Appendix: Statistical Appendix for “Leninism and Local Interests”
1.
Data: sources and summary statistics
2.
Constructing and validating the policy autonomy proxy
3.
Which cities get selected to have concurrent appointments?
4.
Robustness checks for baseline results (Table
2)
5.
Regression results by city type
6.
Causal mechanisms: institutions versus individuals
1. Data: sources and summary statistics
This paper relies on four general sets of data. First, data on which cities have concurrent appointments in any given year come a novel collection of data about the makeup of China’s provincial party standing committees (PPSCs) and the personal and career backgrounds of PPSC members, introduced in Bulman and Jaros (
2018). The dataset spans the years from 1996 to 2013, and contains detailed information on 1443 unique PPSC members. Information on the composition of PPSCs was obtained from annual editions of the
China Directory series published by Radiopress, Inc. To supplement these name lists, biographical and career data were gathered from Baidu Online Encyclopedia (百度百科), available at
baike.baidu.com. Where information was unavailable on Baidu Online Encyclopedia, additional searches were conducted using (1) Xinhua Net (
news.xinhuanet.com), (2) China Vitae (
www.chinavitae.com), and/or (3) China Political Elite Database (
http://cped.nccu.edu.tw/). For the purposes of this paper, there are two relevant variables: the concurrent position of PPSC members (except for party secretaries and deputy party secretaries, all other PPSC members tend to have at least one other concurrent party or government position); and, when the concurrent position is a prefectural-level city party secretary post, the name of the city.
Second, to create our policy autonomy proxy, we collect data from a digital archive of content from the
People’
s Daily (人民日报), available at
http://www.oriprobe.com/PeoplesDaily.shtml. As described in more depth in the next section, we use this archive to collect data on the number of
People’
s Daily articles during each year between 1996 and 2013 that contain both the name of a given prefectural-level city and one or more terms related to local reforms and experimentation: reform (改革), innovation (创新), pilot experiment (试点), and model (示范). We use this measure of reform-associated news frequency as a proxy for the degree of “policy autonomy” of different Chinese cities. The subsequent section in this appendix describes in more depth the data collection process as well as the results of a data validation exercise.
The third set of data relates to the coding of city types, and in particular the identification of “border cities” and “red” (Communist revolutionary heritage) cities. For a border city dummy variable, we code as “1” all prefectures that share a land border with another country, based on Google Maps. For “red” cities, we rely on cities’ status as centers of “red tourism” (红色旅游). Since 2005, the Chinese government has officially promoted “red tourism” involving Communist Party historical sites. We code 13 cities as “red” based on the 2010 signing of a “China Red Tourism Cities Strategic Cooperation Yan’an Declaration.” The 13 signatories were Guang’an, Yan’an, Xiangtan, Jinggangshan, Ruijin, Zunyi, Baise, Shijiazhuang, Linyi, Anyang, Yulin, Qingyang, and Huining. Jinggangshan and Ruijin are county-level cities under the administration of Ji’an and Ganzhou, respectively, while Huining is a county in Baiyin (Gansu).
Finally, we draw data on prefecture-level city socioeconomic characteristics from China Data Online, including GDP, FAI, revenue, expenditure, population, FDI, land area, primary sector GDP, and tertiary sector GDP.
Table
3 below summarizes the variables we use in our analysis, including means, standard deviations, and the numbers of observations in both the complete dataset and the more limited main sample we use to conduct our analysis, which excludes certain outliers and limits analysis to a sample with full information for all independent variables. GDP and FAI data are highly variable from year to year in the CDO data, which leads to some extreme outliers when we construct annual growth measures, as well as extreme fluctuations in the independent variables which rely on GDP for their construction (fiscal shares of GDP and economic structure). So to prevent unreliable outliers from driving our analysis, we exclude observations with an absolute value of GDP growth over 50% as well as observations with an absolute value of FAI growth over 100% (approximately 2% of the sample in each case).
Table 3
Summary statistics for full sample and limited analysis sample
Concurrent PPSC appointment | 0.116 | 0.321 | 5712 | 0.133 | 0.340 | 3886 |
News frequency | 25.643 | 46.051 | 5712 | 28.809 | 48.527 | 3886 |
GDP growth % | 18.482 | 48.712 | 4727 | 14.674 | 8.549 | 3886 |
FAI growth % | 33.435 | 70.531 | 4103 | 24.179 | 23.986 | 3886 |
Revenue share of GDP % | 5.990 | 3.537 | 4093 | 5.635 | 3.024 | 3386 |
Expenditure share of GDP % | 13.361 | 9.349 | 4065 | 12.497 | 8.382 | 3356 |
Fiscal gap share of GDP % | 7.462 | 8.183 | 4021 | 6.964 | 7.552 | 3314 |
Log GDP per capita (1000 RMB) | 9.475 | 0.993 | 4693 | 9.430 | 0.897 | 3881 |
Population (millions) | 3.994 | 2.330 | 4705 | 3.989 | 2.324 | 3881 |
Land area | 1.550 | 2.020 | 4586 | 1.555 | 2.048 | 3851 |
Primary share of GDP (%) | 18.260 | 18.047 | 4667 | 17.835 | 13.380 | 3886 |
Tertiary share of GDP (%) | 35.992 | 13.095 | 4663 | 35.667 | 10.647 | 3886 |
FDI (million USD) | 347.893 | 801.6504 | 438 | 304.111 | 679.567 | 3739 |
Autonomous city | 0.211 | 0.408 | 554 | 40.100 | 0.300 | 3886 |
Border city | 0.134 | 0.341 | 5712 | 0.069 | 0.254 | 3886 |
“Red” city | 0.039 | 0.193 | 571 | 20.044 | 0.205 | 3886 |
Capital city | 0.082 | 0.275 | 555 | 40.100 | 0.300 | 3886 |
Deputy provincial city | 0.015 | 0.123 | 555 | 40.018 | 0.132 | 3886 |
For the data we collect ourselves, there is no missing data for any city-year, although there is a possibility that we have missed concurrent leadership appointments given the potential incompleteness of our online sources. However, the CDO data is less complete: as shown in Table
3, approximately one-quarter of FAI, government revenue, and government expenditure observations are missing. We have no reason to believe that these missing observations are systematically biased; rather than a bias toward missing cities, CDO has temporal gaps, and some years have very limited coverage for key variables: over half of government fiscal data is missing from 1999 to 2002, and most FAI data is missing for 2012 and 2013. Imputing values for these missing observations is difficult given that there are frequently multiple years missing in succession and the missing observations often occur at the beginning or end of the sample. However, as data for FDI (one of our IVs) is missing only for 1 year (2009), we do impute 2009 FDI values by taking the simple average of 2008 and 2010 FDI values. This allows us to keep an additional 265 observations. (Excluding 2009 from the analysis makes no difference to the baseline results.)
Given that a substantial portion of the missing CDO data occurs in our different dependent variables, for the main results we choose not to limit the sample to only those observations that have data for all dependent variables. If we limit to a sample that has a full set of observations across all DVs and IVs—thus yielding a consistent number of observations for each regression—our results do not substantively change (see Table
7).
2. Constructing and validating the policy autonomy proxy
To proxy the policy autonomy of localities vis-à-vis their political-administrative superiors, we take an approach similar to that which Malesky (
2008) uses to assess regional policy autonomy in Vietnam. Malesky’s analysis tracks the relative frequency with which different Vietnamese provinces are mentioned in major national-level newspapers in association with episodes of “fence-breaking” or economic policy reform. Because reform experiments and instances of “fence-breaking” reflect local policy initiative and may run against higher-level preferences, Malesky argues that they proxy for local policymaking autonomy. In our own analysis, we also look for indications of local policy experimentation and reform activism. We do not assume that such reform policies necessarily run counter to the wishes of higher-level authorities, but we do regard them as a reflection of local policy activism and we argue that reform experiments by their nature entail an exercise of local policy autonomy.
We search a digital People’s Daily (人民日报) archive to determine the number of People’s Daily articles for every prefecture-level city and each year from 1996 to 2013 that contain a given city’s name and one or more terms related to local reforms and experimentation: reform (改革), innovation (创新), pilot experiment (试点), and model (示范). The inclusion of reform-related terms is meant to avoid excessive noise and mismeasurement due to collecting news stories about unrelated topics (e.g., natural disasters, corruption cases). Our method of collecting these data involved running a single search for each prefecture-year and recording the number of article hits. This exercise provided data for 6372 city-years and flagged a total of 156,753 relevant news articles, making for an average of 24.6 reform-related news mentions per city-year.
Our analysis had to address the fact that several prefectures change their names over time, and many change their administrative status (e.g., from 地区 or 盟 to 市), over the sample period. We search only for city names, not including administrative status (e.g., 苏州 and not 苏州市) to avoid the latter problem. When cities change names, we change search terms to search for the correct city name in any given year. These include:
-
襄阳 (2011–2013), 襄樊 (1996–2010)
-
普洱 (2004–2013), 思茅 (1996–2003)
-
鄂尔多斯 (2001–2013), 伊克昭 (1996–2000)
-
达州 (1999–2013), 达川 (1996–1998)
-
通辽 (1999–2013), 哲里木 (1996–1998)
-
宿州 (1999–2013), 宿县 (1996–1998)
-
吴忠 (1998–2013), 哲里木 (1996–1997)
The more substantive concern is whether our relative news frequency measure is indeed a valid indicator of policy autonomy. Along these lines, we conduct a validation exercise that involves analyzing a subset of the articles identified in the original exercise and determining if they are indeed about (or whether they at least mention) government reforms/innovations/pilots/models in the specified cities. We expect that the content of many but not all of the articles identified by our searches will be related to the city in question or to reform policies (however broadly defined). We selected 157 articles for closer reading by randomizing city-years and summing the number of articles across them until we reached 0.1% of the total number of
People’
s Daily articles flagged in our searches. This resulted in 12 city-years being represented in our close-reading sample. For each of the 157 articles, we collected the following information:
-
Article name: The original article name in Chinese
-
Article discusses reform: Does the article’s use of the “reform” term (改革, 创新, 试点, 示范) actually correspond to any government reform (however defined and regardless of where the reform takes place)?
-
City is main subject: Is the specified city the main subject of the article (regardless of whether the article is about a reform)?
-
City mentioned only once: Is the specified city only mentioned once in the article (regardless of whether it is mentioned in relationship to the reform)?
-
Reform in specified city: Does the article mention a reform in the specified city?
-
Reform in city is main subject: Is the reform in the specified city the main subject of the article?
-
Reform in city mentioned once: Is the reform in the specified city only mentioned one time?
-
Reform in city mentioned multiple times: Is the reform in the city mentioned multiple times, but still not the major focus of the article?
-
Reform subject: What is the subject of the reform (e.g., land, education, cadre management, etc.)?
-
Reform name: If the reform/pilot/experiment is given a name, write in the original Chinese.
-
Positive or negative story: Is the way the reform in the specified city mentioned in the article generally “positive” or “negative”?
This validation exercise confirms that our measure captures some degree of policy innovation and reform in the specified cities despite a certain amount of background noise. In 13 articles out of 157 in the sample (8%), the city in question is incorrectly specified. In all other cases, the city in question is mentioned, though not always as the location of a policy innovation or reform. Out of 157 articles, 83 (52%) are explicitly related to reforms or innovations, while 51 of the articles (32%) are clearly about reforms in the specified city. Other articles fall into a gray area: the specified city is mentioned and policy reforms are discussed, but it is not clear that these reforms and innovations are taking place in the city of interest. When reforms are discussed, in 96% of cases they are discussed with a positive valence, confirming our interpretation of the reason why national media would focus on these local innovations.
In sum, somewhere between 32 and 92% of articles in the close-reading sample fit our “policy autonomy” interpretation. While the policy autonomy proxy seems broadly valid, it also has a considerable degree of measurement error, with up to half of observations representing noise and not signal. Assuming this measurement error is uncorrelated with our independent variables, it should lead to higher standard errors in our regression coefficient estimates. Because average estimates should not differ, however, our results will not be biased.
3. Which cities get selected to have concurrent appointments?
Table
4 below lists all non-capital and non-deputy-provincial cities that have concurrent appointments over 1996–2013, including the number of years in which they hold these appointments.
Table 4
List of cities with concurrent appointments
Baise | Guangxi | 1 | Ningde | Fujian | 1 |
Baoding | Hebei | 2 | Qiandongnan | Guizhou | 1 |
Baotou | Neimenggu | 8 | Qianxinan | Guizhou | 3 |
Beihai | Guangxi | 8 | Qingyang | Gansu | 1 |
Bijie | Guizhou | 1 | Qinhuangdao | Hebei | 1 |
Changdu | Xizang | 2 | Qinzhou | Guangxi | 1 |
Chifeng | Neimenggu | 1 | Sanya | Hainan | 16 |
Chongqing | Sichuan | 1 | Suqian | Jiangsu | 1 |
Chuxiong | Yunnan | 1 | Suzhou | Jiangsu | 13 |
Daqing | Heilongjiang | 4 | Tangshan | Hebei | 6 |
Enshi | Hubei | 1 | Tongliao | Neimenggu | 2 |
Ezhou | Hubei | 2 | Wenzhou | Zhejiang | 5 |
Foshan | Guangdong | 5 | Wuhu | Anhui | 3 |
Ganzhou | Jiangxi | 10 | Wuxi | Jiangsu | 7 |
Guilin | Guangxi | 1 | Yanan | Shaanxi | 12 |
Guyuan | Ningxia | 2 | Yanbian | Jilin | 18 |
Hainan | Qinghai | 1 | Yantai | Shandong | 3 |
Haixi | Qinghai | 1 | Yichang | Hubei | 2 |
Jilin | Jilin | 1 | Yulin | Guangxi | 3 |
Liuzhou | Guangxi | 2 | Yushu | Qinghai | 2 |
Luoyang | Henan | 10 | Zhuhai | Guangdong | 3 |
Nandu | Xizang | 2 | Zunyi | Guizhou | 22 |
To more systematically test whether considerations of economic and political importance drive concurrent appointments for the larger set of represented cites, we run logit regressions of PPSC concurrent appointments on a set of potential explanatory factors:
$$ C{A}_{it}={\alpha}_{it}+\beta {\mathbf{X}}_{\mathbf{it}}+\gamma {\mathbf{C}}_{\mathbf{it}}+{\delta}_t+{\varphi}_i+{\varepsilon}_{it} $$
Here, our unit of observation is the city-year; i references cities and t references years. CA is a dummy variable for city-years with concurrent appointments. X is a vector of potential city-level predictors of concurrent appointments: city-level gross domestic product (GDP), population, foreign direct investment (FDI), a dummy variable for “red” cities with revolutionary histories, a dummy variable for border cities, and a dummy variable for autonomous cities. In addition, the regressions include a set of controls (C) for whether a city is a capital city or a deputy-provincial city, and include both year fixed effects (δ) and province/region fixed effects (φ). This logit regression helps to determine whether economically important cities (GDP), large cities (population), economically open cities (FDI), historically important cities (revolutionary history), and/or geopolitically important cities (border cities) are more likely to enjoy representation. Border cities are coded “1” when they share a land border with another country (13% of prefectures). Revolutionary cities are coded “1” based on their “red tourism” status (4% of prefectures).
Regression results appear in Table
5 below. The results confirm our expectations that beyond capitals and deputy-provincial cities, two types of city are most likely to gain concurrent appointments: economically important cities and historically/politically important cities. GDP and FDI are associated with a greater likelihood of representation, as might be expected, as is revolutionary status (indicated by cities’ designation as “red tourism” sites). Border cities are if anything
less likely to have representation (though this is a weak and insignificant effect). Column (1) includes macro-regional dummies instead of provincial fixed effects; interestingly, there are no apparent macro-regional differences. These results dovetail with an “eyeball” test from looking at the list of represented cities in Table
4, and underscore the fact that there are multiple logics of PPSC representation for cities.
Table 5
Logit results for city PPSC representation
GDP (billion RMB) | 0.00870* (1.69) | 0.0125** (2.48) | | |
Population (million) | − 0.0235 (− 0.22) | − 0.0654 (− 0.55) | 0.0301 (0.28) | 0.0147 (0.15) |
Revenue (billion RMB) | − 0.0380 (− 0.62) | − 0.0797 (− 1.23) | 0.0549 (1.17) | 0.0535 (1.08) |
FDI (million USD) | 0.000802 (1.44) | 0.00132** (2.07) | 0.000919* (1.68) | 0.00160*** (2.67) |
“Red” city | 2.062** (2.49) | 2.045** (2.51) | 1.963** (2.38) | 1.849** (2.37) |
Border city | − 0.315 (− 0.28) | − 0.861 (− 0.83) | − 0.238 (− 0.21) | − 0.710 (− 0.65) |
Capital city | 6.397*** (12.75) | 7.962*** (9.36) | 6.329*** (12.79) | 7.810*** (9.52) |
Deputy-provincial city | 6.109*** (5.99) | 8.138*** (8.52) | 5.756*** (5.40) | 7.503*** (8.24) |
Autonomous city | 0.894 (1.03) | − 0.558 (− 0.62) | 0.878 (1.02) | − 0.945 (− 1.11) |
Northeast region | − 0.0588 (− 0.10) | | 0.0457 (0.07) | |
Coastal region | − 0.463 (− 0.59) | | − 0.261 (− 0.36) | |
Western region | − 0.0413 (− 0.07) | | − 0.0370 (− 0.07) | |
Constant | − 3.604*** (− 2.82) | − 5.766*** (− 3.12) | − 3.423*** (− 3.01) | − 5.420*** (− 2.93) |
Observations | 3242 | 3242 | 3242 | 3242 |
Year fixed effects | Yes | Yes | Yes | Yes |
Province fixed effects | No | Yes | No | Yes |
4. Robustness checks for baseline results (Table
2)
Here, we provide additional robustness checks for the baseline results in Table
2. In Table
6, we include provincial fixed effects rather than prefecture fixed effects, and cluster standard errors by province. This allows us to more explicitly identify whether outcomes for cities with concurrent appointments differ from other cities within the same province, rather than in comparison to all cities across China. Greater news frequency or GDP growth that occurred across all cities in a province relative to the national average would yield significant effects when including prefecture fixed effects, but this effect would be blunted by including province fixed effects. Given that we are trying to identify policy autonomy from the province itself and distributive outcomes as they relate to provincial authorities, this is an important robustness check. The results confirm the baseline Table
2 findings.
Table 6
Provincial fixed effects with standard errors clustered by prefecture
Concurrent PPSC appointment | 14.98** (2.31) | 0.665* (1.75) | 1.181 (1.37) | 0.326 (0.15) | 1.134** (2.31) | 1.957*** (2.85) |
Log GDP per capita (1000 RMB) | 16.21*** (3.05) | − 6.429*** (− 9.15) | 0.745 (1.31) | 1.956 (1.49) | − 1.366** (− 2.42) | − 7.849*** (− 7.28) |
Population (millions) | 2.520*** (2.99) | − 0.479*** (− 6.37) | − 0.161** (− 2.58) | − 0.0833 (− 0.53) | − 0.243*** (− 5.46) | − 0.711*** (− 7.84) |
Land area | 1.282** (1.97) | 0.210** (2.13) | 0.0662 (1.09) | 0.301 (1.42) | − 0.0202 (− 0.40) | 0.186 (1.51) |
Primary share of GDP (%) | 0.0311 (0.17) | − 0.0166 (− 0.55) | − 0.0581 (− 1.51) | 0.0991 (1.26) | − 0.0771** (− 2.23) | − 0.0955* (− 1.95) |
Tertiary share of GDP (%) | 0.612** (2.01) | 0.0602*** (2.86) | − 0.129*** (− 3.86) | − 0.100 (− 1.57) | 0.0474** (2.06) | 0.106*** (3.40) |
FDI (million USD) | 0.0232*** (3.50) | 0.000102 (0.34) | − 0.000400* (− 1.70) | − 0.00175** (− 2.57) | 0.000818*** (3.45) | 0.000874* (1.88) |
Autonomous city | 27.91** (2.03) | − 2.714* (− 1.74) | − 3.752 (− 1.63) | − 0.617 (− 0.14) | 0.302 (0.14) | 2.708 (0.96) |
Border city | − 5.357 (− 0.87) | 1.356* (1.89) | − 0.0193 (− 0.03) | − 0.260 (− 0.18) | 0.392 (1.03) | 1.777** (2.24) |
“Red” city | 7.837 (1.18) | 0.112 (0.22) | 0.790 (1.02) | − 0.527 (− 0.39) | 1.002 (1.59) | 1.083 (1.13) |
Capital city | 49.26*** (4.25) | − 0.462 (− 0.72) | 1.186 (1.17) | 2.777 (1.16) | − 0.267 (− 0.51) | − 0.787 (− 0.92) |
Deputy provincial city | 36.26 (1.52) | 4.438*** (2.98) | 1.463 (1.45) | − 0.714 (− 0.31) | 1.393 (1.28) | 5.846** (2.48) |
Constant | − 220.8*** (− 3.69) | 76.43*** (10.35) | 13.90** (2.12) | 13.272 (0.94) | 4.76*** (3.52) | 97.98*** (8.03) |
Observations | 3728 | 3161 | 3728 | 3728 | 3233 | 3203 |
In addition, Table
7 repeats the baseline Table
2 regressions but uses a more limited sample of observations. In order to run regressions on a consistent sample for each dependent variable, we drop all city-year observations that have missing observations for any of the relevant dependent variables or independent variables. This results in 2788 observations (nearly 20% fewer than several regressions have in the baseline results). Using this limited sample does not significantly change the results.
Table 7
Baseline results using maximum consistent sample for all regressions
Concurrent PPSC appointment | 2.039 (1.39) | 2.596*** (2.65) | 1.841 (0.59) | 0.734*** (2.76) | 1.147* (1.86) | 0.413 (0.75) |
Log GDP per capita (1000 RMB) | 3.567*** (2.99) | 2.274*** (2.85) | 8.946*** (3.50) | − 5.955*** (− 27.55) | − 13.97*** (− 27.84) | − 8.012*** (− 17.80) |
Population (millions) | − 0.747 (− 0.77) | − 0.182 (− 0.28) | 1.500 (0.72) | − 1.145*** (− 6.53) | − 2.470*** (− 6.07) | − 1.325*** (− 3.63) |
Primary share of GDP (%) | − 0.035 (− 0.74) | – 0.0356 (− 1.13) | 0.460*** (4.55) | − 0.0605*** (− 7.08) | − 0.150*** (− 7.58) | − 0.0898*** (− 5.05) |
Tertiary share of GDP (%) | 0.148*** (3.39) | − 0.0942*** (− 3.23) | − 0.0786 (− 0.84) | 0.0798*** (10.11) | 0.109*** (5.96) | 0.0293* (1.79) |
FDI (million USD) | 0.00708*** (7.19) | − 0.00205*** (− 3.11) | − 0.00761*** (− 3.61) | 0.00189*** (10.62) | 0.000223 (0.54) | − 0.00167*** (− 4.50) |
Constant | − 23.44* (− 1.65) | − 11.44 (− 1.20) | − 80.54*** (− 2.64) | 72.24*** (28.04) | 172.9*** (28.92) | 100.7*** (18.76) |
Observations | 2788 | 2788 | 2788 | 2788 | 2788 | 2788 |
5. Regression Results by City Type
In the paper itself, Fig.
3 presents a plot of the estimated coefficients for the concurrent appointment independent variable from 16 different regressions. To create coefficient plots that are easily comparable, we run regressions using normalized z-scores for all variables. Here, we present the baseline regressions results using raw data (i.e., the identical approach to as used to generate Table
2), for the three city subsets: rich cities (Table
8), red cities (Table
9), and non-rich, non-red cities (Table
10). As in the text, we define rich cities as those with per capita income above the 75th percentile in any given year, and we define red cities based on red tourism status, as described in section 1 of this appendix.
Table 8
Baseline regression (Table
4) repeated on “rich” city sample
Concurrent PPSC appointment | 3.223 (1.10) | 5.369*** (4.80) | 14.85*** (3.17) | 0.441* (1.71) | 0.740*** (4.35) | 0.398 (1.60) |
Log GDP per capita (1000 RMB) | 14.90** (2.51) | 2.252 (0.99) | − 6.665 (− 0.70) | − 1.679*** (− 2.97) | − 3.946*** (− 10.68) | − 2.142*** (− 3.93) |
Population (millions) | − 0.596 (− 0.23) | 0.429 (0.43) | − 1.537 (− 0.36) | − 0.612** (− 2.56) | 0.109 (0.69) | 0.698*** (3.02) |
Primary share of GDP (%) | 0.658* (1.73) | − 0.608*** (− 4.19) | − 0.628 (− 1.03) | − 0.0679** (− 1.97) | − 0.157*** (− 6.90) | − 0.0978*** (− 2.94) |
Tertiary share of GDP (%) | 0.0743 (0.29) | − 0.559*** (− 5.82) | − 0.650 (− 1.61) | 0.0307 (1.23) | − 0.0144 (− 0.87) | − 0.0414* (− 1.72) |
FDI (million USD) | 0.00268 (1.42) | − 0.000265 (− 0.37) | − 0.00587* (− 1.94) | 0.00107*** (6.40) | 0.000311*** (2.82) | − 0.000774*** (− 4.81) |
Constant | − 139.0* (− 1.89) | 1.088 (0.04) | 132.8 (1.13) | 27.75*** (3.97) | 55.96*** (12.22) | 26.98*** (4.00) |
Observations | 664 | 664 | 664 | 538 | 561 | 538 |
Table 9
Baseline regression (Table
4) repeated on “red” city sample
Concurrent PPSC appointment | 28.47*** (4.22) | − 1.480 (− 0.58) | − 21.47 (− 1.60) | 1.038 (0.76) | 2.372 (1.38) | 1.334 (1.23) |
Log GDP per capita (1000 RMB) | 2.610 (0.49) | − 2.206 (− 1.09) | 3.803 (0.36) | − 9.911*** (− 10.72) | − 18.14*** (− 15.74) | − 8.232*** (− 11.28) |
Population (millions) | − 16.95 (− 1.51) | 0.787 (0.18) | − 5.183 (− 0.23) | − 4.879** (− 2.48) | − 10.45*** (− 4.26) | − 5.575*** (− 3.59) |
Primary share of GDP (%) | 0.608 (1.47) | − 0.236 (− 1.50) | 0.398 (0.48) | − 0.628*** (− 8.07) | − 0.826*** (− 8.50) | − 0.198*** (− 3.21) |
Tertiary share of GDP (%) | 0.897** (2.33) | − 0.383** (− 2.62) | − 0.607 (− 0.79) | − 0.047 (− 0.72) | 70.087 (1.05) | 30.135** (2.57) |
FDI (million USD) | − 0.0302*** (− 2.63) | − 0.00403 (− 0.92) | 0.00940 (0.41) | − 0.00283 (− 1.37) | − 0.00216 (− 0.83) | 0.000672 (0.41) |
Constant | 93.88 (1.25) | 55.76* (1.96) | 13.38 (0.09) | 142.9*** (10.50) | 265.2*** (15.62) | 122.3*** (11.39) |
Observations | 140 | 140 | 140 | 118 | 118 | 118 |
Table 10
Baseline regression (Table
4) repeated on non-rich, non-red city sample
Concurrent PPSC appointment | − 2.060 (− 1.49) | − 0.488 (− 0.35) | − 3.736 (− 0.81) | 0.294 (0.74) | − 0.0121 (− 0.01) | − 0.235 (− 0.25) |
Log GDP per capita (1000 RMB) | 3.022*** (3.40) | 3.010*** (3.39) | 10.63*** (3.59) | − 6.561*** (− 25.82) | − 15.76*** (− 24.13) | − 9.295*** (− 15.31) |
Population (millions) | − 1.538* (− 1.95) | 0.956 (1.21) | 3.728 (1.42) | − 1.511*** (− 7.08) | − 2.950*** (− 5.32) | − 1.450*** (− 2.85) |
Primary share of GDP (%) | − 0.0802*** (− 2.64) | − 0.0131 (− 0.43) | 0.394*** (3.89) | − 0.0464*** (− 5.32) | − 0.131*** (− 5.79) | − 0.0859*** (− 4.12) |
Tertiary share of GDP (%) | 0.103*** (3.52) | − 0.0658** (− 2.25) | 0.012 (0.12) | 10.0708*** (8.67) | 0.0979*** (4.63) | 0.0279 (1.43) |
FDI (million USD) | 0.00159 (1.10) | − 0.0000536 (− 0.04) | − 0.00394 (− 0.82) | 0.00285*** (7.25) | 0.00151 (1.48) | − 0.00122 (− 1.30) |
Constant | − 13.59 (− 1.29) | − 23.58** (− 2.24) | − 106.5*** (− 3.05) | 78.82*** (26.52) | 191.6*** (25.08) | 113.8*** (16.03) |
Observations | 2496 | 2496 | 2496 | 2145 | 2164 | 2145 |
6. Causal mechanisms: institutions versus individuals
Our analysis in the paper works from the idea that the institution of concurrent appointment itself causes the effects we see—that the difference in city leaders’ political status and in their relationship to superiors under concurrent appointment is responsible for outcomes that would not otherwise occur. However, this is not the only possible causal logic at work. One alternative possibility is that already high-performing or well-connected leaders are rewarded with concurrent appointments, such that we are conflating the effect of an institution with the achievement of an individual. Another, potentially related, possibility is that concurrent appointment (and the higher rank it confers on an individual) incentivizes a given city leader to behave more ambitiously by opening up previously unavailable career possibilities. In either case, observed outcomes might be driven by the characteristics of specific leaders rather than the institutional empowerment of specific cities. Although our data do not allow us to fully resolve this question of causal logic, additional evidence from the case-study cities and further quantitative analysis increase our confidence in the institutional interpretation.
One piece of evidence that would support an individual-driven as opposed to institution-driven interpretation would be if concurrently appointed leaders tended to be young “rising stars”—individuals who have been specially groomed for rapid career promotion due to either merit or patronage. In terms of ability, such individuals might be more capable or wield more political capital to begin with. In terms of incentives, rising stars might feel greater pressure to make a name for themselves.
9 What we observe empirically, however, is that leaders of various ages, not simply rising stars, get concurrent appointments. We also find that the quantitative relationships between current appointments and enhanced local economic outcomes and policy activism hold across different age categories and are, if anything, driven by older cadres. The city leaders mentioned in our case-study cities were appointed at varying ages, from the relatively young 42 in the case of Pan Yiyang in Ganzhou and 45 for Fan Ruiping in Xiangyang to the more advanced 56 for Shi Wenqing in Ganzhou and 58 for Liang Guangda in Zhuhai (China Vitae; Baidu Baike). Looking more systematically at the age profile of concurrent appointees in our dataset, we find that such leaders average 51.9 years old, younger than the average for PPSC members (54.5 years for regular members), but equal to the average age of all prefecture leaders over 2000–2015.
10 And beyond the case-study cities as well, the age distribution of concurrent appointees is very wide: while some are rising stars in their early 40s, many are in their late 50s and likely serving in their final post. To test whether rising star cadres drive the aggregate results, we re-run our baseline regressions using different age-based samples: rising stars (aged 49 and younger, accounting for 26% of cases), young (52 and younger, 51% of cases), old (53 and older, 49% of cases), and last post (56 and older, 25% of cases). As shown in Table
11, there is no evidence that rising stars drive our aggregate results; if anything, the results are driven by older cadres, though the signs and general results are fairly consistent throughout age groups.
Table 11
Baseline regression (Table
4) repeated on age group sub-samples
Concurrent PPSC appointment | Rising star (49 and younger) | 0.904 | | | | − 2.194 | | | | − 3.303 | | | | 0.273 | | | | 0.215 | | | | − 0.0209 | | | |
(0.51) | | | | (− 1.38) | | | | (− 0.60) | | | | (0.73) | | | | (0.25) | | | | (− 0.03) | | | |
Young (52 and younger) | | − 1.489 | | | | 0.957 | | | | 3.841 | | | | 0.639** | | | | 0.592 | | | | − 0.00982 | | |
| (− 1.08) | | | | (0.77) | | | | (0.90) | | | | (2.21) | | | | (0.90) | | | | (− 0.02) | | |
Old (53 and older) | | | 6.689*** | | | | 3.434** | | | | 6.641 | | | | 0.419 | | | | 1.290 | | | | 0.897 | |
| | (3.62) | | | | (2.08) | | | | (1.17) | | | | (1.14) | | | | (1.54) | | | | (1.18) | |
Last post (56 and older) | | | | 0.870 | | | | 0.879 | | | | 26.35*** | | | | 1.053* | | | | 1.955 | | | | 0.923 |
| | | (0.27) | | | | (0.31) | | | | (2.68) | | | | (1.70) | | | | (1.38) | | | | (0.72) |
Log GDP per capita (1000 RMB) | 3.852*** | 3.840*** | 3.635*** | 3.834*** | 7.160*** | 7.179*** | 7.069*** | 7.165*** | 11.66*** | 11.69*** | 11.47*** | 11.35*** | − 6.385*** | − 6.382*** | − 6.398*** | − 6.400*** | − 14.25*** | − 14.25*** | − 14.29*** | − 14.28*** | − 7.901*** | − 7.901*** | − 7.928*** | − 7.914*** |
(3.95) | (3.94) | (3.73) | (3.93) | (8.23) | (8.25) | (8.12) | (8.23) | (3.90) | (3.91) | (3.83) | (3.80) | (− 31.14) | (− 31.16) | (− 31.17) | (− 31.21) | (− 30.76) | (− 30.75) | (− 30.81) | (− 30.80) | (− 18.61) | (− 18.61) | (− 18.65) | (− 18.62) |
Population (millions) | − 0.987 | − 0.961 | − 0.890 | − 0.975 | 1.733** | 1.696** | 1.750** | 1.706** | 1.296 | 1.217 | 1.340 | 1.259 | − 1.377*** | − 1.379*** | − 1.368*** | − 1.373*** | − 2.669*** | − 2.671*** | − 2.649*** | − 2.666*** | − 1.270*** | − 1.270*** | − 1.258*** | − 1.270*** |
(− 1.13) | (− 1.10) | (− 1.02) | (− 1.12) | (2.23) | (2.18) | (2.25) | (2.19) | (0.48) | (0.46) | (0.50) | (0.47) | (− 7.88) | (− 7.90) | (− 7.83) | (− 7.87) | (− 6.73) | (− 6.74) | (− 6.68) | (− 6.73) | (− 3.51) | (− 3.51) | (− 3.48) | (− 3.51) |
Primary share of GDP (%) | − 0.0672* | − 0.0647* | − 0.0648* | − 0.0662* | 0.120*** | 0.117*** | 0.119*** | 0.118*** | 0.721*** | 0.714*** | 0.720*** | 0.722*** | − 0.0778*** | − 0.0782*** | − 0.0775*** | − 0.0774*** | − 0.166*** | − 0.166*** | − 0.165*** | − 0.165*** | − 0.0896*** | − 0.0896*** | − 0.0894*** | − 0.0894*** |
(− 1.85) | (− 1.78) | (− 1.78) | (− 1.82) | (3.70) | (3.60) | (3.67) | (3.64) | (6.46) | (6.40) | (6.46) | (6.48) | (− 10.25) | (− 10.31) | (− 10.22) | (− 10.21) | (− 9.60) | (− 9.62) | (− 9.58) | (− 9.57) | (− 5.70) | (− 5.70) | (− 5.69) | (− 5.69) |
Tertiary share of GDP (%) | 0.153*** | 0.150*** | 0.150*** | 0.152*** | − 0.119*** | − 0.116*** | − 0.118*** | − 0.117*** | − 0.106 | − 0.098 | – 0.105 | − 0.105 | 0.0942*** | 0.0945*** | 0.0938*** | 0.0939*** | 0.126*** | 0.126*** | 0.125*** | 0.126*** | 0.0329** | 0.0329** | 0.0325** | 0.0328** |
(4.30) | (4.23) | (4.23) | (4.28) | (− 3.76) | (− 3.65) | (− 3.73) | (− 3.69) | (− 0.97) | (− 0.90) | (− 0.96) | (− 0.96) | (12.87) | (12.94) | (12.83) | (12.85) | (7.56) | (7.59) | (7.53) | (7.55) | (2.17) | (2.17) | (2.15) | (2.17) |
FDI (million USD) | 0.00766*** | 0.00781*** | 0.00775*** | 0.00773*** | − 0.00254*** | − 0.00274*** | − 0.00265*** | − 0.00266*** | − 0.0122*** | − 0.0127*** | − 0.0124*** | − 0.0120*** | 0.00201*** | 0.00198*** | 0.00203*** | 0.00204*** | 0.000235 | 0.000207 | 0.000263 | 0.000284 | − 0.00173*** | − 0.00173*** | − 0.00172*** | − 0.00172*** |
(8.93) | (9.13) | (9.13) | (9.06) | (− 3.33) | (− 3.59) | (− 3.50) | (− 3.50) | (− 4.65) | (− 4.83) | (− 4.74) | (− 4.60) | (11.48) | (11.36) | (11.70) | (11.76) | (0.59) | (0.52) | (0.67) | (0.72) | (− 4.78) | (− 4.79) | (− 4.79) | (− 4.77) |
Constant | − 25.18** | − 25.09** | − 23.70** | − 25.12** | − 71.64*** | − 71.73*** | − 70.93*** | − 71.64*** | − 108.3*** | − 108.5*** | − 106.9*** | − 106.9*** | 77.58*** | 77.52*** | 77.65*** | 77.63*** | 176.7*** | 176.6*** | 176.9*** | 176.8*** | 99.29*** | 99.29*** | 99.46*** | 99.34*** |
(− 2.14) | (− 2.13) | (− 2.02) | (− 2.13) | (− 6.83) | (− 6.84) | (− 6.76) | (− 6.82) | (− 3.00) | (− 3.01) | (− 2.96) | (− 2.97) | (31.75) | (31.75) | (31.77) | (31.79) | (31.91) | (31.91) | (31.96) | (31.94) | (19.62) | (19.62) | (19.65) | (19.63) |
Observations | 3410 | 3410 | 3410 | 3410 | 3410 | 3410 | 3410 | 3410 | 3410 | 3410 | 3410 | 3410 | 2909 | 2909 | 2909 | 2909 | 2952 | 2952 | 2952 | 2952 | 2909 | 2909 | 2909 | 2909 |
Although age does not appear to be a decisive factor, other underlying individual characteristics may explain both concurrent appointments and the city outcomes we observe. Perhaps high-performing leaders are selected for concurrent appointments and then continue producing the results they achieved in previous posts. Or perhaps cadres with ties to powerful party factions or well-placed patrons obtain concurrent appointments, and the outcomes we observe are driven by these political connections. The fact that we observe concurrent appointees with a range of career backgrounds and promotion trajectories seems inconsistent with the idea that either previous high achievement or factional alignment is the main factor driving both PPSC appointments and observed city outcomes.
11 But we can begin to assess these possibilities more systematically by analyzing the effect of concurrent appointments for officials who already hold their city leadership posts when they are concurrently appointed to PPSCs. To do so, we track the background of the 69 leaders of non-capital, non-deputy-provincial cities and find that 20 of them (almost 30%) already led their respective cities prior to PPSC elevation, serving nearly 2 years on average. By analyzing the outcomes of these individuals before their PPSC appointment, we can control for individual characteristics—either factional ties or innate ability—while analyzing the effects of institutional arrangements.
12 To check the city-level impacts of these leaders prior to PPSC elevation, we repeat our baseline regression and also include a “pre-PPSC” dummy to check whether the main results hold when controlling for previous city appointment, and also whether the same results hold for these individuals prior to elevation. Table
12 demonstrates that the main results continue to hold for the concurrent appointment dummy, but no results are significant for the pre-PPSC dummy (and some have opposite signs), implying that these individuals on the whole are
not exceptional prior to their PPSC elevation. This provides further support for our institutional interpretation, though the small sample limits the power of this analysis.
13Table 12
Baseline results controlling for pre-PPSC city appointments
Pre-PPSC city appointment | − 0.844 (− 0.48) | − 0.611 (− 0.39) | 7.642 (1.41) | 0.532 (1.34) | 0.769 (0.85) | 0.323 (0.39) |
Concurrent PPSC appointment | 1.515 (1.21) | 2.336** (2.09) | 9.949*** (2.59) | 0.798*** (3.04) | 1.220** (2.05) | 0.501 (0.92) |
Log GDP per capita (1000 RMB) | 3.796*** (3.89) | 7.105*** (8.17) | 11.43*** (3.82) | − 6.405*** (− 31.28) | − 14.28*** (− 30.82) | − 7.913*** (− 18.63) |
Population (millions) | − 0.967 (− 1.11) | 1.721** (2.21) | 1.351 (0.51) | − 1.363*** (− 7.82) | − 2.651*** (− 6.69) | − 1.264*** (− 3.49) |
Primary share of GDP (%) | − 0.0678* (− 1.86) | 0.116*** (3.57) | 0.710*** (6.37) | − 0.0783*** (− 10.33) | − 0.167*** (− 9.65) | − 0.0900*** (− 5.73) |
Tertiary share of GDP (%) | 0.153*** (4.32) | − 0.115*** (− 3.63) | − 0.0958 (− 0.88) | 0.0944*** (12.93) | 0.126*** (7.60) | 0.0332** (2.19) |
FDI (million USD) | 0.00760*** (8.90) | − 0.00283*** (− 3.71) | − 0.0129*** (− 4.92) | 0.00199*** (11.44) | 0.000189 (0.48) | − 0.00175***(− 4.87) |
Constant | − 24.86** (− 2.11) | − 71.26 (− 6.79) | − 107.2*** (− 2.97) | 77.65*** (31.83) | 176.8*** (31.95) | 99.34*** (19.63) |
Observations | 3410 | 3410 | 3410 | 2909 | 2952 | 2909 |
In sum, although we cannot entirely rule out the possibility that unobserved individual characteristics drive our results, the evidence we have presented suggests that the institutional arrangements that come with concurrent appointments are themselves important. Future research, drawing on larger datasets, may be able to further tease apart the effects of institutions and individuals. But it is worth noting that even if the observed effects of concurrent appointments are a function of installing leaders who are more talented, powerful, or incentivized powerful than average leaders, this would remain consistent with the paper’s larger argument that concurrent appointments empower rather than constrain cities.