1 Motivation

What is the critical factor of the cadre’s evaluation in the authoritarian regime? This is a highly debated topic in political science. Former researchers focused on two mainstream theories - “By Performance” and “By Loyalty.” The “Performance Theory” or the more well-known name “Yardstick Competition”(Li and Zhou, 2005) believe the cadres will compete with the economic performance, which is one of the major agendas of Chinese government want to achieve. Those who have better capability to carry out higher growth, whether absolute or relative performance, will receive higher probability of promotion. On the other hand, the “Loyalty Theory” suggested the political connection, as a proxy of factions or loyalty, is the decisive reason why cadres got a promotion. Recent studies are giving us more insightful observations. Jia et al.(2015) claimed the two standards are both critical in the evaluation, while Landury et al.(2017) found no effect of neither factors in the provincial party secretaries promotion.

The contradictory empirical results drive us to investigate what is a more plausible answer to this very fundamental question of China politics. In this project, we replicated two papers that using both connection, performance, and their interaction term, which are Jia et al. (2015) and Landury et al. (2017).(Lia and Zhou 2005 did not investigate the interaction effect)

In the following parts of this document, we will first analyze the different identify strategies and conclusion of these three papers, and then using original data to replicate their results, so that we could be sure we are following the exact procedure. Finally, we will use the data collected and cleaned by ourselves, to see if their conclusions remained stable.

Paper Performance Connection Complementary
Li and Zhou (2005) √ X ø
Jia et al. (2015) X X √
Landry et al. (2017) X X X

The above table show us the basic conclusion of the three papers. We also show the detailed differences of variable definition across three papers.

Variable Li.and.Zhou.2005 Jia.et.al.2015 Landry.et.al.2017
Period 1979 - 1995 1993 - 2009 1999 - 2007
Performance Weighted Average GDP growth Average real GDP growth Relative GDP Growth
Connection Past Experince in Center Former Colleague of PSC Members Former Colleague of GPS
Promotion Poliburo, Vice-Premier, State Councilor Poliburo, Vice-Premier, State Councilor Poliburo, Vice-Premier, State Councilor
% of Cadres 254 275 66
Data Form Cadre-Year Panel Cadre-Year Panel Cadre-Year Panel(Collapsed upon term)
Model Ordered Logistic Logit and Probit Linear Regression

2 Replicate the Results Using Original Data

First of all, we want to check if we could replicate the results of original papers. This is a very important step, as we want to make sure our code is following the specific procedure of these two papers.

2.1 Jia Ruixue, Kudamatsu Masayuki & Seim, David (2015). Political Selection in China: the Complementary Roles of Connections and Performance. Journal of the European Economic Association.

Complementarity Between Connections and Growth (Provincial Party Secretary and Governor)
Dependent variable:
Promotion
(1) (2) (3) (4) (5)
Growth 0.049** 0.048** 0.032 0.018
(0.023) (0.023) (0.023) (0.027)
Connection 0.702 0.690 0.356 3.944
(0.465) (0.447) (0.427) (6.498)
Connection x Growth 2.195** 2.075**
(0.889) (0.839)
Fixed Effect Y Y Y Y Y
Control covariates N N N N Y
Note: p<0.1; p<0.05; p<0.01
Cluster Standard Error on Province Level
Complementarity Between Connections and Growth (Provincial Party Secretary Only)
Dependent variable:
Promotion
(1) (2) (3) (4) (5)
Growth 0.056** 0.057** 0.045** 0.025
(0.024) (0.024) (0.022) (0.019)
Connection -0.035 -0.075 -0.475 8.266
(0.634) (0.597) (0.588) (8.355)
Connection x Growth 2.865*** 2.716***
(0.989) (0.937)
Fixed Effect Y Y Y Y Y
Control covariates N N N N Y
Note: p<0.1; p<0.05; p<0.01
Cluster Standard Error on Province Level

2.2 Landry Pierre F., Lü Xiaobo, & Duan, Haiyan (2017). Does Performance Matter? Evaluating Political Selection Along the Chinese Administrative Ladder. Comparative Political Studies.

Promotion of Party Secretary upon Term Completion (Relative Fiscal Revenue Growth to Competitors)
Dependent variable:
Promotion
(1) (2) (3) (4)
Relative Fiscal Revenue Growth 0.057 -0.020 0.032 -0.046
(0.051) (0.044) (0.048) (0.038)
Connection 0.057 0.031 0.059 0.031
(0.077) (0.070) (0.076) (0.069)
Connection x Growth 0.089 0.110
(0.126) (0.130)
Year Fixed Effect Y Y Y Y
Local Characteristics Y Y Y Y
Politician Characteristics N Y N Y
Note: p<0.1; p<0.05; p<0.01
Cluster Standard Error on Province Level
Promotion of Party Secretary upon Term Completion (Relative GDP Growth to Competitors)
Dependent variable:
Promotion
(1) (2) (3) (4)
Relative GDP Growth 0.169** 0.136* 0.168 0.087
(0.082) (0.080) (0.137) (0.124)
Connection -0.001 -0.016 -0.001 -0.007
(0.082) (0.059) (0.086) (0.067)
Connection x Growth 0.002 0.079
(0.158) (0.160)
Year Fixed Effect Y Y Y Y
Local Characteristics Y Y Y Y
Politician Characteristics N Y N Y
Note: p<0.1; p<0.05; p<0.01
Cluster Standard Error on Province Level

We replicated these papers and results is exactly same. Therefore, we are confident to do the next step.

3 Replicate the Result Using Reconstructed Data

There could be so many reasons why these two papers resulted in contradictory conclusions. As described in the fact table of section 1, the variable choosen, period coverage, model setting, and even the data generation process are all possible candidates. We beliveve this is also the reason why the debating about the mechanism of cadres’ evaluation is still in full swing.

Therefore, the proper strategy is to replicate these differences step by step, to check the stability of the conclusion. The data itself is the start point of empirical analysis. Are they diverged at the starting line？To answer this question, we should reproduce the dataset following the exact codebook of each reasearch. Based on the same data source, we reconstructed the cadres’ resume data and use it to replace the original data of each paper, then replicate them with other setting remaining the same.

In the following process, we will first compare the novel data to the original one and then use the same code to reproduce the results.

3.1 Use our data to run Jia’s model

grp chng_type Province Year Name Promotion Connection
10 + anhui 2002 wangtaihua 0 1
10 - anhui 2002 wangtaihua 0 0
16 + anhui 2008 wangjinshan 0
16 - anhui 2008 wangjinshan 0 0
17 + anhui 2009 wangjinshan 0
17 - anhui 2009 wangjinshan 0 0
18 + beijing 1993 chenxitong 0 1
18 - beijing 1993 chenxitong 0 0
19 + beijing 1994 chenxitong 0 1
19 - beijing 1994 chenxitong 0 0
20 + beijing 1995 weijianxing 0 1
20 - beijing 1995 weijianxing 0 0
21 + beijing 1996 weijianxing 0 1
21 - beijing 1996 weijianxing 0 0
40 + chongqing 2002 heguoqiang 1 1
40 - chongqing 2002 heguoqiang 1 0
45 + chongqing 2007 wangyang 1 1
45 - chongqing 2007 wangyang 1 0
46 + chongqing 2008 boxilai 0 1
46 - chongqing 2008 boxilai 0 0
47 + chongqing 2009 boxilai 0 1
47 - chongqing 2009 boxilai 0 0
62 + fujian 2007 luzhangong 0 1
62 - fujian 2007 luzhangong 0 0
65 + gansu 1993 gujinchi 0 1
65 - gansu 1993 gujinchi 0 0
74 + gansu 2002 songzhaosu 0 1
74 - gansu 2002 songzhaosu 0 0
79 + gansu 2007 luhao 0 1
79 - gansu 2007 luhao 0 0
80 + gansu 2008 luhao 0 1
80 - gansu 2008 luhao 0 0
81 + gansu 2009 luhao 0 1
81 - gansu 2009 luhao 0 0
82 + guangdong 1993 xiefei 0 0
83 + guangdong 1994 xiefei 0 0
84 + guangdong 1995 xiefei 0 0
85 + guangdong 1996 xiefei 0 0
86 + guangdong 1997 xiefei 0 0
97 + guangdong 2008 wangyang 0 1
97 - guangdong 2008 wangyang 0 0
98 + guangdong 2009 wangyang 0 1
98 - guangdong 2009 wangyang 0 0
99 + guangxi 1993 zhaofulin 0 1
100 + guangxi 1994 zhaofulin 0 1
101 + guangxi 1995 zhaofulin 0 1
102 + guangxi 1996 zhaofulin 0 1
103 + guangxi 1997 zhaofulin 0 1
112 + guangxi 2006 liuqibao 0 1
112 - guangxi 2006 liuqibao 0 0
Complementarity Between Connections and Growth (Provincial Party Secretary Only)
Dependent variable:
Promotion
(1) (2) (3) (4)
Growth 0.0003 0.0004 -0.002
(0.027) (0.028) (0.029)
Connection -0.210 -0.292 -1.536**
(0.492) (0.524) (0.617)
Connection x Growth 2.340**
(1.019)
Fixed Effect Y Y Y Y
Control covariates N N N N
Note: p<0.1; p<0.05; p<0.01
Cluster Standard Error on Province Level

3.2 Use our data to run LLD’s model

grp chng_type Province Year Name Promotion Connection
1 + anhui 1999 huiliangyu 0 0
1 - anhui 1999 huiliangyu 1 0
2 - anhui 2003 wangtaihua 0 1
3 + anhui 2004 wangtaihua 0 0
4 - anhui 2006 guojinlong 0 0
5 + anhui 2007 guojinlong 0 0
7 - chongqing 2001 heguoqiang 1 1
8 + chongqing 2002 heguoqiang 1 0
9 - chongqing 2004 huangzhendong 0 0
10 + chongqing 2005 huangzhendong 0 0
11 - chongqing 2006 wangyang 0 1
12 + chongqing 2007 wangyang 1 0
13 - fujian 1999 chenmingyi 0 0
14 + fujian 2000 chenmingyi 0 0
15 + fujian 2003 songdefu 0 1
15 - fujian 2003 songdefu 0 0
16 - gansu 1999 sunying 0 0
17 + gansu 2000 sunying 0 0
18 - gansu 2002 songzhaosu 0 0
19 + gansu 2003 songzhaosu 0 0
20 - gansu 2005 surong 0 1
21 + gansu 2006 surong 0 0
22 - guangdong 2001 lichangchun 1 1
23 + guangdong 2002 lichangchun 1 0
24 - guangdong 2006 zhangdejiang 1 1
25 + guangdong 2007 zhangdejiang 1 0
27 - guangxi 2006 liuqibao 0 0
28 + guangxi 2007 liuqibao 0 1
30 - guizhou 2004 qianyunlu 0 0
31 + guizhou 2005 qianyunlu 0 0
32 - hainan 2000 duqinglin 0 0
33 + hainan 2001 duqinglin 0 0
33 - hainan 2001 baikeming 0 1
34 + hainan 2002 baikeming 0 1
34 - hainan 2002 wangqishan 0 1
35 - hainan 2005 wangxiaofeng 0 0
36 + hainan 2006 wangxiaofeng 0 0
38 - hebei 2001 wangxudong 0 1
39 + hebei 2002 wangxudong 0 1
40 - hebei 2006 baikeming 0 1
41 + hebei 2007 baikeming 0 0
43 - heilongjiang 2004 songfatang 0 0
44 + heilongjiang 2005 songfatang 0 0
45 + heilongjiang 2007 qianyunlu 0 0
45 - heilongjiang 2007 qianyunlu 1 0
46 - henan 1999 mazhongchen 0 1
47 + henan 2000 mazhongchen 0 0
48 - henan 2001 chenkuiyuan 0 1
49 + henan 2002 chenkuiyuan 0 1
50 - henan 2003 likeqiang 0 1
Promotion of Party Secretary upon Term Completion (Relative Fiscal Revenue Growth to Competitors)
Dependent variable:
Promotion
(1) (2) (3) (4)
Relative Fiscal Revenue Growth 0.173** 0.106* 0.161** 0.087*
(0.066) (0.053) (0.064) (0.051)
Connection -0.100 -0.195* -0.102 -0.202*
(0.125) (0.100) (0.130) (0.105)
Connection x Growth 0.088 0.114
(0.219) (0.157)
Year Fixed Effect Y Y Y Y
Local Characteristics Y Y Y Y
Politician Characteristics N Y N Y
Note: p<0.1; p<0.05; p<0.01
Cluster Standard Error on Province Level
Promotion of Party Secretary upon Term Completion (Relative GDP Growth to Competitors)
Dependent variable:
Promotion
(1) (2) (3) (4)
Relative GDP Growth 0.208** 0.155** 0.230** 0.177*
(0.084) (0.073) (0.104) (0.096)
Connection -0.189 -0.264*** -0.186 -0.264***
(0.115) (0.098) (0.114) (0.098)
Connection x Growth -0.089 -0.077
(0.145) (0.137)
Year Fixed Effect Y Y Y Y
Local Characteristics Y Y Y Y
Politician Characteristics N Y N Y
Note: p<0.1; p<0.05; p<0.01
Cluster Standard Error on Province Level

4 Conclusion

To gather things up, we created following table to compare the results we got so far.
Variable Original_Estimation Replicated_Estimation
Jia et.al (2015)
Economic Performance 0.045 -0.002
Political Connection -0.475 -1.536**
Connection x Performance 2.865*** 2.340**
Landry et.al (2017)
Economic Performance 0.087 0.177*
Political Connection -0.007 -0.264***
Connection x Performance 0.079 -0.077
Note:
* p<0.1; ** p<0.05; *** p<0.01

Using our data to replicate the same setting of Jia et al.(2015) and Landury et al.(2017), we find although there is a negative significant effect of political connection, which may because we use a broader definition of connection than the former paper, its conclusion (“complementary of performance and connection decides promotion”) still holds; while for Landury et al.(2017), we find the economic performance still matters on the provincial party secretaries’ promotion. This is not a surprising finding because our reconstructed data has much more differences in Landry’s data than in Jia’s. In other words, the study on this topic differed from each other even on the start point.

We recognized the strong demand for a well collected, carefully processed, and widely accepted dataset for the study of this critical questions. In our further plan, we want to continue carrying on this challenging task and make this dataset a high-quality public good for the academic researchers.