The political origin of differences in long-term economic prosperity: centralization versus decentralization

Abstract

It is important to understand the political determinants of long-term economic prosperity. In Sichuan Province of China, there historically existed two long-standing political institutions: a centralized system, represented by the prefecture system, and a decentralized system, represented by the Tusi system. This study utilizes a unique historical dataset for 181 counties in Sichuan Province and implements a regression discontinuity design to estimate the impact of different institutions on long-term economic development. The results indicate that the Tusi system significantly suppressed the performance of the local economy and impeded current regional openness and urbanization. By tracing the channels of its persistent effect, we document that the influence of these historical institutions has persisted mainly through the shaping of human capital and state capacity. The Tusi system is not only harmful to cultural communication and local educational progress but also inhibits the government’s capacity and the formation of authoritarianism.

Appendices

Appendix 1

Figure 18 in Appendix 1 shows the historical changes in China’s central dynasties and the TS.

Fig. 18

The historical changes in the TS

Appendix 2

Appendix 2 shows the establishment of the Tusi units in the Sichuan area of the Yuan dynasty and the number of counties under TS and PS jurisdiction. The outcomes are shown in Table 10 and Fig. 19.

Table 10 Establishment of the Tusi units in the Sichuan area of the Yuan dynasty

Fig. 19

Number of counties under the TS and PS in Sichuan

Appendix 3: OLS and instrumental variable estimations

To further confirm the conclusions of RD strategy, we follow two approaches. First, we use the policy of the TTPS during the Ming and Qing dynasties as a natural experiment. As the central government feared that centralization in the southwestern regions would be hindered by the complex geographical environment, we compare the samples within the group under the governance of the TS to resolve the potential bias of the geographical environment. The TTPS exactly provided an exogenous shock for this purpose. We find that in the same geographical environment, the samples that did not receive the TTPS still have lower current income than those that received the TTPS. Second, using the IV method, we choose the number of stone bridges that existed in the early years of each county as an instrumental variable of the political institutions.

We use OLS to make a preliminary inference on the relationship between political institutions and counties’ current income. The specific model is as follows:

$$Y_{it} = \alpha + \rho {\text{Manage}}\_{\text{tusi}}_{i} + \varepsilon_{it}$$

(5)

where Y_it is the focus of our concern, indicating the proxy variable for income, and ${\text{Manage}}\_{\text{tusi}}_{i}$ indicates whether the sample area was once ruled by the TS, i.e., if political decentralization is implemented in the jurisdiction, the value is 1; otherwise, 0. i is a sample of the county, and ε_it represents a random disturbance term. Based on model (5), we also add control variables X_it to control other factors. The new model is as follows²³:

$$Y_{it} = \alpha + \rho {\text{Manage}}\_{\text{tusi}}_{i} + \varphi X_{it} + \varepsilon_{it}$$

(6)

By using model (6), we try to obtain the net effect of TS as much as possible. However, it is difficult for us to control all factors that may affect both Y_it and ${\text{Manage}}\_{\text{tusi}}_{i}$ at the same time. To solve this problem, we establish a 2SLS for regression analysis.

The first stage of the model is as follows:

$${\text{Manage}}\_{\text{tusi}}_{i} = \phi + \beta {\text{bridge}}_{i} + \varphi X_{it} + \varepsilon_{it}$$

(7)

The second stage is as follows:

$$Y_{it} = \alpha + \rho {\text{Manage}}\_{\text{tusi}}_{i} + \varphi X_{it} + \varepsilon_{it}$$

(8)

where we choose the number of bridges that existed in the early years of each county as the instrumental variable. We choose this variable because the early construction of stone bridges required a large manpower investment and advanced bridge technology that were not available in the Tusi autonomous region. Therefore, this variable meets the requirements for the instrumental variable.

Baseline regression Then, we use OLS to analyze the long-term effect of decentralization; the results are shown in Table 11. Columns (1)–(3) examine the effect of political decentralization on GDP, while (4)–(6) analyze the influence on GDP per capita. After controlling for other factors, the effects of decentralization on GDP and GDP per capita are − 1.2 and − 0.2, respectively, and they show a negative relationship that is significant at the confidence levels of 1% and 5%. The results show that, whether GDP or GDP per capita is used as an indicator of economic outcomes, decentralization is significantly negatively related to long-term economic development. In addition, the results show that the development of industrialization has a significant role in promoting economic development. At the same time, the development of the private economy has made important contributions to China’s rapid economic performance in the past decades, and this effect is significant at the 1% confidence level.

Table 11 The OLS return of political decentralization to counties’ current income

Based on this result, another concern is that historically some counties experienced a transformation of the political institution and were transformed from the TS to the PS under the pressure of the central government (e.g., the TTPS policy); does this transformation affect local income? To test this, we manually collect the data which are relative to the year that counties transferred to the PS from the TS, making a total of 53 queries. Based on the year of policy transfer, we further divide the sample into two categories. First, if the county experienced transformation during the Ming dynasty and before the Ming dynasty, it is given the value 1; if otherwise, 0; Second, if the county experienced transformation during the Qing dynasty, it is given the value 1; if otherwise, 0.²⁴ These two variables are used to illustrate the length of time of the transformation from TS decentralization to PS centralization in all counties²⁵; the results are shown in Table 12. The counties decentralized before the Ming dynasty experienced less hindrance to current economic prosperity than other counties paying allegiance to the Qing dynasty. It supports our claim that centralized rule is indeed conducive to economic development.

Table 12 OLS on current income in political decentralization and centralization reforms

Although the above analysis suggests that the decentralization represented by the TS has hindered counties’ current income, in fact, this political institution may be affected by the geographical environment and early economic factors. The central dynasties were more likely to conquer the geographically flat regions with convenient transportation and economic development and set up the PS, but they may have voluntarily given up the regions with poor geographical conditions. This shows that Sichuan’s political institutions may come from differences in geographical conditions; that is, the differences in the objective environments of the east and west create institutional differences. To test this, we map the geographical distribution of altitude, number of rivers and number of mountains in each county in Sichuan, as shown in Figs. 20, 21, and 22. These figures reveal obvious differences in the natural environment of the western and the eastern regions; these may be the root cause of the early institutional differences and then lead to bias in our results. How can we remove the influence of this sample selection bias? We select the counties that had the TTPS policy before 1840 as the control group and assign them the value 0.²⁶ Then, the sample of counties that had not undergone the TTPS policy before 1840 is used as the treatment group and given the value 1. Both the treatment group and the control group are from the TS region. If the political institutions in Sichuan differ due to the geographical environment and early economic factors, then these two groups should have almost no difference in geographical environment and economic development conditions. Overall, the specific results are shown in Table 13, which indicate political decentralization still has a significant negative impact.

Fig. 20

Geographical distribution of altitude in Sichuan

Fig. 21

Geographical distribution of the number of rivers in Sichuan

Fig. 22

Geographical distribution of the number of mountains in Sichuan

Table 13 Results of the internal comparison in decentralized regions

The Instrumental Variable We choose the number of stone bridges that existed in the early years of each county as the instrumental variable and adopt the 2SLS method to obtain models (7) and (8). The results are shown in Table 14, where we examine not only the entire sample from 2003 to 2013 but also the impact of political decentralization in 2005 and 2013. The first-stage regression shows that the number of stone bridges as an instrumental variable is significantly negatively related to whether the local county is under the TS, which indicates that a county with a large number of stone bridges is more likely to be controlled by the centralized PS, and vice versa. At the same time, the F statistic in this regression is much larger than 10, which means that the instrumental variable is satisfactory. The second-stage regression shows that decentralization acts as a significant impediment to both GDP and GDP per capita, and the results obtained are shown in columns (4)–(9).

Table 14 Instrumental variable regression

Appendix 4

See Table 15.

Table 15 Variable selection and description

Appendix 5: robustness tests

5.1 Continuity of control variables (a): a nonparametric test

See Figs. 23, 24, 25, 26, 27, 28 and 29.

Fig. 23

Distribution of asset investment levels

Fig. 24

Distribution of labor input

Fig. 25

Distribution of industrialization

Fig. 26

Distribution of household savings

Fig. 27

Distribution of education level

Fig. 28

Distribution of financial development

Fig. 29

Distribution of private enterprise development

5.2 Continuity of control variables (b): a parametric test

See Table 16.

Table 16 Continuity test of control variables

5.3 Changes in bandwidth selection size

See Table 17.

Table 17 Effect of bandwidth selection size on RD results

5.4 Changes in bandwidth selection method

See Table 18.

Table 18 Effect of change in bandwidth selection method on RD results

5.5 The spillover effect of political institutions

See Table 19.

Table 19 The test of spillover effect of political decentralization

5.6 Boundary movement

The above analysis uses RD to estimate the persistent effect of political decentralization, but it is uncertain whether there is only one explicit discontinuity or whether multiple discontinuities exist. To test this, we use the placebo test to illustrate that there is only one discontinuity. We assume that there are two discontinuities a and b; that is, the original boundary lines are moved 200 km to the east and west. The specific discontinuity distribution is shown in Figs. 30, 31, 32 and 33, from which we can see that the change in the location of the threshold did not lead to a significant jump in GDP and GDP per capita. We also conduct a more rigorous test and generate the new variables East and West. If the distance_i is greater than 200 km, East takes the value of 1, indicating the treatment sample of counties beyond hypothetical threshold a on the right; if otherwise, 0. In addition, if the distance_i is greater than − 200 km, West has the value of 1, indicating the treatment sample of counties on the right side of hypothetical threshold b; if otherwise, 0. If the regression analysis conducted yields a significant effect, there may be other discontinuities; if the effect is not significant, there are no other discontinuities. The specific results, shown in Table 20, indicate that whether the boundary line is moved to the east or west, the results at the hypothetical threshold are not significant, which fully demonstrates that the above estimated results are reliable.

Fig. 30

Distribution of GDP after moving the discontinuity 200 km westward

Fig. 31

Distribution of GDP per capita after moving the discontinuity 200 km westward

Fig. 32

Distribution of GDP after moving the discontinuity 200 km eastward

Fig. 33

Distribution of GDP per capita after moving the discontinuity 200 km eastward

Table 20 Placebo test

5.7 The counterfactual test

The series of tests and arguments described above demonstrate that political decentralization acts as a significant impediment to local economic prosperity. The concern that there are other random factors that interfere with this deterrence effect is important to address. Therefore, we use a counterfactual method to identify the existence of random factors according to the following approach. We randomly select 58 counties and 20 counties in the centralized and decentralized regions as the imaginary TS sample and generate the variables random1 and random2, respectively. If the samples are randomly selected, they are assigned the value 1; otherwise, 0. Then, they are compared and tested in their respective samples. If a significant result is obtained, it means that random factors in addition to differences in political institutions interfere with the effect; that is to say, the hindrance to economic development is not entirely due to political differences. The results, shown in Table 21, indicate that the hypothetical treatment samples have no significant effect in either the TS sample or the non-TS sample. This indicates that there are no other random factors that hinder current income and further strengthens the importance of the long-term effect of political decentralization.

Table 21 Counterfactual test

5.8 Exclusion of historical events

Like the above factors, major historical events are unlikely to bias the estimated results. To correct potential estimation errors, we remove the major historical events that occurred in Sichuan. First, given that the “Huguang Tianchuan” migration has had a great impact on Sichuan’s population and may have thus affected the current economic performance,²⁷ we remove samples affected by the “Huguang Tianchuan” migration from the full sample. According to the results, shown in Table 22 columns (1)–(6), even when the impact of “Huguang Tianchuan” is excluded, decentralized institutions still significantly hinder local current income.

Table 22 RD results after excluding major historical events

Second, China’s southwestern border has historically been the location of other countries and powers established by ethnic minorities, such as the Dali, Tubo and Nanzhao. The political structure of these countries is similar to the centralized institution of the Chinese central government, which is completely different from the decentralization implemented within the TS jurisdiction. These regimes historically controlled parts of some counties of Sichuan; thus, some counties in the TS sample were also controlled and affected by their centralized political institution, which may affect the results. Hence, we manually search Baidu Encyclopedia to determine whether counties in the TS region were historically controlled by these regimes.²⁸ Furthermore, the samples with “centralized experience” were removed from the TS sample. The regression results, shown in Table 22 columns (7)–(12), show that removing these samples does not change the negative impact of political decentralization.

5.9 Other historical factors

Since the above analysis considers only the impact of present-day factors and ignores the effect of historical events on current economic development, this part focuses on the impact of decentralization controlling for historical factors. We explore the following three aspects: the complexity of the geographical environment, social stability and human capital.

First, the geographical environment of each county not only has a significant impact on early economic development, but also causes development differences in the long run. Therefore, to prevent bias caused by geographical factors, we choose “whether the counties have (mountain) passes,” the number of passes and the number of mountains to measure the geographical environment, and we control for these factors in Table 23 columns (1)–(3). The results indicate that even when controlling for geographical factors, the adverse effect of decentralization on current economic development still exists.

Table 23 The impact of geographical environment and social stability

Second, we also consider social stability as an essential factor to ensure long-term economic development. Because of the many various ethnic groups in Sichuan, armed conflicts and wars historically have been frequent. Therefore, we use military facilities denoted as military post stations (pudi) to represent the social stability of the local area. According to the Qinghui Dian and Guangxu Huidian Cases, in the Guangxu years, there were 13,935 pudi in China and 1426 pudi in Sichuan Province. To convey important military news, the central government set up special traffic routes in various provinces and cities, and to prevent information interruption, the military forces were stationed there and equipped with developed infrastructure public goods, etc. Therefore, the more military facilities there are in an area, the more stable the local social situation. Moreover, we choose the number of conflicts within the counties, the number of conflicts between the PS and TS areas, and the number of deaths in the conflicts to reflect social stability. In Table 23 columns (4)–(7), we control for these variables, and in column (8), we control for them together; doing so does not change the original results.

Finally, we also consider the human capital and institutional factors that may cause bias. We use the construction date of Confucian schools to represent Confucian culture. Considering that the human capital of the TS areas may also be accumulated through other channels, we also collect the number of local schools in each county. In addition to cultural differences, we choose the number of Jinshi, the number of Juren and the number of Jinshi in the Qing dynasty to represent the cultural level of each county. Moreover, we select the date of county founding as the proxy variable for the institutional factor. Then, in Table 24, we control for these variables and examine the impact of decentralization. The results show that the above conclusions remain unchanged.

Table 24 The Impact of Human Capital and Institutional Factors