^{1}

^{1}

Based on the data of China’s 28 provinces and regions during 1987-2007, this paper is trying to explore the role of the composition of the human capital on the economic growth with the spatial effect among provinces. The results indicate that there are significant spatial effects of human capital and economic growth in China; there is a significant positive correlation between the GDP per capita and the average schooling years; when the education of human capital is divided into primary education, secondary education and higher education, only the higher education and the primary education show significantly positive correlation with the GDP per capita.

It is commonly agreed that high-level human capital can improve labor productivity, and furthermore enhance the overall national strength. For a country, the most important problem of education development strategy is to decide what kind and what level of education should be developed, and also it closely relates to the efficiency and effectiveness of public expenditure. The relevant theoretical problems are: what is the relationship between education and economic growth? What kind of education has the greatest contribution to economic growth? For example, the Chinese government began to give priority to the development of education especially higher education from 1990s, and the number of freshmen admitted by college entrance exam increased with an average annual growth rate of 15.2%. Is this education development strategy beneficial to economic growth?

With the development (Romer, 1986, 1987) [

One reason is that space is not considered. It is well known that large-scale population movements provide rich laborforce, promote the development of the economy and society, lead to the proliferation and spread of technology, and produce social economic effects. Each province is no longer an independent economic entity. Instead, all provinces have become an economic community and have mutual influence on each other. Thus, the space among provinces is an indispensable factor which should be considered in both theoretical research on human capital and economic growth.

In addition, the selection of indicators of human capital is a key issue affecting the determination of the relationship between human capital and economic growth. Becker [

This paper attempts to examine the relationship between China’s human capital and economic growth by using different variables and constructing different weight matrixes. In this paper, educational human capital is divided into the capital of primary education, secondary education and higher education, and meanwhile health human capital is taken into consideration to study the effect of different types of human capital. This dissertation applies geography weight matrix which reflects provincial location adjacency. Owing to space spillover effects caused by the geographical distance among China’s eastern, central and western regions, the distance attenuation matrix is set and spatial lag model and spatial error model for specific analysis are applied as well.

Rey and Mountuori (1999) [

Chinese scholars believe that there is significant positive correlation between China’s human capital and economic growth. As a carrier of knowledge expansion and technology progress, human capital has strong spatial spillover effect on regional economic growth. Researchers (Xiao Zhi-yong [

With regard to the studies of human capital in China, most of empirical literatures treated human capital as a homogeneous concept. Some papers incorporated human capital in the growth accounting and found a positive relationship between growth and education (Fleisher & Chen [

In summary, a huge progress has already been made in both the direction and the depth of the research. There is still much room for research on spatial econometric model.

First, this paper plans to use spatial lag model and spatial error model which are commonly used to conduct the research. In doing so, the most important thing is to set spatial structure and an appropriate spatial structure weight matrix is the key. In current empirical studies, most of researchers would use spatial adjacency to set weight which is very simple. However, the spatial spillover of human capital is not limited only between adjacent areas. Interactions also exist between locations that are not adjacent but close. Therefore, this paper not only set geographic weighting matrix considering the adjacency of the geographic location, but also set distance decay matrix considering that the eastern, the central and the western regions are actually three overalls, and that the spatial effect caused by geographic distance also exist between these three huge regional economies.

Besides, educational human capital is divided into primary education, secondary education and higher education in this paper. The author is trying to investigate whether these three types of education have spatial spillover effect; and if it does, is it positive or negative? The results could offer a reference for government’s policies on education and educational resource distribution. By conducting empirical analysis with those two space models and those two spatial weight matrix models mentioned above, the real relationship between human capital, spatial spillover and economic growth can be studied more accurately.

In addition, this paper also considers educational human capital and health human capital. As an important part of human capital, the impact of population’s health status and its change on economic growth should not be neglected.

The data used in this paper involves population and economic variables of 28 provinces in Chinese mainland over the period of 1987-2007. Although there are 31 provincial-level regions in mainland China, Hainan and Tibet are excluded for missing data, and Because Chongqing became a municipality city drawn from Sichuan Province in 1997, Chongqing and Sichuan are accounted as one province.

The form in this paper is different from the original Solow model since it has introduced education and health human capital in the economic growth model. Adopting production function of Cobb-Douglas, we can get the expanded Solow model consisting with human capital, whose form is:

where Y represents total output, A represents technology stock, K represents physical capital stock, H represents human capital stock which consists educational human capital E and health human capital M.

where

This can also be written as:

where y represents per capita GDP of labor population, k represents per capita physical capital of labor population, h represents per capita human capital which consists of the average years of schooling and life expectancy.

Taking (4) into (3), we can get:

where m represents per capita health capital,

where i represents the i-th province, t represents the t-th year. I refers to the investment of the year, and its indicator is the gross fixed capital formation (data from the National Bureau of Statistics Database: http://www.stats.gov.cn); δ represents the economic depreciation rate (calculate as 9.6%); Additional capital investment for the year will be converted to a constant price, using the price index of investment in fixed asset published by “China’s Statistical Yearbook”.

Health human capital is measured by life expectancy (M) in this paper. As China’s Statistical Yearbook only contains data for every ten years, so this paper applies liner interpolation method to obtain the estimates. In this article, the average year of schooling is used to measure educational human capital. Each provincial data of the average level of education and data of the proportion of the higher educated, the secondary educated and the primary educated are provided by Chen Zhao and other scholars’ research results.

In accordance with China’s education system, we divided the workforces who are beyond age of 15 into 5 groups according to their educational level. Namely, the population of tertiary and higher education H1, the population of high school education H2, the population of junior high school education H3, the population of primary school education H4, and the population of illiterate and semiliterate H5. According to the years of schooling, the author sets weights for each group. Specifically: 0 for illiterate and semiliterate, 6 for primary school education, 9 for junior high school education, 12 for high school education, and 16 for tertiary and higher education. At last, calculate the total stock of human capital with weighted sum method. It can be expressed as:

Total stock of human capital = (human capital stock of higher education H1 + human capital of high school education H2 + human capital of secondary education H3 + human capital of primary education H4 + human capital of illiterate and semiliterate H5) = Σ (the population of different education level × weight) = 16 × H1 + 12 × H2 + 9 × H3 + 6 × H4 + 0 × H5.

Dividing the total stock of human capital by the number of educated people at different level, we can get the average stock of human capital (the average years of schooling for working population). Expressed in formula is:

The average stock of human capital = the total stock of human capital/the population of different education level.

Using China’s provincial administrative panel data, this paper is trying to construct a spatial econometric model to interpret the correlation between China’s provincial economic growth and human capital. We assume that the economic growth of each province in China depends not only on the level of human capital and physical capital, but also relies on the human capital level of adjacent regions, that is to say, both education human capital and health human capital have spatial “spillover effect” either within the region or between regions. Moran I index is a common method to test whether spatial autocorrelation exist between region variables. In order to determine whether spatial autocorrelation exist between the variables, this paper tested all the variables with Moran method before the empirical spatial econometric analysis. The formula of Moran I is:

where, _{i} represents the observed value of the i-th region, n is the total

number of regions,

The specific variables of Moran I index are shown in

year | GDP per capita | Life expectancy | Physical capital | The average years of schooling | Primary education | Secondary education | Higher education |
---|---|---|---|---|---|---|---|

1987 | 0.0949*** | 0.1026*** | 0.0846*** | 0.1089*** | 0.0879*** | 0.1369*** | 0.0916*** |

1988 | 0.0926*** | 0.1049*** | 0.0844*** | 0.1083*** | 0.0879*** | 0.1370*** | 0.0916*** |

1989 | 0.0904*** | 0.1073*** | 0.0844*** | 0.1083*** | 0.0879*** | 0.1371*** | 0.0916*** |

1990 | 0.0822*** | 0.1097*** | 0.0660** | 0.1126*** | 0.0730*** | 0.1329*** | 0.0760*** |

1991 | 0.0785*** | 0.1121*** | 0.0842*** | 0.1083*** | 0.0879*** | 0.1374*** | 0.0916*** |

1992 | 0.0798*** | 0.1144*** | 0.0842*** | 0.1083*** | 0.0879*** | 0.1375*** | 0.0916*** |

1993 | 0.0817*** | 0.1166*** | 0.0914*** | 0.1085*** | 0.0759*** | 0.1322*** | 0.0968*** |

1994 | 0.0847*** | 0.1187*** | 0.0848*** | 0.1083*** | 0.0879*** | 0.1377*** | 0.0916*** |

1995 | 0.0866*** | 0.1207*** | 0.0862*** | 0.1085*** | 0.0792*** | 0.1321*** | 0.0906*** |

1996 | 0.0895*** | 0.1224*** | 0.0784*** | 0.1045*** | 0.0838*** | 0.1396*** | 0.0806*** |

1997 | 0.0900*** | 0.1240*** | 0.0853*** | 0.1006*** | 0.0880*** | 0.1289*** | 0.0923*** |

1998 | 0.0887*** | 0.1253*** | 0.0877*** | 0.1015*** | 0.0902*** | 0.1337*** | 0.0909*** |

1999 | 0.0888*** | 0.1264*** | 0.0816*** | 0.1042*** | 0.0970*** | 0.1388*** | 0.0844*** |

2000 | 0.0901*** | 0.1272*** | 0.0802*** | 0.1076*** | 0.0955*** | 0.1192*** | 0.0867*** |

2001 | 0.0901*** | 0.1271*** | 0.0714*** | 0.1067*** | 0.0955*** | 0.1327*** | 0.0794*** |

2002 | 0.0901*** | 0.1270*** | 0.0778*** | 0.1077*** | 0.0961*** | 0.1371*** | 0.0795*** |

2003 | 0.0907*** | 0.1268*** | 0.0626** | 0.1032*** | 0.1024*** | 0.1321*** | 0.0714*** |

2004 | 0.0916*** | 0.1263*** | 0.0559 | 0.0997*** | 0.1059*** | 0.1389*** | 0.0631** |

2005 | 0.0943*** | 0.1258*** | 0.0724*** | 0.1062*** | 0.1027*** | 0.1302*** | 0.0793*** |

2006 | 0.0943*** | 0.1252*** | 0.0644** | 0.1003*** | 0.0951*** | 0.1235*** | 0.0737*** |

2007 | 0.0947*** | 0.1244*** | 0.0580* | 0.0946*** | 0.1000*** | 0.1148*** | 0.0676** |

Note: ***, **, * respectively denotes the significance level of statistical measurement test results 1%, 5%, 10%.

According to different reflection methods of model setting for “space”, there are two kinds of spatial econometric models: SAR and SEM. The first model (SAR) is mainly used to study situations when the behavior of neighboring institutes or regions has an impact on other institutes or regions within the whole system. The specific formula is:

W is spatial weight matrix of N-N order, namely a matrix which shows the web-structured relationship of several (N) institutes or regions. The weight matrix W need to be standardized to make the sum of each row in the weight matrix equals 1. W_{y} is the spatial lag of the dependent variable,

In the second model (SEM), the relationship between organization or regions are reflected through the error term. This model is often used when the interaction between organizations or regions varies with different relative positions. The specific formula is:

where y is the dependent variable, X is exogenous explanatory variable matrix,

This article uses two kinds of spatial weights matrix to do empirical analysis, the first one is the geo-spatial weight matrix; the weight of geographic adjacent area is 0, while non-adjacent areas is 1; the second one is the distance decay weight matrix, which uses distance decay function to construct spatial weights matrix. The specific formula is:

where, i, j, represent any two different regions;

Taking spatial matrix of all regions in china as an example, the form is as follows:

where,_{ }respectively represents the spatial weights matrix of 28 regions over the period of 1987-2007. Since the space distance is calculated based on latitude and longitude, therefore,

In this paper, four aspects were considered while constructing the spatial econometric model. First, we divided human capital into two categories and constructed two models respectively. One model measured the educational human capital by the average years of schooling. Another model subdivided the educational human capital into primary education, secondary education and higher education. Besides, the spatial econometric panel model setting consists of SAR Panel Data and SEM Panel Data; in addition, the spatial weights matrix setting includes distance decay weights matrix and geographic weights matrix; at last, the error component of spatial panel data model includes unfixed effect, spatial fixed effect, time fixed effect and temporal-spatial fixed effect. In summary, there can be 2 × 2 × 2 × 4 = 32 aspects.

Models were separated into two categories in this paper. First, we consider the model which measures the educational human capital by the average years of schooling. More details can be seen in

Measuring the educational human capital by the average years of schooling,

Test method | Number of samples | Test value | Probability |
---|---|---|---|

Lmerr | 588 | 945.2212 | 0 |

Lmsar | 588 | 890.4653 | 0 |

Moran | 588 | 0.3575 | 0 |

Lratios | 588 | 208.8682 | 0 |

Walds | 588 | 304.1064 | 0 |

Model | Variable | Un-fixed effect | Spatial fixed effect | Time fixed effect | Time-spatial fixed effect | ||||
---|---|---|---|---|---|---|---|---|---|

Two kinds of spatial weights | Distance decay weights | Geographic weights | Distance decay weights | Geographic weights | Distance decay weights | Geographic weights | Distance decay weights | Geographic weights | |

Spatial lag panel data model | Constant term | −23.99*** | 23.72*** | −23.37 | −23.11 | −27.45 | −21.09 | −26.47 | 18.86 |

Life expectancy | 9.21*** | 9.23*** | 9.17*** | 9.20*** | 8.20*** | 8.37*** | 8.43*** | 8.64*** | |

Physical capital | 0.66*** | 0.65*** | 0.68*** | 0.67*** | 0.55*** | 0.55*** | 0.61*** | 0.61*** | |

The average years of schooling | 0.23 | 0.31 | 0.01 | 0.10 | 1.60*** | 1.54*** | 1.17*** | 1.05*** | |

W*dep. var. | −0.06 | 0.13*** | 0.03 | 0.11*** | 0.41*** | 0.41*** | 0.35*** | 0.62*** | |

R-squared | 0.82 | 0.82 | 0.84 | 0.84 | 0.89 | 0.89 | 0.91 | 0.91 | |

sigma^2 | 0.17 | 0.17 | 0.15 | 0.15 | 0.10 | 0.10 | 0.09 | 0.09 | |

log-likelihood | −321.70 | −318.27 | −289.65 | −286.78 | −167.08 | −166.69 | −127.02 | −121.96 | |

Spatial error panel data model | Constant term | −24.44*** | −23.58*** | −23.88 | 23.17 | −24.27 | −25.28 | −23.74 | −24.51 |

Life expectancy | 8.46*** | 8.12*** | 8.63*** | 8.38*** | 8.30*** | 8.63*** | 8.52*** | 8.78*** | |

Physical capital | 0.58*** | 0.56*** | 0.63*** | 0.62*** | 0.56*** | 0.57*** | 0.62*** | 0.63*** | |

The average years of schooling | 1.43*** | 1.61*** | 1.04*** | 1.16*** | 1.61*** | 1.47*** | 1.16*** | 1.04*** | |

Spat.aut. | 0.814*** | 0.790*** | 0.846*** | 0.785*** | 0.424*** | −0.974*** | 0.448*** | −0.985*** | |

R-squared | 0.804 | 0.803 | 0.826 | 0.825 | 0.895 | 0.895 | 0.908 | 0.908 | |

sigma^2 | 0.103 | 0.116 | 0.089 | 0.102 | 0.102 | 0.097 | 0.089 | 0.084 | |

Log-likelihood | −186.316 | −218.553 | −140.419 | −180.283 | −166.249 | −157.139 | −125.451 | −114.951 |

Note: ***, **,* respectively denotes the significance level of statistical measurement test results 1%, 5%, 10%.

In all cases, the absolute values of the log-likelihood function estimated by the equation are relatively large. The largest absolute value is in the unfixed effect model and the smallest is in the temporal-spatial fixed effect model. The R-squared which represents the goodness-of-fit of the model is significant high in temporal-spatial fixed effect. Besides, the goodness-of-fit of spatial lag panel data in space fixed effect is higher than it in time fixed effect. In spatial error panel data, the situation is the same. The R-squared of temporal-spatial effect model almost remains the same either in spatial lag panel model or spatial error panel model.

Since estimates of the W*dep.var which marks the spatial dependence and estimates of the Error Spat.aut. both reached a significant level of 1%, it further confirmed the spatial correlation among data. On the whole, in panel data models, the coefficient estimates based on distance decay weights matrix is larger than the ones based on geographic weights matrix. (For example, in the unfixed effect model which belongs to the spatial error model, the value of Spat.aut. based on distance decay weights is 0.814, yet the Spat.aut. value based on geographic weights is 0.790) This probably because that the spillover and diffusion of human capital are limited by geographical positions in some degree, and they (the spillover and diffusion effect) will decay with the increase of spatial distance (Keller [

According to the four spatial lag regression models, the R-squared value and the Log-likelihood value affected by spatial and time fixed effect shows significant improvement. Meanwhile, other regression variables are also significant, even the spatial lag term passed the test. In light of this, we can conclude that not only spatial fixed effect, but also time fixed effect exists between regions. It also shows us that some ignored factors such as the regional production technology, management abilities, have strong diffusion effect on surrounding areas. In the spatial error regression model, the fitting efficiency of temporal-spatial fixed effect is the optimal among the four kinds of effects.

In the spatial lag panel model and the spatial error panel model, the coefficients of explanatory, namely the average years of schooling, physical capital and life expectancy, show significant positive influence. The average years of schooling is positively correlated with economic growth. This conclusion is similar with the results of Marcelo Soto, Daniel Cohen [

Then consider models of the second category. Among the explanatory variables, educational human capital is measured by primary education, secondary education and higher education. The specific results are shown in

Based on results in

Diving educational human capital into primary education, secondary education and higher education,

In spatial lag panel data model, the R-squared values of temporal-spatial fixed effect are 0.920 and 0.922

Test method | Number of sample | Test value | probability |
---|---|---|---|

Lmerr | 588 | 511.0945 | 0 |

Lmsar | 588 | 518.1999 | 0 |

Moran | 588 | 0.2629 | 0 |

Lratios | 588 | 124.8417 | 0 |

Walds | 588 | 302.3148 | 0 |

model | Variable | Unfixed effect | Spatial fixed effect | Time fixed effect | Time-spatil effect | ||||
---|---|---|---|---|---|---|---|---|---|

Two kinds of spatial weights | Distance decay weights | Geographical weights | Distance decay weights | Geographical weights | Distance decay weights | Geographical weights | Distance decay weights | Geographical weights | |

Spatial lag panel data model | The constant term | −33.574*** | −33.142*** | −33.825 | −33.381 | −40.211 | −34.161 | −40.657 | −33.536 |

Life expectancy | 6.817*** | 6.844*** | 6.968*** | 6.993*** | 7.362*** | 7.541*** | 7.586*** | 7.815*** | |

Physical capital | 1.154*** | 1.156*** | 1.074*** | 1.081*** | 1.114*** | 1.111*** | 1.079*** | 1.061*** | |

Primary education | 2.976* | 2.426 | 4.471*** | 3.906** | 4.797*** | 4.727*** | 6.452*** | 6.379*** | |

Secondary education | 0.716*** | 0.696*** | 0.822*** | 0.797*** | 0.046 | 0.082 | 0.194 | 0.250* | |

Higher education | 12.215*** | 12.194*** | 11.888*** | 11.905*** | 11.299*** | 11.270*** | 11.304*** | 11.152*** | |

W*dep.var. | −0.027 | 0.0830** | 0.008 | 0.0545 | 0.389*** | 0.401*** | 0.353*** | 0.564*** | |

R-squared | 0.863 | 0.864 | 0.881 | 0.882 | 0.906 | 0.906 | 0.920 | 0.922 | |

sigma^2 | 0.136 | 0.135 | 0.118 | 0.118 | 0.093 | 0.094 | 0.079 | 0.078 | |

log-likelihood | −248.499 | −246.658 | −206.012 | −205.029 | −139.718 | −139.188 | −92.285 | −87.379 | |

Spatial error panel data model | The constant term | −37.184*** | −36.661*** | −37.994 | −36.918 | −37.485 | −38.143 | −38.217 | −38.972 |

Life expectancy | 7.469*** | 7.109*** | 7.724*** | 7.336*** | 7.487604*** | 7.755*** | 7.708*** | 7.938*** | |

Physical capital | 1.101*** | 1.201*** | 1.056*** | 1.127*** | 1.111*** | 1.064*** | 1.076*** | 1.050*** | |

Primary education | 5.051*** | 4.580*** | 6.807*** | 5.998*** | 5.126*** | 5.387*** | 6.804*** | 7.2611*** | |

Secondary education | 0.136 | −0.125 | 0.261* | −0.288** | 0.052 | −0.060 | 0.205 | −0.231 | |

Higher education | 11.397*** | 11.961*** | 11.317*** | 11.746*** | 11.371*** | 11.096*** | 11.376*** | 11.246*** | |

Spat.aut. | 0.753*** | 0.726** | 0.801*** | 0.681*** | 0.447*** | −0.978*** | 0.423*** | −0.985*** | |

R-squared | 0.850 | 0.850 | 0.869 | 0.876 | 0.9043 | 0.904 | 0.9186 | 0.919 | |

sigma^2 | 0.095 | 0.106 | 0.079 | 0.092 | 0.093 | 0.088 | 0.079 | 0.074 | |

log-likelihood | −159.449 | −186.034 | −106.236 | −145.823 | −138.491 | −129.591 | −90.541 | −77.399 |

Note: ***, **, * respectively denotes the significance level of statistical measurement test results 1%, 5%, 10%.

respectively, and the log-likelihood values are −92.28 and −87.37. Compared to the un-fixed effect, spatial fixed effect and time fixed effect, these values are the greatest, which indicates that the temporal-spatial fixed effect model has better goodness-of-fit than any other effect model. Therefore, this paper lays a lot of emphasis on the results of temporal-spatial fixed effect model.

From the temporal-spatial fixed model in spatial lag panel data model, we can find out that the coefficient of primary education and higher education are both positive, and that p values are both below the significance level of 5%. Therefore, we can conclude that there is a significant positive correlation between these two different levels of education and economic growth. Where, the coefficient of higher education in distance decay matrix and geographical weights matrix are 11.304 and 11.152 respectively. The coefficient of secondary education is not significant in distance decay matrix, while it is significant in geographical weights matrix as its value is 0.250. The coefficient of primary education in distance decay matrix is 6.452, while 6.379 in geographical weights matrix. The results indicate that the workforce who received higher education contributes the most to economic growth, and workforce who received primary education comes second while workforce who received secondary education contribute the least. There are several reasons for this result. First, as one kind of “mature” labor force, workforce who received higher education can adapt to the liquidity faster and better, which is good for the technology innovation for the inflow area, moreover, they have better abilities in learning new technology, which further promotes economic growth. As for the workforce who only received primary education, they are usually engaged in labor-intensive industries due to their low education level. The liquidity increases since they have to move around to get better opportunities, and finally attributes a lot to economic growth. According to Cai Fang’s research, rural migrant workers are usually not the ones who have received secondary or higher education, because rural workforce who have received better education have already had a good occupation in rural areas, so the motive to migrant to other places is weak. This explained why the spatial spillover effect of secondary education is not significant.

The result of spatial error model is similar to the result of spatial lag model. It is also presented as a result that the temporal-spatial fixed effect model is the best. The coefficient and significance of the three different education levels agree with the ones in spatial lag panel model, from which we can conclude that higher education contributes the most to economic, primary education comes second and secondary education contributes least.

Through the spatial statistics based on data provided by China’s 28 provinces, municipalities and autonomous regions over the period of 1987-2007 and the empirical results of econometric analysis, we can conclude that: the spatial dependence between Chinese human capital and economic growth is statistically significant; among the empirical results of different fixed effect model settings, the magnitude and direction of correlation between human capital and economic growth vary, not only including the positive impact of the development of human capital in time dimension, but also involving the negative impact of regional differences of human capital in space dimension; through the spatial lag model, we can see that the provincial human capital will affect the provincial economic growth, and the effect will be further passed to the economic growth of the neighboring provinces or even the whole region. Compared to the traditional statistical model, spatial statistical model and spatial panel data model can reveal the spatial autocorrelation and spatial clustering of human capital much deeper since it includes both the vertical movement and the horizontal differences in time and space dimensions.

In order to further promote the free flow of human capital among the eastern region, central region and western region of China, to realize the rational allocation of human capital, to make full use of the positive spatial spillover effect of human capital, and finally to contribute to economic growth, the government should make efforts from the following aspects:

First, the government should make full use of spatial spillover effect and promote the coordinated regional development of human capital. Only flow capital can produce benefits, so does human capital. The flow of human capital directly affects its operational efficiency. Seeing from the perspective of space transfer, a certain transaction cost must be paid if human capital is about to migrate or flow to other regions or industries. However, this cost is also a kind of “investment” because the migration for better opportunities is the payment for higher “future income”, and this kind of flow has influence on the operational efficiency of human capital. Meanwhile, liquidity also denotes the merit and maturity of the economic system. According to the analysis above, we can see that the spatial mobility of workforce will always have positive impact on economic growth regardless of the different education levels. Currently, the household registration system reform hasn’t been completed yet. When rural labors move out for off-farm employment, they can’t transform their identity to citizenship at the same time. Therefore, the movement of rural labors is inadequate and incomplete, which is not conducive to motivate the enthusiasm of rural labors. Moreover, it also leads to the co-existence of labor mobility and income disparities. So, the government should continue the reform of the household registration system, break down the spatial mobility barrier of elements, product and workforce, and resolve the spatial structure problem through the interaction between people and material. Specifically, first of all, the household registration system reform must continue under the auspices of central government and a series of unified household registration management laws and regulations must be established so that the comprehensive and effective management can be conducted. Population migration and flow are individual’s rational economic behavior since it is a measure that an individual takes to seek for better opportunities. However, due to the imbalanced distribution of public goods in local areas, the population flow is also imbalanced, which requires more financial support for the backward regions from the government so that the supply of public goods can be improved. The government also needs to narrow the inter-regional differences at maximum and avoid some population flow that is not necessary. Meanwhile, the government should use some economic leverage to regulate the population flow to large cities, such as tax and the price of real estate. Namely, under the premise of non-discrimination, the government should use macroeconomic measures to increase the cost of migration, in case the density of urban population becomes too high and the healthy development of the city gets affected. As for the population that has already migrated to other regions, government should ensure they have equal welfare with local people, and also ensure that their accumulation can be transferred to the resettlement place. By doing so, migrant labors will be motivated to work harder in the new places, and the probability of occurrence of mass incidents will be reduced, which has positive impact on the maintenance of social stability and harmony.

Besides, while making policies of human capital investment, government should take spatial correlation into consideration, pay attention to the spatial association between the central western inland areas and the eastern coastal areas, and strengthen the economic interaction between the eastern, the central and the western areas. Compared to the eastern region, the central and western regions are at a disadvantage in economic development, employment environment and development opportunities, which not only makes it difficult to attract high-qual- ity talents from outside, but also makes it hard to keep their own talents. To solve this problem, it is imperative to establish a long-term mechanism of human capital mobility and introduction. It is also important to improve both hard and soft conditions for talent development and growth through the implementation of a variety of policies. On one hand, we can introduce foreign talents into our country. But more importantly, we should build a long-term mechanism of nurturing internal talent to inject new vitality to our economy. The central government should further strengthen the financial support for the central and western regions, such as increasing transfer payments, supporting the education in rural underdeveloped areas and ethnic minority areas, strengthening the key areas and weak links, and solving outstanding problems.

All in all, as to promote the development of Chinese human capital and achieve economic growth by making full use of the spatial spillover effect of human capital, we need to establish a multi-level policy framework from government and society down to individuals, and all relevant policies and measures must be combined with the determinants of economic and social development of human capital.