LONG LIVE KEJU! THE PERSISTENT EFFECTS OF CHINA’S CIVIL EXAMINATION SYSTEM

LONG LIVE KEJU! The persistent effects of China's civil examination system (Ting Chen et al) – Intensive Reading

Overview Methodology

This paper studies the continuous impact of the ancient imperial examination system on contemporary human capital.

The most noteworthy thing in this paper is the selection of instrumental variables. The shortest river distance from each county to the nearest bamboo and pine and cypress production areas is selected as the instrumental variable of the strength of the imperial examination.

This paper also explores the possible path of the influence of the ancient imperial examination system on contemporary human capital.

Model

$$y_i=\beta kej{u}_i+c_1{X}_i^c+c_2{X}_i^h+a_p+{ϵ}_i$$

Among them, $i$ means county,$y_i$Then it represents the logarithmic average education years of each county, $kej{u}_i$Indicates the logarithm of the density of Jinshi obtained by each county during the Ming and Qing Dynasties (1371-1905) (density = the total number of people who obtained Jinshi divided by the total number of each county to standardize, with 10,000 as the unit), that is, ln(1 + jinshi /population), $a_p$Indicates the provincial fixed effect;

Control variable $X_i^c$: time of night lighting (economic prosperity), distance from county center to coast (diffusion of Western technology), roughness of terrain (key historical events);

Control variable $X_i^h$: Population density, urbanization rate, and potential agricultural output (a proxy variable for historical economic prosperity), excluding people who have been admitted to Jinshi but whose birthplace is different from the place of examination (in order to control regional migration, because they tend to go to places where the examination is easy) );

Baseline results

• Column (1): only control the province fixed effect, the result is significant
• Column (2): further control $X_i^c$, the result is significant
• Column (3): further control $X_i^h$, the result is significant (main result) For every 10,000 people, doubling the number of Jinshi during the Ming and Qing Dynasties would lead to a 6.9% increase in the contemporary average years of education.
• Column (4): Samples that exclude regional migration, the results are still significant;

Columns (5)-(8) replace the proportion of uneducated, proportion of elementary and middle school educated, proportion of high school educated and proportion of university educated with different explanatory variables. For the first two, the results are negative and significant, but not as expected. For the latter two, the results are positive and significant, with a larger coefficient for college than for high school.

Spatial autocorrelation in the residuals

The authors argue that the error terms between neighboring counties may be correlated, and thus the long-term effects of the imperial examinations may be biased.

标准做法是： Conley (1999) standard errors adjusted for two-dimensional spatial autocorrelation(reported in brackets)，in addition to clustering the standard errors at the province level (in parentheses)；

Additional tests: Kelly (2019), Moran test for spatial autocorrelation tests the residuals of column (1) and column (4), restricting to only 5 adjacent counties, the results (z-score of the Moran test) are -0.659 (column (1)) and -0.486 (column (4)), rejecting the null hypothesis at the 5% significance level that there is no spatial autocorrelation.

Tests of Artificial Spatial Noise: Kelly (2019), assume a spatial interpolation
that is normally distributed with the covariance matrix… The elements of the covariance matrix are related to the distance between counties. The parameters of these noises are estimated using county per capita GDP data. These noises are then used to replace the explanatory and explained variables, respectively.

Panel A：

• Column (1) shows the proportion when a noise variable has better explanatory power than the original persistent variable.
• Column (2) indicates that at the 5% significance level, the noise variable has 5.4% explanatory power for contemporary years of education;
• Column (6) indicates that at a significance level of 5%, the imperial examination density has an explanatory power of 4.8% for noise variables;

Panel B is the result of adding control variables.

In summary, in the residuals, the likelihood of a persistent effect of Jinshi density driven by spatial autocorrelation is low. That is, the influence of spatial autocorrelation is excluded.

Robustness check

1. The counties along the Yangtze River Delta are rich in Jinshi, and the regression results may be dominated by these counties. Excluding the 30 counties along the Yangtze River Delta, the results are still stable.

2. Because the data of the entire Ming and Qing Dynasties are used, the impact of different periods may be different, so all the data are divided into groups at 50-year intervals, and regression is performed separately to see the size of the coefficients. (The influence was small before the 17th century, but it reached a peak in the early and middle Qing Dynasty, and decreased due to internal and external troubles in the later period)

3. In terms of explanatory variables, in addition to the density of Jinshi, there are also similar densities of recruits and students. The conclusions are consistent when Juren density is used as an explanatory variable, but the effect is not as good as that of Jinshi (when Juren and Jinshi are added at the same time, Jinshi is significant but Juren is not); The number of students in the county remains relatively stable, so it cannot reflect the county's imperial examination level, so it is not significant. Column (4) puts the three explanatory variables together. In the end, only Jinshi is significant in 'horse race' , indicating that the density of Jinshi is the most suitable variable to measure the culture of imperial examinations.

(From the distribution map, the distribution of Jinshi and Juren is very similar, while the distribution of students is mostly in the backward areas of the southwest, for the sake of regional balance...)

endogeneity test

For the reasons that may lead to endogeneity, it is obviously impossible to be reciprocal causation, because the explanatory variable is the Jinshi density long ago, and the explained variable is the average years of education in the present (2010). So the possible cause of the endogeneity problem is the omitted variable problem. This paper uses the instrumental variable method to solve this problem.

• Instrumental variable selection: the shortest river distance from each county to the nearest bamboo and pine and cypress production areas
• Rationale: The Imperial Examination Examination of the Four Books and Five Classics is too difficult, many reference books and explanations are needed to understand the subtle differences, one needs to learn how to write stereotyped essays, the production of these books is closely related to the printing factory; the main printing center is near the pine and bamboo habitat Locally, the components required for printing are transported through tributaries of major navigable rivers. (The figure below shows the distribution of bamboo and pine and cypress with printing centers)

Then the author made some correlation regressions to show the correlation between instrumental variables and various variables (printed books and Jinshi density, instrumental variables and printing centers, instrumental variables and printed books, and instrumental variables and Jinshi density), indicating that instrumental variables are not weak. instrumental variable.

The authors argue that the habitats of bamboo and cypress are formed according to geographical features and are strictly exogenous (there is only a small amount of evidence that they were planted because of commercial printing). It is therefore independent of economic prosperity and uncorrelated with the omitted variables affecting modern mean years of schooling. In order to test this statement, the author made some regressions (instrumental variables and commercial center, tea center, silk center, population density, urbanization rate, suitability of agricultural products, terrain, drought, distance to commercial center, distance to provincial capital distance, etc.)

Then the author divides the time into 50-year intervals to examine the impact of instrumental variables on the density of Jinshi in different periods. From the 15th century to the middle of the 19th century, the impact of instrumental variables on the density of Jinshi was significant, and then became less and less significant (the main reason It was the new printing technology that we adopted in the 20th century).

The results of the instrumental variable regression are as follows, reporting the simple form and the two-stage least squares form, focusing on column (6), with a coefficient of 0.085, which is slightly higher than the previous use of Jinshi density measurement (previously 0.069)

The Possible Channels of Imperial Examinations Affecting Contemporary Human Capital

Transmission of Keju Culture

The authors use 6 interview questions as explained variables:

1. Do you think education is "the most important determinant of social status"
2. Do they "prefer their government to prioritize education spending"
3. how many years of education their children should (ideally) have
4. Do they often give up watching TV to spend time with the kids
5. Total hours per week they spend on homework tutoring their children
6. Whether parents communicate effectively with (their) children

Using patrilineal or maternal ancestor Jinshi density as an explanatory variable (measured by matching surnames with counties), while controlling for heritability (using memory, logic test scores as control variables) controlling for education level, family income, age, gender, rural or urban etc. The result is as follows

In addition, the author also used the children's sample to perform regressions on math and literacy test scores, absenteeism, and time spent on learning per week, and the results were consistent with expectations.

Educational Infrastructure

The author uses the Confucius Institute, a key educational institution in ancient times, as the explained variable, and also selects the number of modern primary and secondary schools and universities as the explanatory variable. It can be seen that Jinshi density has a marginally significant effect (OLS) on colleges, while the IV estimator is not significant. It is significant for the number of primary and secondary schools in 1900 but not significant for 2010 (mainly because of the popularization of compulsory education in 1986) and significant for universities both in 1947 and 2010.

Social Capital

The author believes that the officials who passed the Jinshi examination are likely to jointly create a large amount of social capital by providing public goods and organizing charitable activities. Clans and family ties formed by consanguinity have the primary goal of providing public goods and a social safety net for their members. So the author uses the number of genealogy compiled to represent the strength of a family (resourceful families tend to revise their genealogy more frequently to strengthen the sense of belonging and honor), the number of charitable organizations (relief from the famine, orphanages), the number of non-profit organization as the explained variable. The results are as expected.

Social Elites (Political Elites)

Political elites can also influence county test results by influencing the distribution of hometown educational resources (such as the size of colleges). The author uses the number of people with official positions above the provincial governor as the explained variable (currently it is membership of the Central Committee—the party's highest level of political organization). Jinshi density had a significant effect on the number of Qing officials, and this effect persisted even after the fall of the imperial regime in 1911. This effect was insignificant in the subsequent period, when the political elite survived the deadly Long March or fought in various patriotic guerilla warfare, and in (1949–77), during the initial land reform and agricultural collectivisation or the Cultural Revolution that targeted the traditional elites.

Imperial Examinations and Intergenerational Mobility (Social Mobility)

Over time, the accumulation of higher human capital may have facilitated intergenerational mobility simply by raising the average educational level of various social groups. But inheritance will inhibit intergenerational mobility (for rich people, the imperial examination will strengthen the inequality between these social groups, resulting in a decline in social mobility), in order to explore the impact of the imperial examination on intergenerational mobility, the author uses the 2005 census The data analyzes the flow of education and income. Both Jinshi density and family background have a significant positive impact on the average years of education. Columns (2)-(3) In counties with strong imperial examination influence, parents' educational background has less influence on individuals' educational outcomes. (In fact, what the author wants to express is that education is conducive to class leapfrogging)