Machine-learning analysis of leadership formation in China to parse the roles of loyalty and institutional norms
Edited by Jean Hong, University of Michigan, Ann Arbor, MI; received March 29, 2023; accepted September 27, 2023 by Editorial Board Member Margaret Levi
Significance
Machine-learning applications have great potential to offer valuable insights into a country’s political dynamics. This study represents an attempt to examine the promotion outcomes of Chinese high-ranking officials by adopting machine-learning techniques. For this purpose, we established a comprehensive database containing biographical and career information for 5,110 cadres of the Communist Party of China. By incorporating 250 individual features constructed from the database, our machine-learning model achieved an impressive accuracy level of almost 80% for promotion outcomes. This study highlights that Xi Jinping promoted loyalists to top leadership roles, even if they deviated from traditional promotion norms. The methodology and findings of this research shed additional light on the typically clandestine process of political selection in authoritarian regimes.
Abstract
A thriving cottage industry has long tried to predict the selection outcomes of the Chinese leadership using qualitative judgments based on historical trends and elite interviews. This study contributes to the discourse by adopting machine-learning techniques to quantitatively and systematically evaluate the promotion prospects of Chinese high-ranking officials. By incorporating over 250 individual features of approximately 20,000 high-ranking positions from 1982 to 2020, this paper calculated predicted probabilities of promotion for the 19th Politburo members of the Communist Party of China. The rankings of the promotion probabilities can be used not only to identify candidates who would have traditionally advanced within the party’s promotion norms but also to gauge Xi Jinping’s personal favoritism toward specific individuals. Based on different specifications for positions and periods, we developed measurements to quantify candidates’ levels of perceived loyalty and promotion eligibility. The empirical results demonstrated that the newly formed 20th Politburo Standing Committee was predominantly composed of loyalists who would not have risen to such positions under conventional promotion standards. We further found that, even within his circle of known allies, Xi Jinping did not opt for candidates with strong credentials. The findings of this study underscore the increasing emphasis on loyalty and the diminishing role of institutional norms in China’s high-ranking selections.
Sign up for PNAS alerts.
Get alerts for new articles, or get an alert when an article is cited.
Due to the unprecedented advancement in computing power, machine-learning technology has been integrated into various aspects of our lives, being widely used in sciences, engineering, and industry. In the study of politics, machine-learning applications to political outcomes hold great promise, particularly when paired with a consistent flow of high-quality data. It can deliver dynamic insights into the ever-evolving political landscape, often providing a more profound understanding than the static theoretical frameworks that much of the current scholarship relies upon.
Machine-Learning Adoption to Authoritarian Selection
Political selection is a key area of study in political science. In democratic systems, there is thriving literature on predicting electoral outcomes, with researchers using public surveys and other publicly accessible information to identify voter preferences (1). Noticeably, cutting-edge methodologies, including machine-learning models, have been gradually adopted in these electoral predictions to analyze extensive and complicated data from diverse sources, such as social media interactions, facial recognition, and web-traffic information (2–4). In authoritarian regimes, selection mechanisms are generally kept secret from the public. However, the increasing accessibility of data and advancement in computing techniques may offer new opportunities for examining the hidden selections made by authoritarian rulers.
Classical machine-learning models, such as naive Bayes and logistic regression, emphasize the importance of large sample sizes to guarantee prediction accuracy (5). While recent studies on electoral predictions in democracies can benefit from machine-learning methods due to the growing availability of data sources, the opaque nature of selection in authoritarian regimes poses a challenge in obtaining the necessary sample sizes for machine-learning predictions. However, it is essential to note that small sample sizes are not uncommon in machine-learning applications, especially in human-related domains, such as neuroimaging, genomics, and motion tracking. Recent advancements in these fields, facilitated by the integration of more advanced models such as deep learning algorithms, demonstrate the potential to overcome the problem of small sample sizes by incorporating a large number of features for each sample (6).
Given these considerations, this study focuses on elite selections of the Communist Party of China (CPC). The CPC has released detailed career profiles for high-ranking officials, providing ample information to generate myriad individual features for selectors and selectees. Such an abundance of official profiles allows for the creation of high-dimensional data that mitigates the small sample bias in the machine-learning process. Moreover, as one of the oldest and largest political parties in the world, the CPC has a long history of producing numerous rounds of selection outcomes. Although the sample sizes may not rival those in democracies, this relatively large number of observations renders China an ideal starting point for machine-learning applications to authoritarian selection.
Political Selection in China
In October 2022, Li Qiang, Cai Qi, Ding Xuexiang, and Li Xi were newly appointed to the 20th Politburo Standing Committee, the highest decision-making body of the CPC. However, concerns quickly arose about their qualifications, as it appeared that the primary selection criterion was personal connections with Xi Jinping rather than official credentials or performance records. All members of this committee seem to have previous work or family ties with Xi, either directly or indirectly. However, close associations with Xi do not automatically imply that these officials are unqualified. It is possible that they cultivated particular sets of skills and acquired specific credentials through their collaborative endeavors with Xi, thereby aligning them with the established norms of promotion within the party.
Many scholars specializing in Chinese politics attempted to identify the mechanisms behind Xi Jinping’s selections (7). However, these studies primarily depend on qualitative evidence, such as case studies or historical analysis, which may make inferences on the basis of a limited period or perspective. Instead, this study calculates predicted probabilities of promotion to quantitatively and systematically evaluate an official’s qualifications for the current positions using historical promotion patterns and individual characteristics of preceding cohorts of officials. However, merely relying on these predicted probabilities might not fully elucidate the intricacies of the CPC’s promotion logic. Rather, comparing the predicted outcomes dictated by established promotion norms with actual outcomes observed under Xi Jinping’s more personalistic leadership offers clearer evidence of the shifting promotion criteria within the CPC.
Studies on authoritarian governance suggest that a key factor contributing to authoritarian longevity is the promotion norms embedded in the ruling party, which motivate members to actively perpetuate the party’s dominant roles in the regime (8). In the context of the CPC, since the 1980s, the leadership has set up explicit guidelines for promoting cadres to influential positions, prioritizing qualities, such as being “politically reliable, young, educated, and professional” (9). On the other hand, research on authoritarian selections proposes that paramount rulers may value loyalty over competence, even favoring those who might not necessarily meet the highest standards of established promotion norms (10). In essence, if high-ranking officials are selected based on institutional norms, the ruler faces the risk of delegating substantial authority to those who might not remain loyal or support the ruler during moments of crisis (11).
Measuring Loyalty and Institutional Norms
As outsiders, how can we indirectly measure a candidate’s loyalty to the ruler and compliance with existing institutional norms? Previous research on Chinese elites focuses on a limited set of indicators to evaluate the above characteristics. For example, sharing personal traits, such as birthplace, university alma mater, or work experience, is interpreted as a sign of loyalty (12). In a similar vein, adherence to promotion norms has traditionally been determined using a few biographical and career-related variables, such as age, education level, administrative experience, and regional growth performance (13).
Table 1 presents the measurement schemes for loyalty and institutional norms in this study. Notably, this paper develops a unique method to evaluate Xi’s assessments of his subordinates’ loyalty. As the regime becomes more personalistic, the ruler tends to value loyalty over institutional norms to compensate for its lack of legitimacy (14). Accordingly, ample anecdotal evidence suggests that Xi has placed more of his loyalists in influential positions than his predecessors (7). Therefore, by contrasting Xi’s promotion patterns with those of his predecessors, this study measures the levels of Xi’s favoritism toward candidates.
Table 1.
| Loyalty | Institutional norms | |
|---|---|---|
| Theory | Favoritism and trust | Qualification and eligibility |
| Meaning | How much a candidate has the incentive to be loyal to Xi | How much a candidate is eligible to acquire a promotion |
| Measurement | Differences in promotion prospects between under Xi and under his predecessors | Promotion prospects based on promotion patterns during the post-Mao period |
| Representation |
For this purpose, this study first calculates a predicted probability of promotion, , as a proxy to measure an individual’s promotion prospects. The ranking of the predicted probability of promotion, , indicates a candidate’s relative promotion prospects compared to the contemporaries. Then, the dataset is divided into two parts: pre-Xi and Xi periods. The pre-Xi data include promotion patterns from Xi’s predecessors, such as Deng Xiaoping, Jiang Zemin, and Hu Jintao, while the Xi data capture the promotion patterns under Xi Jinping’s leadership. By comparing a candidate’s ranking of promotion prospects based on models trained on the pre-Xi (i.e., ) and Xi (i.e., ) periods, we can measure the counterfactual promotion prospects of the candidate without Xi Jinping’s influence. In other words, the differences in these rankings quantify how much candidates’ potential for promotion would differ hypothetically if they were working under the pre-existing promotion norms of Xi’s predecessors.
To illustrate, if a candidate’s ranking of promotion probabilities is notably higher in the model derived from the Xi data compared to the pre-Xi data, it indicates that the candidate is likely favored by Xi, presumably due to perceived loyalty. Conversely, if a candidate’s ranking under Xi is lower than under previous leaders, the candidate may not be a trusted ally of Xi but would have fared better under established promotion norms. By directly measuring Xi’s preference for his subordinates, this study can mitigate the occurrence of false-positive cases in assessing an official’s level of loyalty (e.g., categorizing a candidate as a Xi loyalist based on one or two proxies despite not being trusted by Xi).
In evaluating a candidate’s alignment with the existing norms for promotion, we measure how much the candidate is eligible for promotion based on historical patterns of promotion during the post-Mao era (1982 to 2020). Promotion eligibility has been closely associated with a candidate’s credentials, political reliability, experience, and performance (15). Throughout the entire post-Mao period, officials who excelled in these aspects were more likely to acquire promotions (16). It is a sufficient time frame for Chinese officials to develop a collective understanding of the shared promotion norms, partly embedded in internal guidelines.
It is important to note that this study does not entirely dismiss the impact of patronage as a factor in existing institutional norms for promotion. Along with official qualifications and performance records, many Chinese officials believe that political connections play a role in gaining recognition and appreciation from superiors (17). However, during the pre-Xi period, when personnel decisions were made collectively, officials could not rely solely on a single patron. Instead, rulers needed to consider the broader sentiments of the elite group by skillfully balancing various political interests (7). This complex nature of promotion eligibility, encompassing both formal and informal components, is expected to be captured by the ranking of predicted promotion probabilities derived from the complete data of the post-Mao period (1982 to 2020), represented as .
Machine-Learning Process
The primary target of this study is the 19th Politburo members of the CPC. As a building block to measure the levels of loyalty and adherence to institutional norms, this study first derives the predicted probability of promotion by employing machine-learning techniques. This promotion probability is estimated based on a plethora of both formal and informal characteristics of high-ranking officials at central and provincial levels, spanning the years 1982 to 2020. The analysis encompasses three levels of central officials: Politburo members (POLI), Central Committee members (CC), and Alternate Central Committee members (ACC). In the case of provincial officials, there are four levels: Provincial Party Secretaries (PS), Governors (GOV), Vice Party Secretaries (VPS), and ordinary Provincial Standing Committee members (SC).
From 1982 to 2020, 1,578 (8.7%) promotions occurred out of 18,179 observations for central and provincial high-ranking officials. Setting the promotion outcomes of Chinese officials as an outcome variable, this study incorporates over 250 individual characteristics as predictors, including biographical (e.g., age, gender, education level, and ethnicity), career (e.g., seniority, tenure period, administrative experience, and job diversity), context (e.g., geopolitical conditions of workplaces), factional (e.g., competing affiliations in top-level Chinese politics), and network (e.g., political connections and centrality indices) information. Detailed variable descriptions can be found in SI Appendix, 1 and 2.
The accuracy of machine-learning predictions is heavily dependent on the breadth of data used to train the model. In this study, the baseline model utilizes the complete dataset of central and provincial officials from 1982 to 2020 in order to assess an official’s adherence to pre-existing promotion norms within the party. Subsequently, to measure the levels of loyalty, different training sets are employed to better capture the rulers’ personal preferences in high-ranking selections. The training set of the loyalty model includes only central officials (i.e., POLI, CC, and ACC) and is partitioned into two time periods: pre-Xi and Xi. This setting allows us to juxtapose an official’s promotion probabilities under Xi to the counterfactual probabilities under his predecessors, such as Deng Xiaoping, Jiang Zemin, and Hu Jintao.
There are numerous machine-learning algorithms that estimate predicted probabilities of promotion in various fashions. Among them, we constructed an ensemble model that synthesizes seven different techniques, including Generalized Linear Model, Least Absolute Shrinkage and Selection Operator, K-nearest Neighbor, Gradient Boosting Machine, Random Forest, Support Vector Machine, and Neural Net. Each technique has its own algorithm for selecting important variables, removing irrelevant ones, and determining a sufficient set for predictions. The final model for each technique is chosen based on the highest area under the curve (AUC) of the test set among 1,000 iterations using 5-fold cross-validation. The ensemble model combines the results of the individual methods into a single model by a weighted majority vote to maximize its prediction accuracy. For detailed information on the machine-learning procedures used in this study, please refer to SI Appendix, 4.
The final ensemble model is harnessed to generate predicted probabilities of promotion for the prediction set (i.e., the likelihood of the 19th Politburo members advancing to the 20th Politburo Standing Committee members). In our analysis, the accuracy of the ensemble model was over 0.7 for most specifications, as measured by the AUC. While there is no statistical test to decide whether this value represents a good fit, a value of 0.7 is generally considered a strong effect (18). In fact, the model trained with the full dataset achieved a peak prediction accuracy of 0.846, underscoring the effectiveness of the machine-learning methods in predicting promotion outcomes for Chinese high-ranking officials.
Machine-Learning Results
The result of the baseline model, trained using the promotion patterns of central and provincial officials from 1982 to 2020, is presented in Table 2. Results from different specifications for positions and times can be found in SI Appendix, 6. According to the baseline model, the top four predictions from the 19th Politburo to the newly formed 20th Politburo Standing Committee are Hu Chunhua, Wang Chen, Ding Xuexiang, and Huang Kunming. Given the realized four-seat turnover in the 20th Politburo Standing Committee, these four officials would likely have been elevated under the promotion norms established during the post-Mao period (1982 to 2020).
Table 2.
| Central+Provincial Officials (POLI+CC+ACC+PS+GOV+VPS+SC) | |
|---|---|
| Rank | 1982 to 2020 |
| 1 | Hu Chunhua |
| 2 | Wang Chen |
| 3 | Ding Xuexiang |
| 4 | Huang Kunming |
| 5 | Yang Jiechi |
| 6 | Chen Min’er |
| 7 | Chen Quanguo |
| 8 | Liu He |
| 9 | Guo Shengkun |
| 10 | Zhang Youxia |
| N | 18,179 |
| AUC | 0.846 |
Undoubtedly, the credentials of Hu Chunhua are extraordinary. Hu served as the vice premier of China in the 19th Politburo and was the youngest member of the 18th Politburo. After completing his tenure as the head of the Communist Youth League, he worked in provinces, such as Tibet, Hebei, and Inner Mongolia, which were considered economically impoverished. Recruiting young officials with experience in these underprivileged regions is a prominent strategy by the CPC leadership to cultivate potential national leaders (9). Moreover, officials groomed by the Communist Youth League typically have promising career trajectories, due to the CPC leadership’s recognition of the distinctive experiences the Youth League offers in championing communist values with Chinese characteristics (19).
The absence of Hu Chunhua in the 20th Politburo Standing Committee signals a decline in the importance of traditional selection criteria such as age and administrative experience under Xi Jinping’s leadership. The appointment of Li Qiang, Cai Qi, Ding Xuexiang, and Li Xi, all recognized as Xi’s allies, to the new standing committee highlights Xi’s significant reliance on personal connections. As depicted in Table 2, even among Xi’s known followers, the top four predictions are Wang Chen (connected with Xi through experience in Shaanxi), Ding Xuexiang (Shanghai), Huang Kunming (Zhejiang), and Chen Min’er (Zhejiang). However, among these names, only Ding Xuexiang could secure a position in the new Politburo Standing Committee.
Strikingly, most of the actual appointees, namely Li Qiang, Cai Qi, and Li Xi, do not stand out as particularly qualified according to the pre-existing promotion norms. This difference between the baseline predictions and the actual selections provides insight to evaluate Xi’s perceived loyalty toward candidates. To this end, this study divides the training sets into pre-Xi (1982 to 2011) and Xi (2012 to 2020) periods. The analysis exclusively focuses on central officials (i.e., POLI, CC, and ACC) as the relatively small size of the Central Committee is expected to better reflect Xi Jinping’s personal preferences.
Table 3 shows a significant discrepancy between the lists generated by the pre-Xi and Xi models. When the model was trained solely on promotion patterns during Xi Jinping’s leadership, its top predictions are Li Xi, Li Qiang, Chen Min’er, and Cai Qi. In contrast, the model trained on data from Xi’s predecessors produced similar results to the baseline model. By contrasting the pre-Xi and Xi models, we can assess Xi’s personal favoritism toward candidates and the corresponding incentives for candidates to be loyal to him. If a candidate’s promotion prospects are notably higher under Xi than under his predecessors, it suggests that the candidate holds a vested interest in upholding Xi’s leadership.
Table 3.
| Central officials (POLI+CC) | ||
|---|---|---|
| Rank | Pre-Xi (1982 to 2011) | Xi (2012 to 2020) |
| 1 | Hu Chunhua | Li Xi |
| 2 | Chen Quanguo | Li Qiang |
| 3 | Yang Jiechi | Chen Min’er |
| 4 | Wang Chen | Cai Qi |
| 5 | Zhang Youxia | Yang Xiaodu |
| 6 | Liu He | Zhang Youxia |
| 7 | Guo Shengkun | Sun Chunlan |
| 8 | Ding Xuexiang | Huang Kunming |
| 9 | Li Hongzhong | Hu Chunhua |
| 10 | Huang Kunming | Xu Qiliang |
| N | 1,147 | 198 |
| AUC | 0.809 | 0.704 |
As a result, the 19th Politburo members can be divided into two groups with differing incentives for loyalty. Fig. 1 illustrates the changes in promotion prospects for each member under the pre-Xi and Xi standards. The Left plot shows a group of candidates who have a higher probability of promotion under Xi’s leadership, indicating a greater incentive to align themselves with Xi. For instance, Li Xi’s promotion probability is the highest in the Xi model, whereas his ranking drops sharply to the 15th in the pre-Xi model. This disparity implies that Li Xi’s loyalty might arise from Xi’s favoritism, an advantage he would not have enjoyed under the party’s conventional promotion norms.
Fig. 1.
On the other hand, the right plot demonstrates the candidates who were penalized by Xi’s selection mechanisms. One vivid example in this group is Hu Chunhua. Traditionally, meritocratic criteria that prioritize official qualifications and performance records would favor Hu Chunhua. The pre-Xi model also confirms this perspective by bestowing on him the highest predicted ranking. However, in the Xi model, Hu’s promotion prospects plummeted to the ninth position, indicating Xi’s potential indifference or even aversion to candidates with Hu’s credentials. More importantly, Hu Chunhua may have less incentive to be loyal to Xi because his promotion opportunities were deprived by Xi. In reality, Hu failed to retain his position in the Politburo after the 20th Party Congress.
The result of the baseline model in Table 2 is used to evaluate a candidate’s level of adherence to pre-existing promotion norms. It aims to encapsulate the collective wisdom of Chinese officials regarding promotion eligibility, which they have nurtured throughout their careers, spanning from the post-Mao era to the present. Since most of the time period in the baseline model predates the Xi Jinping period, the prediction outcomes turned out to be similar to those of the pre-Xi model. However, it is crucial to acknowledge the importance of evaluating promotion eligibility within the frameworks of both provincial contexts and Xi Jinping’s leadership. For example, officials who joined the CPC in the 1980s still remain politically active, having witnessed promotion trends across a range of central and provincial roles throughout the entire post-Mao era. Consequently, when assessing adherence to institutional norms, the model that encompasses the entire time span and all positions proves to be more appropriate than the pre-Xi model, which only incorporates the pre-Xi period and central promotions.
Evidently, this paper also illustrates that candidates identified as norm-compliant (determined by the ranking of the baseline model) align with the traditional understanding of qualifications according to the party’s existing norms. This stands in contrast to candidates identified as loyalists (determined by the differences in rankings between the pre-Xi and Xi models). Upon examining the variable importance indexes and descriptive statistics of both groups, it becomes apparent that norm-compliant candidates generally possess stronger attributes in terms of age, education, and professional competency. For more detailed information, please refer to SI Appendix, 7.
Loyalty versus Institutional Norms in Leadership Formation
As a selector of his own high-ranking subordinates, Xi Jinping must take into account two aspects of candidates: loyalty and eligibility. Fig. 2 visualizes the levels of loyalty and eligibility for the 19th Politburo members, showing four categories of candidates: loyal and norm-compliant, loyal but norm-deviating, disloyal and norm-deviating, and disloyal but norm-compliant. Considering the “seven up and eight down” (qishangbaxia) referring to the convention that officials who are 67 or younger are eligible for consideration as members of the Politburo Standing Committee, nine Politburo members of the 19th Party Congress can be potential candidates.
Fig. 2.
Li Xi, Cai Qi, and Li Qiang belong to the loyal but norm-deviating group. Given their relatively weaker eligibility under the existing party norms, Xi’s direct involvement would be necessary to ensure their appointments to high-ranking positions. The selection outcomes of the 20th Party Congress substantiate the findings of this study, as all three individuals were appointed as members of the Politburo Standing Committee. Under a more personalized leadership, rulers like Xi Jinping are prone to choosing officials who exhibit loyalty but have comparatively weaker credentials for prominent roles, due to their perceived lower threat and greater support for his regime.
While Chen Min’er and Huang Kunming also possess strong incentives to remain loyal to Xi Jinping, their staunch adherence to the existing promotion norms ironically might have hindered their advancement under Xi. According to traditional selection mechanisms, they are eligible candidates whose qualifications do not deviate far from the conventional standards for promotion. However, their high levels of norm-compliance also render them potential threats to Xi’s rule, given that their current political standing is not solely dependent on Xi Jinping’s personal support. In this context, Xi’s preference for loyal but norm-deviating candidates becomes more understandable, as this group is less likely to pose challenges to Xi’s authority. Consequently, despite Chen and Huang being well-known followers of Xi Jinping, they indeed remained in the Politburo without receiving promotions during the 20th Party Congress.
Li Hongzhong stands as the sole member of the disloyal and norm-deviating group, marked by a weak propensity for loyalty and limited alignment with the existing norms. Xi would show a preference for such candidates over those who are disloyal yet strongly comply with the established promotion norms. This preference is mainly because the latter group may present a greater threat given their potential ability to challenge Xi’s authority. Hu Chunhua and Chen Quanguo serve as examples of candidates in the disloyal but norm-compliant group. Xi acknowledges that they have little motivation to be loyal to him, but their qualifications are robust enough to garner support and admiration from circles outside Xi’s direct sphere of influence. Indeed, both Hu and Chen were expelled from Politburo at the 20th Party Congress.
The case of Ding Xuexiang in the disloyal but norm-compliant group provides valuable insights into both the potential implications and the inherent limitations of this study. Compared to other allies of Xi, Ding’s incentive to remain loyal may be weaker, as his credentials are already sufficient to secure a high-ranking position. Even under Jiang Zemin or Hu Jintao, Ding’s noteworthy background encompassing his tenure in Shanghai, graduation from a prestigious university, expertise in science and technology, and various administrative and party roles would have positioned him as a qualified candidate for promotion, irrespective of Xi’s personal support. The appointment of Ding Xuexiang to the new Politburo Standing Committee contradicts the prediction of our model. One explanation might be that Xi underestimated Ding’s credentials when he initially took Ding under his wing during his brief stint in Shanghai. Another explanation, one beyond the scope of this study, is that Ding performed an unobserved service for Xi to earn his trust, which apparently paid off handsomely. Due to the secret nature of certain behind-the-scenes maneuvers in authoritarian politics, the data-driven approach to predicting promotions primarily gauges the ruler’s preferences toward candidates, rather than achieving flawless prognostication of actual outcomes.
Implications and Concluding Remarks
How much a ruler values institutional norms over loyalty in political appointments largely hinges on the domestic political environment of the country. In regimes characterized by a more personalistic nature, the ruler is inclined to attach greater importance to loyalty as a way to offset the lack of legitimacy and stability (14). China serves as an excellent example in this regard. In the post-Mao era, Deng Xiaoping initiated a shift toward more collective-based institutions for selecting qualified and competent subordinates (20). However, under the leadership of Xi Jinping, there has been a reversal of Deng’s decisions with an accentuation on personalistic and authoritative rules (7). Consequently, the significance of having qualified subordinates has diminished, while there is now a heightened incentive to prioritize loyalty.
The majority of the top predictions from the baseline model were not able to secure positions in the Politburo Standing Committee. The popular explanation for Hu Chunhua’s failure, positing his affiliation with the Youth League, is not sufficient. Other top predictions, such as Wang Chen and Huang Kunming, were also not chosen for high-ranking positions despite their personal ties with Xi Jinping. They might be in a sense too qualified for Xi’s comfort. Instead, Xi appointed Li Qiang to the premier, the second-highest rank in the leadership. Unlike Hu Chunhua whose credentials were often regarded as the prime contender for the premiership, Li Qiang even lacked the experience as a vice premier, a role traditionally deemed crucial for assuming the premier position.
Furthermore, Li Qiang’s level of competence has not been proven through the established norms of promotions that value administrative experience in less-developed provinces (9). Li Qiang has held diverse administrative roles in regions, such as Zhejiang, Jiangsu, and Shanghai. However, given that these areas are already well developed, it is hard to discern whether his achievements can be attributed solely to his capability rather than the inherent advantages of those regions. Interestingly, the machine-learning model detects the distinct characteristics of each experience and factors them into its calculation of promotion prospects. According to the baseline model in Table 2, Li Qiang, who accumulated professional experiences in prosperous provinces, received a considerably lower ranking compared to Hu Chunhua, who operated in relatively underdeveloped regions, even though both had an equivalent number of administrative roles (Li Qiang in Zhejiang, Jiangsu, and Shanghai, and Hu Chunhua in Tibet, Hebei, and Inner Mongolia).
The fact that the new Politburo Standing Committee members were generally ranked lower in the baseline model (e.g., Li Qiang ranked 14th out of 19 candidates, Li Xi 16th, and Cai Qi 17th) suggests that they were not the most qualified candidates according to the historical patterns of promotion. This outcome implies that Xi Jinping has deviated from the established meritocratic norms and instead prioritized his personal relationship in high-ranking selections. Such a move toward a more personalistic leadership style raises concerns about the dictator’s dilemma for Xi: The more power one acquired, the greater the likelihood of facing challenges and being overthrown. Xi can never be entirely sure of his subjects’ loyalty, as they tend to conceal their true preferences and only feign support for the ruler. Hence, Xi may prefer loyalty over qualifications, fearing that highly credentialed subordinates could be more rebellious. Overall, Xi’s personalization of leadership may temporarily stabilize his position, but it may undermine the CPC’s long-term capability to address the country’s intricate socio-economic challenges.
Materials and Methods
Database Description.
We established a CPC elite database containing biographical and career data for 5,110 CPC cadres, including all the central committee and provincial standing committee members from 1976 to 2021. In the database, we provide detailed biographical information including birth year, gender, ethnicity, birthplace, education level, university alma maters, year of entering and retiring from the CPC, and all the job trajectories that each cadre had experienced since 1949. The database encompasses exceptionally comprehensive career information, as it can virtually identify thousands of different affiliations, including 645 party, 550 government, 394 military, and 77 social organizations in addition to 2,411 administrative units and 1,040 universities. Overall, it can track all of the ranks of the positions that individual officials held during their tenure, totaling over 13,500 job specifications. The data collection process is included in SI Appendix, 3.
Measuring Promotion.
Promotion is defined as whether an official is promoted or transferred to a more desirable position in a given year. One example is an official appointed to a one-level or higher rank in the administrative hierarchy (e.g., CC to POLI, ACC to CC, GOV to PS, VPS to GOV, and SC to VPS). Additionally, transfers to central organizations are also treated as promotions, as working at any central department grants greater political authority and a higher probability to gain a leading role in the near future. However, promotions or transfers to the nominal and “second-line” positions in the center, such as the National People’s Congress and the National People’s Political Consultative Conference, are not included in the analysis. Please kindly note that alternative specifications for promotions are presented in SI Appendix, 5.
Statistical Package.
This study mainly used R with the Classification and Regression Training package for machine-learning analysis.
Data, Materials, and Software Availability
The data and associated descriptions are available at https://github.com/jonghyuklee/ml_cpc (21).
Acknowledgments
We thank Susan L. Shirk and Margaret E. Roberts at the University of California San Diego in the United States, Mingjiang Li at Nanyang Technological University in Singapore, Heeok Lee at Sungkyunkwan University in Korea, and Kentaro Fukumoto at Gakushuin University in Japan for their valuable comments.
Author contributions
J.L. and V.C.S. designed research; J.L. performed research; J.L. contributed new reagents/analytic tools; J.L. analyzed data; J.L. and V.C.S. data contribution; J.L. wrote the paper; and J.L. and V.C.S. reviewed and edited the manuscript.
Competing interests
The authors declare no competing interest.
Supporting Information
Appendix 01 (PDF)
- Download
- 683.76 KB
References
1
P. Hummel, D. Rothschild, Fundamental models for forecasting elections at the state level. Elect. Stud. 35, 123–139 (2014).
2
C. C. Ballew, A. Todorov, Predicting political elections from rapid and unreflective face judgments. Proc. Natl. Acad. Sci. U.S.A. 104, 17948–17953 (2007).
3
T. Yasseri, J. Bright, Wikipedia traffic data and electoral prediction: Towards theoretically informed models. EPJ Data Sci. 5, 1–15 (2016).
4
A. Hasan, S. Moin, A. Karim, S. Shamshirband, Machine learning-based sentiment analysis for twitter accounts. Math. Comput. Appl. 23, 1–15 (2018).
5
S. J. Raudys, A. K. Jain, Small sample size effects in statistical pattern recognition: Recommendations for practitioners. IEEE Trans. Pattern Anal. Mach. Intell. 13, 252–264 (1991).
6
A. Vabalas, E. Gowen, E. Poliakoff, A. J. Casson, Machine learning algorithm validation with a limited sample size. PLoS One 14, 1–20 (2019).
7
S. L. Shirk, The return to personalistic rule. J. Democracy 29, 22–36 (2018).
8
B. Geddes, What do we know about democratization after twenty years? Annu. Rev. Polit. Sci. 2, 115–144 (1999).
9
F. C. Teiwes, Normal politics with Chinese characteristics. China J. 45, 69–82 (2001).
10
A. V. Zakharov, The loyalty-competence trade-off in dictatorships and outside options for subordinates. J. Polit. 78, 457–466 (2016).
11
G. Egorov, K. Sonin, Dictators and their viziers: Endogenizing the loyalty-competence trade-off. J. Eur. Econ. Assoc. 9, 903–930 (2011).
12
V. C. Shih, J. Lee, Locking in fair weather friends: Assessing the fate of Chinese communist elite when their patrons fall from power. Party Polit. 26, 628–639 (2020).
13
P. F. Landry, X. Lu, H. Duan, Does performance matter? Evaluating political selection along the Chinese administrative ladderCompar. Polit. Stud. 51, 1074–1105 (2018).
14
B. Magaloni, R. Kricheli, Political order and one-party rule. Annu. Rev. Polit. Sci. 13, 123–143 (2010).
15
M. Edin, State capacity and local agent control in China: CCP cadre management from a township perspective. China Quart. 173, 35–52 (2003).
16
O. Blanchard, A. Shleifer, Federalism with and without political centralization: China versus Russia. IMF Staff Papers 48, 171–179 (2001).
17
H. Li, L. L. Gore, Merit-based patronage: Career incentives of local leading cadres in China. J. Contemp. China 27, 85–102 (2018).
18
M. E. Rice, G. T. Harris, Comparing effect sizes in follow-up studies: ROC area, Cohen’s d, and r. Law Hum. Behav. 29, 615–620 (2005).
19
J. Doyon, The strength of a weak organization: The communist youth league as a path to power in post-mao China. China Quart. 243, 780–800 (2020).
20
Y. Zhang, The successor’s dilemma in China’s single party political system. Eur. J. Polit. Econ. 27, 674–680 (2011).
21
J. Lee, V. C. Shih, Machine-learning analysis on elite selection in the Communist Party of China. CPC Elite Database. https://github.com/jonghyuklee/ml_cpc. Deposited 15 October 2023.
Information & Authors
Information
Published in
Cover image: Pictured is a 2D titanium carbide MXene loaded with nickel nanoparticles. Hongming Sun et al. developed a nickel nanoparticle catalyst for hydrogen oxidation in alkaline conditions, a reaction which is involved in the production of electricity by hydrogen fuel cells. The authors used titanium carbide MXene to tune the electronic structure of the nickel nanoparticles, resulting in a catalyst that outperformed commercial platinum-based catalysts. The results could inform the development of improved catalysts for renewable hydrogen technologies. See the article by Sun et al., e2308035120. Image credit: Hongming Sun.
Classifications
Copyright
Copyright © 2023 the Author(s). Published by PNAS. This article is distributed under Creative Commons Attribution-NonCommercial-NoDerivatives License 4.0 (CC BY-NC-ND).
Data, Materials, and Software Availability
The data and associated descriptions are available at https://github.com/jonghyuklee/ml_cpc (21).
Submission history
Received: March 29, 2023
Accepted: September 27, 2023
Published online: October 30, 2023
Published in issue: November 7, 2023
Keywords
Acknowledgments
We thank Susan L. Shirk and Margaret E. Roberts at the University of California San Diego in the United States, Mingjiang Li at Nanyang Technological University in Singapore, Heeok Lee at Sungkyunkwan University in Korea, and Kentaro Fukumoto at Gakushuin University in Japan for their valuable comments.
Author Contributions
J.L. and V.C.S. designed research; J.L. performed research; J.L. contributed new reagents/analytic tools; J.L. analyzed data; J.L. and V.C.S. data contribution; J.L. wrote the paper; and J.L. and V.C.S. reviewed and edited the manuscript.
Competing Interests
The authors declare no competing interest.
Notes
This article is a PNAS Direct Submission. J.H. is a guest editor invited by the Editorial Board.
Authors
Metrics & Citations
Metrics
Citation statements
Altmetrics
Citations
If you have the appropriate software installed, you can download article citation data to the citation manager of your choice. Simply select your manager software from the list below and click Download.
View Options
View options
PDF format
Download this article as a PDF file
DOWNLOAD PDFLogin options
Check if you have access through your login credentials or your institution to get full access on this article.
Personal login Institutional LoginRecommend to a librarian
Recommend PNAS to a LibrarianPurchase options
Purchase this article to access the full text.