of the agricultural support policy on the association between urbanization and URIG. However, only less than one-fourth primary studies have estimated the effect of agricultural support policy. More research can be conducted to understand the role of such a policy.
Additionally, the significant negative effect of human capital level, which is often measured by the number of college students or the average years of education, reveals that the enhancement of human capital plays a critical role in narrowing URIG. There are mainly two explanations. Firstly, the Chinese government has increased investment in rural education and improved the level of human capital in rural areas (Li, 2015). Secondly, although the quality of education in rural areas is relatively low, most young people in rural areas choose to work in urban areas at an earlier age to improve their income, thus reducing URIG (Qin and Liu, 2016). While the policy implication of these two explanations is consistent, that is, to increase the investment in rural education, we suggest more analysis of the long-term effect of education on the human capital level as well as URIG in future research.
Furthermore, we found a significant negative coefficient on the farmland scale. Farmland is an important source of income for most farmers. The more farmland they have, the higher the possibility for them to realize scale management and yield more income (Yang et al., 2013). Meanwhile, while it is argued that urban land scale might enlarge URIG during the process of urbanization (Wang et al., 2015), our regression results find that urban land scale is statistically insignificant. However, we suggest that the potential role of urban land scale cannot be ignored as it refers to the ongoing debate about China's existing urban-rural dual land tenure system (Ye et al., 2018). The transformation of agricultural land into urban construction land is a flagship embodiment of China's urbanization process (Zhong et al., 2011;Feng et al., 2019;Chen et al., 2019;Huang and Li, 2020). At the same time, studies have found that farmers' rights and interests were compromised during the urbanization process (He et al., 2014;Wang et al., 2020). We argue that the protection of farmers' rights and interests should be paid more attention in both policy design and academic research to promote urban-rural equality and reduce URIG.
The findings regarding control variables also have several policy implications for remedying income-based urban-rural inequality during the urbanization process. With the analysis of those three control variables with significant effects, we argue that it is necessary to improve the supply of financial resources for agricultural activities and expand financing channels for rural households (Zhong et al., 2019). Besides, the government should spend more fiscal expenditure on rural human capital by investing more in the development of rural education. In the meantime, reform of employment and household registration system should be accelerated to alleviate institutional obstacles of migration and establish a unified urban-rural labor market (Fields and Song, 2020). Last but not least, during the agricultural land to nonagricultural land conversion, policy needs to focus on the long-term protection of farmers' rights and interests, such as giving reasonable compensation and providing employment guidance .
This meta-analysis work provides synergy of previous studies regarding the effect of urbanization on URIG from a quantitative perspective. st Of course, there are some caveats and limitations of this study. First, this research draws on studies on the average effect of urbanization and URIG in China. Regional differences in such a relationship reported in some studies (Cao, 2010;Liu et al., 2013) are not taken into account. Studies using the data of regional areas of China could be included in MRA in future research to provide more nuanced evidence. Second, while some scholars argue that URIG is an important driving factor of urbanization (Lall et al., 2006;Zeng and Zuo, 2013), we did not analyze the effect of URIG on urbanization due to limited empirical studies in the reverse direction.

Conclusion
This article synergizes primary studies on the relationship between urbanization and URIG in China. Generally, empirical studies tend to report that urbanization narrows URIG. Our MRA reveals the heterogeneous effectsfrom 94 effect-size estimates in 29 studies. The jointed interpretation of PET-FAT and PEESE tests shows that publication bias is not a significant concern of those included primary studies. The results of MMR show a general negative association between urbanization and URIG in China. However, we found more nuanced evidence on the impact of different URIG indicators and set of control variables. More specifically, urbanization shows a positive effect on URIG amongst articles where the urban-rural income/consumption ratio is adopted to measure URIG. By contrast, urbanization is negatively associated with URIG amongst articles where inequality index is used to measure URIG.
These findings suggest that research priorities of urbanization's effect on URIG in China are twofold. First, future analysis can be conducted to understand how and why different indicators play a role in the association between urbanization and URIG. Second, different factors (e.g., financial development, agricultural support policy, human capital, farmland, and urban land scale, etc.) can be examined in detail to understand the pathways and mechanisms related to URIG. Additionally, factors related to urban spatial structure (e.g., Liu and Wang (2016) and Liu et al. (2018)) may provide nuanced implication regarding the relationship between urbanizaiton and URIG. For example, Wang et al. (2019b) found a mixed relationship between polycentric urban strucutre and economic productivity. More recently, Wang (2020) revealed that urban polycentricity is associated with a higher quantity and diversity of urban consumption places.
Based on the empirical findings, this work has the following policy implications regarding remedying income-based urban-rural inequality during urbanization. First, the supply of financial resources for agricultural activities can be improved, and financing channels for rural households can be expanded. Second, fiscal expenditure on rural human capital can be increased through investing more on rural education in conjunction with removing institutional obstacles of migration. Third, actions are needed to protect farmers' rights and interests in the long run, especially those whose agricultural land would be converted to non-agricultural land during the urbanization process.