4.1. Examining Spatial Autocorrelation with ESDA
Since the reform and opening up in 1978, economic development has become the central task of the Chinese government. During the reform period, the exchange of production factors—including labor, capital, resources, and technologies—between regions became increasingly frequent. This has led to a transformation of the regional economy, from self-governing to interdependence. We calculated global and local Moran’s I to investigate the change in the spatial dependence of total factor productivity from 2000 to 2010, in an attempt to capture such a change. The results of global Moran’s I and the responding significance tests with different spatial weight matrices are reported in
Table 2.
It can be seen that the Moran’s I values of total factor productivity are significantly positive for both years, indicating there is a strong similarity or dependence between the total factor productivity of neighboring cities in these two years in China. The difference between using a different weight matrix, however, remains negligible, which suggests the spatial autocorrelation of productivity in China is the result of increased interconnectedness among cities after the economic reform, regardless of how a neighborhood is defined. When looking at the change from 2000 to 2010, we find that the Moran’s I of total factor productivity becomes larger, which means a general trend of increasing spatial autocorrelation in these ten years (increased interconnectedness). The other interesting finding is that the difference in Moran’s I in 2010 with different weighting strategies is less than that in 2000. In other words, the impact of geographical distance on the spatial dependence of productivity becomes smaller in 2010. This might be explained by the improvement in China’s transport facilities leading to the decrease in the impediment effect of geographical distance, which is a relatively important factor in the exchange of technologies, talents and production factors. In 2000, the road mileage in China was 1.4 million km and railway mileage 58,700 km. These two figures increased to 4 million km and 90,000 km in 2010. Moreover, the high speed railway construction project was launched in 2008, and its mileage reached beyond 5000 km in 2010, which promoted a more integrated regional productivity landscape.
The local Moran’s I is mapped in
Figure 1 and
Figure 2. From the figures, we can see that the number of L-L clustered cities in 2010 is less than that in 2000, which is opposite to the H-H type. The rapid growth in total factor productivity in Xinjiang has led to the change from a gradient pattern decreasing from east to west in 2000 to a sandwich like pattern in 2010. The grand western development program initiated in 2000 helped Xinjiang develop at a high speed. In addition, Xinjiang is a province with net immigrants in western China and the number of net immigrants was more than 1.5 million in 2010. Xinjiang also has 10,635 people with junior college and above degrees per 100,000 persons in 2010, ranking sixth among the 31 provincial administrative units in mainland China. Population immigration and the large proportion of people with higher education might be a critical factor for Xinjiang to promote its productivity. We will examine the mechanisms for productivity growth in the following section.
4.2. Regression Results
For the regression analysis, we conducted a regular OLS estimation with the four demographic factors, namely, population size (POP), aging of workers (OLD), human capital stock (HCS) and migration (MIG), and other control variables, and report the results for 2000 and 2010 in
Table 3. The Lagrange multiplier (LM) tests were then conducted with different weight matrices and reported in
Table 3.
Results from
Table 4 reveal that there is significant spatial autocorrelation in the OLS’s regression residuals. As the LM tests suggest, the likely cause for the spatial autocorrelation in the regression residuals is from the spatially autocorrelated dependent variable. According to selection criteria proposed in Anselin [
28] and Elhorst [
68], the spatial lag autoregressive model might be a more appropriate alternative, indicating a strong spatial dependence and connection among neighboring cities in terms of productivity. We then present the spatial lag estimation results in
Table 5. The additional models’ results are presented in
Table 6.
By observing the results from
Table 5 and
Table 6 and comparing them with
Table 3, we have a few interesting results. First, as shown by the log likelihood and AIC values in
Table 5, the spatial regression models have significantly improved over the OLS models, regardless of which spatial weight matrix we use. The log likelihood values increase by nearly 30 (40) in 2000 (2010) and AIC values decrease by nearly 50 (70) in 2010. The ρ values in spatial lag models are significant at 99% confidence level, indicating that a strong spatial autocorrelation does exist and the total factor productivity in the surrounding regions has a strong positive diffusive effect. The spatial autocorrelation tests for the spatial lag model’s residuals suggest the residuals are no longer autocorrelated, which means that the spatial effect is well dealt with by adding the spatial lag terms. In addition, compared with the results of OLS models (
Table 3), the coefficients of the independent variables have changed in different directions in the spatial lag models. For example, the coefficient of population size (POP) turns to positive in the spatial lag model for 2000 and the coefficient of migration (MIG) becomes smaller in the spatial lag models for both years. The correction of the coefficients by the spatial regression models provides potentially a more realistic understanding of the impact of each explanatory variable.
Second, population size seems to have no effect on productivity growth in both years. The population size in China is large. The average population size of cities in 2010 is 3.72 million. The least populous city is Shennongjia in Hubei Province, which had more than 70 thousand people in 2010. The reason why population size does not improve productivity might be that, in the post-Malthusian period, innovation and technological progress mainly come from the secondary and tertiary industries rather than agriculture, which do not depend on population size. Another explanation, from Klasen and Nestmann [
26], is that population density, instead of size, plays an important role in knowledge creation and diffusion, market expansion and technological progress. We will examine this argument in further exploration. On the contrary, human capital stock (HCS) is found to have a significant positive impact on productivity growth in both years. The coefficient in 2010 is larger than that in 2000, indicating an increasing growth effect of human capital stock on productivity. The rapid accumulation of human capital might be largely due to the expansion of China’s higher education. According to official statistics, the annual enrollment in higher education in China has increased from 3.90 million in 2000 to 9.56 million in 2010, an increase of 2.5 times within a decade. The other two population factors, aging and migration, have changed their impact on productivity growth with varying symbols and significance levels. As shown in
Table 5, the coefficient of aging workers (OLD) is negative in 2000 and is not significant with the simple spatial contiguity matrix (
). However, in 2010, the coefficient becomes positive and significant at a 5% level with both weight matrices, indicating the increase in the proportion of older workers in the labor force, and they became a significant driving effect on productivity growth. As argued by Ang and Madsen [
42], the productivity growth effects of elderly workers with higher education are substantially higher than those of their younger counterparts, due to the accumulation of experience and knowledge. Compared with their counterparts in 2000, the elderly workers born between 1951 and 1960 have a better chance to gain higher education. As the college entrance examination system in China was interrupted from 1966 to 1976 and was not restored until 1977, people born between 1941 to 1950 were often less exposed to higher education. As a result, the elderly workers in 2010 might have higher creativity and management quality to improve productivity than their counterparts in 2000. This change in the impact of different cohorts on productivity has also been observed in the United States [
14].
Third, the coefficients of migration (MIG) become larger and significant in 2010, though they are positive in 2000, but not statistically significant (
Table 5). This might suggest that migrants in 2000 are mainly unskilled surplus labor from rural areas (the typical “
Nongmin Gong”, or peasant workers) and they often work in labor intensive industries, such as clothing and construction, which contributes little to productivity improvement. In 2010, the proportion of skilled workers, such as high tech talents and college students, has increased substantially. Migration becomes an important way of human resource reallocation. Migration factor in 2010, hence, generated a significant impact on local productivity. In summary, human capital stock might be the most important population factor for productivity growth, and the impact of other factors largely depends on it.
Fourth, as for the other control variables, some meaningful findings emerge from the regression results as well. Fixed asset investment (INV) shows a significant negative effect in both years, which might be a result of China’s economic policies changing during that time. For a long time after the reform and opening up, China served as a “world factory.” China’s economic growth has been mainly dependent on massive investments of labor, physical capital, mineral and energy resources. The development pattern with high material consumption is obviously harmful to productivity growth in 2000. In 2008, to ease the downward pressure on the economy caused by the world financial crisis, the Chinese government put forward the four trillion investment fiscal policy. The strong stimulus policy successfully completed the task of “maintaining the growth.” Firm investment became less efficient due to the increase of bank loans and government subsidies [
69]. The problems of a rigid economic structure, high energy consumption and high pollution in China have not been effectively addressed, and still play a restraining role in the productivity growth. The change in the symbol of government expenditure (GOV) and the change in the significance of industry structure (SEC or TER), might also be attributed to the change in economic policies during this period, for similar reasons. The openness (FDI) is always positive at the 99% confidence level, which is in line with previous studies [
9,
10]. In today’s globalized world, openness promotes market expansion, technology introduction and talent exchange between countries, and it plays an increasingly important role in productivity growth.
Fifth, from the model summary in
Table 6, compared with their corresponding OLS models, the log likelihood values become larger and AIC values become smaller in all spatial regression models, indicating the spatial models fit the data better. The direction and significance of the coefficients of most independent variables—including the same population variables, control variables and spatial lag term—in Model 2–4 are consistent with those in Model 1. This suggests that the findings are relatively robust and credible. However, despite the addition of the squared term, the population size is always insignificant concerning affecting productivity growth, which suggests that the inverted-U effect of population size on productivity growth does not exist at the prefecture level in China. On the other hand, the estimation results of population density and its squared term in Model 3 and Model 4 show that population density does affect productivity growth and that there is an inverted U-shaped relationship between them. In other words, when population density is below a threshold point, a higher density improves productivity growth. Once population density reaches and passes the threshold point, higher density exerts a negative impact on productivity growth. This result suggests that the concentration of the population, instead of the sheer number of population, is one of the primary demographic factors in driving the prefectures’ productivity growth. The coefficient of human capital inequality has changed from significantly negative in 2000 to insignificantly positive in 2010. The Chinese government attaches great importance to education and has invested heavily to raise the national education level. In 2006, China fully implemented the nine year free compulsory education system to promote fairness and eliminate inequality in education. This is also reflected in the change in the human capital Gini coefficient, which has decreased from 0.250 in 2000 to 0.213 in 2010. Apparently, increased human capital equality practically removed the negative effect it had on TFP in the early 2000s, though the potential positive effect still needs to fully manifest in the future.