In regimes that lack strong institutions to manage the problem of power sharing, charges of malfeasance are often exercises in high-level power politics to eliminate rivals—and this is how the literature conventionally studies them (see, for example, Chen and Weiss Reference Chen, Weiss, Chen and Weiss2019; Gillespie and Okruhlik Reference Gillespie and Okruhlik1991; Halfin Reference Halfin2003; Malesky and Phan Reference Malesky, Phan, Chen and Weiss2019; Zhu and Zhang Reference Zhu and Zhang2017). Such charges fit straightforwardly in a substantial literature on purges as “coup-proofing” (see, for example, Quinlivan Reference Quinlivan1999; Sudduth Reference Sudduth2017; Svolik Reference Svolik2012) in authoritarian regimes by the supreme political boss. However, when purges reach far below the inner circle, as often occurs, the coup-proofing logic leads us conceptually astray.
In this article, we contribute conceptually, theoretically, and empirically to our understanding of purges by developing several linked ideas. First, we introduce a new concept, putting some purges in a class we call “broad purges,” with characteristics and implications that the coup-proofing logic fails to illuminate. Broad purges extend below powerholders at the top to encompass a large number of plausible but indeterminate targets in an expansively defined elite. We find instances of them in many countries, in the past, and at present. Secondly, we focus on how broad purges affect political decision making: briefly, they grossly exacerbate uncertainty at the same time as they raise both the likelihood and the cost of political error. This encourages strategies of risk reduction throughout the political system. We present a theoretically based argument about bias against clients as a specific, always safe, strategy of risk reduction in political appointments during a broad purge. Thirdly, we shift analytic interest away from purge casualties to a large set of players overlooked in the literature, namely, individuals whose careers are neither ended nor stalled in the course of broad purges. These players are the purge winners. By definition, unlike casualties, their influence grows as purges progress and may continue long after purges end. Lastly, we consider the consequences of purges for political institutions. Even as risk reduction in decision making can be economically costly to the regime, broad purges can promote cohesion of the authoritarian party by weakening informal networks. The counterintuitive possibility of this indirect impact also distinguishes broad purges from coup-proofing purges.
We develop these ideas and test their implications empirically by analyzing decisions on political appointments during a massive corruption crackdown in China that has investigated and punished some 1.5 million officials since late 2012 (Gan and Choi Reference Gan and Choi2018). We characterize the crackdown as a broad purge. Scholars mostly study whether or not the crackdown disproportionately targets those who threaten the “factional network” of Xi Jinping, supreme leader of the Communist Party of China (CPC) (see, for example, Fu Reference Fu, Rose-Ackerman and Lagunes2015; Lorentzen and Lu Reference Lorentzen and Lu2018; Murong Reference Murong2015; Wedeman Reference Wedeman2017; Xi, Yao, and Zhang Reference Xi, Yao and Zhang2018; Zhu and Zhang Reference Zhu and Zhang2017). Others study it as a campaign to boost public support (see, for example, Sun and Yuan Reference Sun and Yuan2017; Wang and Dickson Reference Wang and Dickson2022; Zhu, Huang, and Zhang Reference Zhu, Huang and Zhang2019) or, indeed, to check endemic corruption (see, for example, Manion Reference Manion2016; Orlik Reference Orlik2014). More relevant to our research question is the idea that officials below the apex of power in Beijing lack a reliable basis to bet on any of these in an environment when the professional and personal stakes are vastly higher than before—and this greatly affects their political decision making.
We focus here on decisions on political appointments during the purge: the emergence of a cohort of purge winners. A broad purge creates more than the usual number of political vacancies to fill. In addition to these extraordinary vacancies, decisions on appointments below the inner circle routinely arise over any number of years, including purge years. At any time, such decisions are opportunities for local CPC bosses, who have appointment control in China, to build their own fiefs of patronage networks. Political appointment decisions also offer opportunities for local party bosses to signal to the party center in Beijing that they pose no political threat, however—which can be imperative in the uncertain environment of a broad purge. We theorize that signaling conformity to party organizational discipline by not investing in local networks is a safe strategy of risk reduction. We test our theory with political appointment decisions affecting the careers of chief party secretaries of China's 333 prefectures for the purge years 2013–17.Footnote 1 We theorize that the provincial party bosses of these prefectural officials bias appointments against officials who are easily identifiable as their own clients so as to signal to Beijing they are not building “factional networks.” More specifically, we expect variation in anti-client bias in appointments to reflect heterogeneity across provincial party bosses in how serious a threat the purge seems to present to their careers. We measure this threat as purges of centrally managed (CM) officials in each province. In the provinces, CM officials are senior political leaders whose careers are directly controlled from Beijing by the Central Organization Department (COD) of the CPC. Examples are vice-governors, standing committee members of provincial party committees, and party secretaries of cities at a half-rank below the provincial level.
The analysis using prefecture-year data shows a salient impact of these CM officials' purges on political appointments by provincial party bosses. Our findings, robust to a variety of specifications, support our theory of risk reduction through anti-client bias. Our most conservative estimate suggests that the probability of client appointment decreases by 2.5 percentage points with one additional centrally managed purge. As mean observed promotion probability is about 8 per cent, the effect of purge intensity is not only statistically significant, but also very substantial.
In sum, the distinctive feature of the cohort of purge winners is that before their appointments, they are perceptibly not beholden to their respective provincial bosses. We show that this, not any feature intrinsic to them as individuals, is what makes their appointments a credible signal of conformity to party organizational discipline for provincial party bosses who strive to protect their own careers and personal safety as they observe mounting purge casualties around them. This resembles the “connection penalty” for politburo appointments noted by Fisman et al. (Reference Fisman2020).
We conduct two key extensions to our analysis to better elucidate features of the broad purge. First, we investigate whether the only patronage network left intact in the purge's wake is Xi Jinping's own. We show that anti-client bias exists for both Xi's clients and others, and the difference in such bias is not statistically different between the two groups. Secondly, as a placebo test, we gauge the impact of anticorruption intensity on prefectural careers in the years 2008–12, as many Chinese officials were felled on corruption charges before the ongoing crackdown (Manion Reference Manion2004; Quade Reference Quade2007; Wedeman Reference Wedeman2005). We find no statistically or substantively significant relationship between purge intensity and appointments (or bias against appointments) of clients in crackdowns before the broad purge. Scale seems an essential condition for anti-client bias.
Our findings have important implications for the configuration of elite politics in China. The direct impact of the purge is reflected in the large number of its casualties. Our research measures an important indirect impact. To the extent that the purge has induced anti-client bias in appointments, it has consolidated the formal authority of the hierarchically organized CPC and undermined informal networks that challenge it.
Our concept of broad purge, prompted in part by reading and thinking about the historical extreme of Stalin's Great Purge, contributes to our understanding of purges under authoritarianism. Svolik (Reference Svolik2009; Svolik Reference Svolik2012) formulates the central problem of authoritarian governance as the conflict of interest between the dictator and a ruling coalition whose significant power poses a threat to them. Purges in this context are coup-proofing purges that the dictator launches against the inner circle. Yet, as we describe later, many authoritarian regimes are built on a broad coalition of organized followers—and many purges extend far below the top. Such broad purges trigger defensive action throughout the political system, such as risk reduction in decision making of all sorts.
Purges
In this article, we focus, conceptually, theoretically, and empirically, on purges in authoritarian regimes. We define a purge as the nonroutine removal of an official from power at the direction of a supreme political leader (or collection of leaders), invoking some standard that fits within regime norms.
Coup-Proofing Purges
The classic view of totalitarian dictators engaged in a constant struggle to survive the threat of regime overthrow by mass uprising (see, for example, Friedrich and Brzezinski Reference Friedrich and Brzezinski1965) turns out to be inaccurate. Svolik (Reference Svolik2009; Svolik Reference Svolik2012) examines all 316 authoritarian rulers who held office for at least one day and who lost power by any nonconstitutional means between 1945 and 2002. He finds mass public involvement in only sixty-eight such exits from power; by contrast, 205 dictators were ousted by insiders in the party, government, military, or security forces. Accordingly, he formulates the central problem of authoritarian governance as the conflict of interest between the dictator and his ruling coalition: “Although the dictator may be the most powerful member of the ruling coalition, he rules in the shadow of the threat of a coup” (Svolik Reference Svolik2012, 482).
This formulation has guided research on authoritarian purges since. That is to say, the obsession of dictators has to be “coup-proofing,” a term coined by Quinlivan (Reference Quinlivan1999). Not inappropriately, then, the literature on purges in authoritarian regimes focuses almost exclusively on coup-proofing purges of the inner circle, with a particular focus on purges to thwart threats by individuals “in key positions that have legitimate access to the use of armed force” (Sudduth Reference Sudduth2017, 1172).Footnote 2 As Roessler (Reference Roessler2011, 308–9) puts it: “The imminence, proximity, and secrecy of the threat, coupled with its incredibly high costs, have forced rulers to be on the defensive at all times.… Rulers pursue policies designed to safeguard their hold on power and neutralize the first-strike capabilities of those within their regimes.” Questions about why coup-proofing purges occur thus become questions about when they occur, as the fundamental insecurity of the dictator vis-à-vis his fellows and vice versa is a given.
Stalin's Great Purge
Not all authoritarian purges are directed at the inner circle of powerholders, however. Nor do all coup-proofing purges end with the inner circle. Stalin's monstrous Great Purge of 1937–38 is sui generis, but we consider some of its features in the following to situate the distinction between the less extreme broad purges that are coup-proofing purges and our focus.
Most crucial for us is the immense scale of the Great Purge. It not only eliminated the top echelon of the Communist Party, Red Army generals and officers, and ranking officials in the security forces, but also went far beyond these circles. It targeted rank-and-file party members—the political elite by the broadest definition. It also employed extreme violence to silence them, sanctioning deaths on an immense scale. That a preponderance of the more than 1.5 million arrested and more than 800,000 executed during 1937–38 (Kotkin Reference Kotkin2017, 305, 437) were linked in real networks of danger to Stalin himself is inconceivable.
It was a top-down purge, insisted on from the center. After the second public trial in Moscow in 1935, local party branches were enjoined to seek out and punish enemies among party members who had ever criticized Stalin or his ruling group. The decades-old routines of party purification to root out members who had proved themselves “unworthy” laid the groundwork for the extraordinary violence. Long-standing practices that had led to party punishment led during 1937–38 to “certain arrest and possible execution” (Kotkin Reference Kotkin2017, 439). Once the purge moved downward to target rank-and-file party members to break an “omnipresent counterrevolutionary plot,” it created an uncertainty that shook the entire political system. Party members were fearful not because they opposed Stalin or his ruling group, but because the executions of prominent revolutionaries suggested that no party member could be sure that he himself was up to standard (Halfin Reference Halfin2003). Actions that were not punished before 1930 became retroactively traitorous during 1937–38. Plausible culprits were many; yet, which party members would ultimately be arrested and executed was anyone's guess.
Broad Purges
A broad purge is a disruptive shock that reverberates throughout the political system in the form of grossly increased uncertainty about career security and personal safety, which creates fear. Examples of broad purges in modern history include Hitler's purge of the Nazi Party's paramilitary auxiliary, the Sturmabteilung (Campbell Reference Campbell1993). In Latvia, thousands were systematically purged from the Communist Party during 1959–63 (Loader Reference Loader2018). In China's Cultural Revolution, Mao Zedong set off a violent purge of the Communist Party in 1966 that provided the ideological pretext for factions at all levels to seize power in the name of his radical vision (MacFarquhar and Schoenhals Reference MacFarquhar and Schoenhals2006; Walder Reference Walder2015; Walder Reference Walder2019). In Vietnam, Le Duan orchestrated a purge of “North Firsters” in the 1960s, which expanded to a broad attack on anyone who had ever criticized party leaders after 1963 or whose family included anyone guilty of “bourgeois” sentiments (Nguyen Reference Nguyen2012). Other examples include a 1931 party purge in Fascist Italy (Morgan Reference Morgan2012) and a purge of the Syrian Ba'ath Party in the 1980s (Van Dam Reference Van Dam2011).
The defining feature for our distinction between coup-proofing and broad purges is the scale of broad purges: they extend below powerholders at the top to encompass a large number of plausible but indeterminate targets in the elite, broadly defined. The adoption of a coup-proofing paradigm leads us astray when most casualties of a broad purge are situated too far below the top to be a direct threat to the supreme leader's power. They cannot conceivably be linked to the safety of the supreme political leader. Indeed, the farther down the broad purge goes, the less basis there is to conceptualize it as a purge directed at enemies of the supreme leader.
In addition to their scale, we define broad purges by several other features. They are top-down. They are aimed at the elite itself, broadly defined. They are accompanied by greater enforcement effort and more severe punishment. They invoke familiar standards and build on routine procedures, but newly interpret the standards more strictly and apply the new interpretation retroactively, such that the number of plausible culprits increases immediately and immensely. In these circumstances, a broad purge gives urgent cause for many (perhaps most) to fear repercussions for conduct that by prior norms probably constituted business as usual. Moreover, as past failure to measure up becomes newly actionable, predicting actual targets from the great number of plausible targets is by no means obvious. In broad purges, it is difficult or impossible for any plausible target to reliably estimate their career security or personal safety.
In sum, for any member of the elite, who is by no means necessarily situated close to power holders at the top, the purge's large scale raises the likelihood of falling victim to it and the increase in enforcement strictness raises its expected personal cost. Yet, whether and when a plausible target becomes an actual target remains fundamentally unpredictable. We theorize that these features of broad purges create extreme uncertainty for the elite, expansively defined. Broad purges grossly raise both the likelihood and cost of past and present political error. We theorize that this encourages strategies of risk reduction in decision making of all sorts at all levels of power.
Xi Jinping's Broad Purge
Xi Jinping ascended to leadership of the CPC in November 2012. His corruption crackdown began in earnest in early 2013. Several features of the crackdown lead us to classify it as a broad purge.
First, the purge has vast reach, our key defining feature for a broad purge. The Central Discipline Inspection Commission (CDIC), the party's anticorruption agency, has investigated and punished some 1.5 million officials since the purge began (Gan and Choi Reference Gan and Choi2018). It has felled more officials and more high-ranking officials than any other crackdown in the party's history.
Secondly, the party has recentralized the anticorruption effort. Before 2013, local party bosses controlled their local party anticorruption departments, which gave them say over targets of corruption crackdowns and severity of punishment, but Xi's purge is strongly centralized in the CDIC in Beijing. In a structural redesign, the CDIC now exercises line leadership over local anticorruption departments, making it difficult for local party bosses to protect their clients or themselves, as was common practice (Manion Reference Manion2004; Manion Reference Manion2016). For local party bosses, there is a seeming randomness to the investigations: the CDIC's main investigatory form is roving inspection teams sent from Beijing to the localities, with timing and targets undisclosed until the last moment (Fu Reference Fu, Rose-Ackerman and Lagunes2015; Yeo Reference Yeo2016).
Another distinctive feature is the campaign's heightened rhetoric of Leninist party discipline, with corruption also defined organizationally, as a problem of factionalism. Wang Qishan, then CDIC chief, pronounced corruption and factionalism conjoined evils, with factionalism the most serious form of corruption (Wang Reference Wang2017). The anti-factionalist rhetoric distinguishes Xi's campaign from previous crackdowns. It clarifies for us, and may clarify for local party bosses, that the party center is particularly sensitive to the threat of informal networks, but our argument here is that signaling an anti-client bias is a safe strategy in a broad purge, with or without such rhetoric.
Finally, the recent broad purge includes various forms of punishment. As an initial step of investigation against corruption, party-led detention and interrogation often inspire fear, as law is “emphatically not tolerated as a hurdle” (Fu Reference Fu, Rose-Ackerman and Lagunes2015, 148). Hundreds (at least) of officials under investigation are estimated to have committed suicide (Wang Reference Wang2018). After the investigations, officials may face party punishments that target all sorts of commonplace “decadence,” such as overly elaborate celebrations, luxury purchases, and extravagant dining (Quah Reference Quah2015). In addition, the CDIC has reported that investigations exposed not only corruption, but also countless “collusive networks,” “gangs,” and “factions.” The last step is criminal sentencing. The party hands corrupt officials over to the courts for criminal punishment, including imprisonment, confiscation of property, and, in some cases, the death penalty. According to official statistics, of the 1.41 million officials investigated by the party between late 2012 to mid-2017, most punishments (1.40 million) were within the party, with the harshest punishment being expulsion from the party. Only 5,400 officials investigated received a criminal sentence.Footnote 3
In this article, we focus on purges of provincial-level CM officials who work in the inner circle around provincial bosses. The left panel of Figure 1 shows purges of CM officials felled by corruption charges from 2008 through 2017.Footnote 4 During 2008–12, before Xi's broad purge, only fourteen such officials lost office in corruption investigations. From 2013 through 2017, 112 such officials were investigated and punished. The right panel of Figure 1 shows that the spatial distribution of the purge extends across all China's thirty-one provinces. In addition to its greater intensity and geographical pervasiveness, dismissals of CM officials in the current crackdown lead to more serious penalties than those for lower-rank officials. Of 107 dismissed CM officials with punishment information available, ninety-six officials (90 per cent) were expelled from the party and eighty-seven officials (81 per cent) faced criminal charges. The criminal charges were severe: one official was sentenced to death with immediate execution; four obtained a death sentence with reprieve; and sixteen were sentenced to life imprisonment. Officials sentenced to a limited prison term were sentenced to thirteen years on average. Our point here is to show that the purge constituted a major shock, not only to these dismissed officials or the public, but also to incumbents who survived the purge. In the next section, we elaborate the decision making of these political survivors in response to this broad purge.
Figure 1. Time trend and spatial distribution of dismissed CM officials.
Political Selection in Broad Purges
Our analysis of political appointments fits into an existing literature on political selection in China. Different from the standard perspective in that literature, however, we view political selection as dynamic, being responsive to changing circumstances, especially circumstances that introduce political shocks for party bosses, who make the political selection decisions in China. Specifically, we theorize here that a broad purge environment induces risk reduction in political appointments and that risk reduction will be especially sensitive to local purge intensity.
Political Selection
CPC control over all careers of any importance from top to bottom is the linchpin of party rule in China. The hierarchically organized CPC monopolizes appointments, promotions, transfers, and dismissals of all officials of even moderate importance. The party center in Beijing, through its COD, directly manages a few thousand officials and delegates management of others to CPC committees, each dominated by a party secretary, below it. Tier by tier, these powerful local party bosses play the decisive role in appointments of officials one level down the political hierarchy (Burns Reference Burns1989; Landry, Lü, and Duan Reference Landry, Lü and Duan2018; Manion Reference Manion1985).Footnote 5 Provincial party secretaries—the bosses in our analysis—manage officials at the prefectural rank. What we analyze here as appointments are promotions of prefectural party secretaries to an office at the vice-governor rank.
The party center sets standards for career advancement (Edin Reference Edin2003; Whiting Reference Whiting, Naughton and Yang2004; Zuo Reference Zuo2015). Advancement is supposed to be largely based on performance, not personal connections. A large quantitative literature on political selection in China analyzes the actual relative weight of performance and connections on career advancement of officials. Findings do not all point in one direction, but rather vary by level of official, time period, measures, and models (see, for example, Jia, Kudamatsu, and Seim Reference Jia, Kudamatsu and Seim2015; Landry, Lü, and Duan Reference Landry, Lü and Duan2018; Li and Zhou Reference Li and Zhou2005; Maskin, Qian, and Xu Reference Maskin, Qian and Xu2000; Shih, Adolph, and Liu Reference Shih, Adolph and Liu2012). We theorize that political selection decision making (for example, appointments) may not follow a single static principle (that is, connections or performance). Instead, we expect party bosses to respond strategically (and probably differently) to changing circumstances, especially political shocks. Xi's broad purge is one such shock.Footnote 6
Anti-client Bias: Signal of Conformity to Party Organizational Discipline
In more normal times, provincial party bosses can signal loyalty to the party center with talk, for example, in exaggerated echoes of the supreme leader's ideological campaign in the local media they control (Shih Reference Shih2008). In the extreme political environment of Xi's broad purge, rhetorical displays are insufficient. With their power over political selection, however, local party bosses command a highly credible resource to signal organizational loyalty. Specifically, appointments offer them strategic opportunities to signal to the party center in Beijing that they are not building their own informal networks of patronage in their respective jurisdictions. Clients are the building blocks of informal networks. Advancing the careers of clients builds networks by strengthening the obligations of clients to bosses. We argue that an anti-client bias in appointment decision making is a safe strategy for party bosses to reduce risk in a broad purge like the current corruption crackdown.
An anti-client bias is a singularly appropriate signal because of the extraordinary sensitivity to factionalism that runs through Chinese communist politics. In CPC history, factional challenges to the authority of the party hierarchy have been an enduring concern (Huang Reference Huang2006). Chinese party rhetoric routinely warns local party leaders against building “independent kingdoms.” Actual examples of such regional threats to party hierarchy occurred as early as the 1950s (Teiwes Reference Teiwes1990). Organizationally, factionalism flouts the supreme Leninist principle of party discipline, which requires strict adherence to party hierarchy (Meyer Reference Meyer1957). Moreover, as noted earlier, rooting out factionalism is an explicitly proclaimed aim of Xi's broad purge.
In addition, the particular anti-client bias we study here is a highly legible signal of conformity to party organizational discipline. This is because promoted prefectural party secretaries ascend to a rank of officials directly managed by the party center, with the transfer of their files containing all career information to the COD.
Certainly, echoing Xi's broad purge, provincial party leaders can also use their political selection power to signal loyalty by actively following the center with their own purges of officials they manage. We study appointments (not dismissals) because we are interested in purge winners (not purge losers).Footnote 7 Purge winners survive to exercise influence on politics in the future. We are also particularly interested in the institutional implications of Xi's broad purge: an analysis of anti-client bias in appointments is more fruitful from this perspective too.
Purge Intensity
We theorize that variation in anti-client bias in political appointments reflects heterogeneity across provincial party bosses in the threat the purge presents to their careers. We analyze this threat as jurisdictional purge intensity. As we argued earlier, a broad purge creates grossly increased uncertainty for all: essentially, no official has a sense of career security or personal safety. For example, Lorentzen and Lu (Reference Lorentzen and Lu2018) show that connections to politburo members do not protect officials from Xi's broad purge.Footnote 8 As Xi's purge differs in major ways from previous corruption crackdowns, provincial party bosses cannot reliably gauge their vulnerability with information from the pre-purge years. Their own corruption and that of subordinates on their watch probably increases the threat of purge; however, we argue, they can best gauge their prospects for purge in the near future by paying attention to purge actions in the near past. We capture the threat with our measure of jurisdictional purge intensity, which we showed in Figure 1 earlier. Empirically, this is the number of CM officials recently purged in the province. Conceptually, jurisdictional purge intensity bundles together information about the various unknown predictors of purge. For provincial party bosses, it is a highly visible action by the CDIC that offers them individualized, up-to-date information about their own vulnerability to purge.
For party bosses, the felling of CM officials in their respective provinces implies (at least) their mismanagement of personnel matters. As the chief executives of their respective provinces, they assume responsibility for actions of all subordinate officials, which strongly implicates them in the quality of these officials. Moreover, other than small numbers of vice-governor appointments from outside the province, provincial party bosses manage the vetting, nomination, and appointment process that puts vice-governor-ranked officials in office, which implicates them in their quality too.Footnote 9 At worst, purge of CM officials in the province is a step in a typical CDIC investigatory encirclement that ends in the downfall of provincial leaders. For example, in early 2014, the CDIC purged two CM officials in Yunnan province, including four provincial party committee deputy secretaries. A few months later, the provincial boss of Yunnan, Bai Enpei, was dismissed for corruption. In Shaanxi, the CDIC purged three senior subordinates (CM officials) of provincial party boss Zhao Zhengyong when he served as the provincial party boss (2013–16). Zhao eventually was purged in 2019, and charges against him included not only corruption, but also clientelistic behaviors, such as building political gangs.Footnote 10
In the purge environment, we argue that party bosses take purge intensity in their respective provinces as a signal from the party center of their own vulnerability to purge. We formulate this argument as our main hypothesis:
Hypothesis 1 (H1): Experienced purge intensity in their respective provinces is directly associated with anti-client bias by provincial party bosses in appointment decisions.
Data and Measures
To test our main hypothesis, as well as other implications of our theory, we construct a prefecture-year panel dataset that includes political turnover for all 667 individuals who served as CPC secretaries of China's 333 prefectures at some point from 2013 through 2017. We focus on the promotion of prefectural party secretaries because these promotions are relatively effective signals by provincial party bosses to the party center. They are effective signals because these promotions must be approved by the COD of the CPC; promoted prefectural party secretaries will then come under COD management.Footnote 11
We obtain biographical information for prefectural party secretaries from: a database created and managed by Pierre Landry at the Chinese University of Hong Kong; the Chinese Political Elite Database developed by Jiang (Reference Jiang2018); Peking University's China Center for Economic Research Official Dataset (Yao and Xi Reference Yao and Xi2019); and Baidu Baike, a web-based Chinese language encyclopedia. We cross-check the biographical information from these sources and construct a comprehensive dataset of prefectural party secretaries. In addition, we add some basic prefectural-level administrative data obtained from China's City Statistical Yearbooks. We obtain data on purged officials from the Tigers and Flies Database developed by China File and China's Corruption Investigations Dataset developed by Wang (Reference Wang2020).
Dependent Variable: Political Turnover
The dependent variable for our baseline specification is political turnover. We code seven categories of turnover: no change, lateral transfer, promotion, retirement, dismissal for misconduct, demotion, and death. Table D.1 in the Online Appendix presents summary statistics for the 1,655 political turnovers experienced by the 667 prefectural party secretaries in our five-year period of study. By far the most common category (72.5 per cent) is no change. In descending order, lateral transfer is next most common (13.2 per cent), then promotion (8.3 per cent), then dismissal (2.8 per cent).
Independent Variables of Theoretical Interest
Our independent variable of key theoretical interest is patronage connection between a provincial party boss and party secretaries of prefectures in his province. Following Jiang (Reference Jiang2018), we define a patronage connection between provincial party boss A and prefectural party secretary B only if B was promoted to prefecture rank during the term of provincial party boss A. Simply put, this indebts B to A for his rank. We are particularly interested in how the broad purge affects the relative weights of patronage connection in a provincial party boss's promotion of prefectural party secretaries to a higher rank. As argued earlier, we expect that provincial party bosses take provincial intensity of the purge by the CDIC as information from Beijing about their own political vulnerability.
We focus on individual-level variation in purge intensity across provincial party bosses. Specifically, we measure purge intensity as the number of CM officials purged in the province since the provincial party boss assumed office. As noted earlier, careers of CM officials are directly controlled by the COD in Beijing. In the provinces, CM officials include provincial party committee members, governors, deputy governors, heads, and deputy heads of the provincial people's congresses. In our data, twenty-seven of forty-eight provincial party bosses experienced a purge of one or more CM officials between 2013 and 2017.
Confounding Covariates
We control for relative performance of prefectural party secretaries in our analyses. Specifically, following the quantitative literature on China's performance evaluation system, for each prefecture-year in our dataset, we construct two measures of prefectural relative performance: (1) economic performance, reflected in gross domestic product (GDP) growth rate; and (2) fiscal performance, reflected in fiscal revenue growth rate. We follow the literature in measuring these two types of performance by computing the deviation of a prefectural party secretary's performance from the average of peers in all other prefectures in the province.Footnote 12 In addition to prefectural-level covariates, we control for several individual-level covariates for prefectural party secretaries. We include five demographic variables: age and its squared term, ethnicity, gender, and college education. We also include seven variables reflecting work experience: years of work experience; years of party membership; and work experience in the discipline inspection system, courts and law enforcement, Communist Youth League, organization department, and propaganda department. To address the concern that connected prefectural party secretaries are less likely to be promoted if they were only recently promoted to prefectural rank by their provincial party bosses, we control for tenure of prefectural party secretaries by years in office. We also include the quadratic term of years in office, which is standard in the literature. Lastly, we control for provincial-level confounders. We include the term of provincial party bosses, measured by years in current office. Including this measure adjusts for the mechanical relationship between time in office and number of purges experienced. We also address the concern that forward-looking provincial party bosses might make risk-averse decisions in response to the ex ante risk of corruption dismissals. The CDIC conducted twelve rounds of inspections across all thirty-one provinces between 2013 and 2017.Footnote 13 According to the official inspection summary, roving inspections discovered evidence of corruption for over 60 per cent of CM officials.Footnote 14 To take into account the effect of ex ante risk reflected in roving inspection, we include a binary measure, coding all prefectures in a province under CDIC inspection as 1 and 0 otherwise.Footnote 15 Summary statistics are in Table E.1 in the Online Appendix.
Empirical Strategy
We study how Xi's broad purge affects political appointment decisions by provincial party bosses by analyzing prefecture-year panel data for 2013 through 2017. We use linear probability models that allow us to incorporate various fixed effects and compare our results with existing findings. The model is specified as follows:
$$\eqalign{{\rm Promotio}{\rm n}_{( i( jk) t + 1) } &= \beta _1\;{\rm Connectio}{\rm n}_{i( j) t} + \beta _2\;{\rm PurgeIntensit}{\rm y}_{kt} \cr & \quad + \beta _3\;{\rm PurgeIntensit}{\rm y}_{kt} \, \ast \, {\rm Connectio}{\rm n}_{i( j) t} + \beta _4\;X_{it} + \beta _5\;Z_{\,jt} \cr & \quad + \beta _6\;P_{kt} + \lambda _i + \gamma _t + {\rm \epsilon }_{ijt}, \;} $$ where Promotion(i(jk)t+1) is political turnover of party secretary j of prefecture i in year t + 1. Here, we use a forward term to guarantee that all purges and patronage connections counted in our independent variables occur before the provincial party boss makes the appointment decision. In the baseline specification, we use a dichotomous measure, coding promotion as 1 and other types of career changes (including no change) as 0. Connectioni(j)t is a dichotomous measure of patronage connection. We code it as 1 if party secretary j of prefecture i is connected to provincial party boss in year t and 0 otherwise. PurgeIntensitykt denotes the number of purged CM officials in the province ruled by provincial party boss k. We interact PurgeIntensitykt and Connectioni(j)t to produce the estimate, β 3, which we interpret as the effect of purge intensity on the weight of connections in promotion. X it is a set of prefectural-level time-variant covariates, specifically, relative economic and fiscal growth rate of prefecture i in year t. Z jt is a set of individual-level covariates: female; minority ethnicity; college education; years in office and its squared term; age and its squared term; years of working experience; years of party membership; and work experience in discipline inspection, courts and law enforcement, organization, and propaganda. P kt denotes provincial-level covariates, including the term of provincial party bosses and roving inspections. λ i and γ t capture prefecture and year fixed effects, respectively. 
$\epsilon _{ijt}$ is the error term. We cluster standard errors at the prefectural level to account for the serial correlation of observations in each prefectural unit.
Regression results for our test of H1 are reported in Table 1.Footnote 16 Before estimating the interaction model, we estimate the effect of patronage connections and performance on political selection, which can be compared to previous analyses. To do so, we first regress political turnover on our key independent variable, that is, connection, controlling for individual-level covariates, as well as prefecture and year fixed effects, in Column 1. The coefficient on connection is negative but statistically insignificant. In Column 2, we add relative economic and fiscal performance indicators into the model. The estimates of two performance indicators are small and statistically insignificant. The null effect of performance on political turnover is consistent with the mixed findings in the literature on political selection for prefectures in China (see, for example, Landry, Lü, and Duan Reference Landry, Lü and Duan2018). Column 3 includes connections, purge intensity, and their interaction. The estimate of the interaction term is negative and statistically significant at the 0.01 level. In Columns 4 and 5, we gradually add prefectural-level and provincial-level covariates in addition to the interaction term. All provincial-level and prefectural-level covariates are statistically insignificant. By contrast, our key variable of interest, that is, the interaction term, remains negative and significant at the 0.05 level.
Table 1. Effect of purge on political appointments of prefectural party secretaries
Notes: Dependent variable is political turnover (promotion = 1: otherwise = 0). FE = fixed effects. Standard errors clustered at prefectural level reported in parentheses. Individual-level controls are: female; college; years of work experience; years of party membership; age; age squared; years in office; years in office squared; and work experience in discipline inspection, courts and law enforcement, organization, and propaganda. * p < 0.1; ** p < 0.05; *** p < 0.01.
To visualize these results, we plot the marginal effect of connections on political turnover over different levels of purge intensity in Figure 2.Footnote 17 Figure 2 shows that the marginal effect of connections on promotion becomes negative when purge intensity increases to 1. In other words, our model predicts that the effect of connections on promotion becomes negative when more than one subordinate prefectural leader of provincial party bosses is purged. Shanxi is a case that illustrates the anti-client bias. In Shanxi, where nine CM officials were purged, neither of the two provincial party bosses, Wang Rulin and Luo Huining, granted promotion to any of their connected subordinates in the broad purge period.
Figure 2. Marginal effect.
Overall, the results suggest that the greater the number of purge casualties in their respective provinces and terms of office, the less likely are provincial party bosses to promote their clients. The conditional effect of purge intensity is not only statistically significant, but also sizable: our most conservative estimate suggests that the probability of client appointment decreases by 2.5 to 2.8 percentage points with one additional CM official purge. As the mean probability of promotion is 8.3 percentage points, the effect of purge intensity is considerable. We see this as clear support for our argument that provincial party bosses make promotion decisions that are biased against their clients in the broad purge environment. We interpret this as risk-reducing political selection decision making of the sort theorized earlier.
We further verify the robustness of our baseline specifications by conducting several sets of robustness tests. Our first concern is about the measurement of dependent and independent variables. In the Online Appendix, we show that our results are robust to an alternative coding scheme of political turnover (see Table F.1), ratio measure of purge intensity (see Table F.2), and alternative measures of political connections (see Table F.3). We are also concerned about the model specification. We show that our results on the anti-client bias hold when using provincial-level clustered (see Table F.4) and bootstrap standard errors (see Table F.5), as well as a probit model (see Table F.6). In addition to model specification, we show that our results are generally robust when conducting subsample analysis (see Table F.7). As our dependent variable, that is, political turnover, varies across different terms of prefectural party secretaries, we follow Landry, Lü, and Duan (Reference Landry, Lü and Duan2018) by collapsing the data to the term level. Table F.8 shows that the results remain consistent in the term-level analysis. Overall, none of the preceding analyses challenge our main findings. Detailed discussions of these tests are in Online Appendix F.
Addressing Alternative Explanations
Beyond demonstrating the robustness, we exclude three alternative explanations of the observed anti-client bias. First is the incompetence explanation that provincial party bosses may promote non-clients because clients are less competent or are corrupt. We test this conjecture by examining the effect of patronage connections on prefectural GDP growth, prefectural fiscal growth, and the corruption dismissal rate for prefectural party secretaries. Table F.9 In the Online Appendix presents the results. All estimates of connections are negative but statistically insignificant, suggesting that clients perform as well as non-clients and do not have a higher corruption dismissal rate than non-clients.
A second explanation is that promotions are eye-catching events that may attract the attention of the CDIC, leading to future corruption investigations. We empirically examine this “promotion curse” conjecture by taking two steps. First, we focus on how frequently clients and non-clients move. If a “promotion curse” does exist, a natural response of provincial party bosses is to keep a low profile for their clients, making fewer career changes for clients. An observable implication is that clients stay longer in office than non-clients. We test this conjecture by conducting a t-test comparing the years in office of clients and non-clients. Table F.10 in the Online Appendix shows the result. We show that clients spend less time in a post (0.7 years) than do non-clients, providing suggestive evidence against the promotion curse conjecture. As a second step, we directly examine the linkage between promotion and future corruption dismissal by conducting a term-level analysis. We construct a binary measure of future dismissal that is coded as 1 if the prefectural leader is purged in the following three years and 0 otherwise. We regress this future dismissal measure on our political turnover measure, which codes promotion as 1 if one is granted promotion after the term, controlling for individual, prefectural, and provincial controls. Table F.11 in the Online Appendix shows the result. In contrast to the speculation of the promotion curse conjecture, we show that promoted prefectural leaders are less likely to be dismissed for corruption than their peers who do not experience career advancement. In summary, we present compelling evidence to exclude two alternative explanations for the observed anti-client bias.
Last but not least, the anti-client bias might be driven by career rotation patterns. In China, political leaders are regularly rotated across different posts in their career (Guo Reference Guo2009; Landry, Lü, and Duan Reference Landry, Lü and Duan2018). Career rotation might be a confounder in our study because it affects both promotion and connections. For example, party bosses who have newly assumed office in a province may lack connections with any prefectural subordinates in the province. Moreover, they may also lack a timely opportunity to promote anyone. While controlling for the provincial party boss's term in the baseline specification, we further address this rotation effect by adding a binary indicator of whether a provincial party boss has prior work experience in the province. Column 1 of Table F.12 in the Online Appendix shows results similar to that in our baseline results. We also conduct a subsample analysis, estimating the anti-client bias for cities ruled by provincial party bosses with and without prior working experience in the province. Columns 2 and 3 of Table F.12 in the Online Appendix show the results. The estimate of Purge Intensity × Connection is −0.045 and statistically significant at the 5 per cent level in cities ruled by provincial leaders with experience in the province. In the sample of cities ruled by party bosses without prior local experience (see Column 3), the estimate of the interaction term is smaller but still statistically significant at the 10 per cent level. This suggests that anti-client bias exists in both samples. To further demonstrate that our key findings do not result from the rotation of provincial leaders, we replace provincial-level covariates with provincial leader fixed effects to explore within-leader variation in connections and promotions. As shown in Column 4, the within-leader estimation is still negative and statistically significant. In sum, the evidence in these further analyses suggests that our baseline findings are not driven by rotation of provincial party leaders.
Key Extensions
We conduct two key extensions. First, we investigate how (if at all) Xi's clients differ from their non-client peers in their appointment decisions. We present suggestive evidence that anti-client bias is not less evident among Xi's clients compared to other provincial bosses. Secondly, we conduct a placebo test, showing that purge of CM officials in each province is not associated with anti-client bias before 2013, essentially because purge intensity of previous corruption crackdowns lacks the large scale of Xi's broad purge. Overall, no extensions in this section challenge the main results. We are, therefore, all the more reassured about the support for our theory in our baseline specifications.
Anti-client Bias among Xi's Clients
As we noted in our introduction, scholars have analyzed Xi's broad purge as a factional struggle, examining Xi's use of the corruption crackdown to sweep away potential enemies and strengthen his own faction in the party by appointing loyal clients to important posts (see, for example, Fu Reference Fu, Rose-Ackerman and Lagunes2015; Lorentzen and Lu Reference Lorentzen and Lu2018; Murong Reference Murong2015). An implication relevant to our study is that the anti-factionalist rhetoric of Xi's broad purge is not to promote party organizational cohesion per se, but rather aimed at leaving only one faction standing: Xi's own. Xi's provincial clients can extend Xi's reach in their respective provinces by promoting their own clients. This line of argument suggests we should find weaker anti-client bias among Xi's provincial clients than among peers who are not Xi's clients.
Our theorized effect of purge on political selection presents a different view and different empirical expectations of Xi's clients. We argue that a broad purge creates extreme uncertainty for the elite, encouraging strategies of risk reduction. Specifically, even politically connected officials cannot reliably estimate their career security or personal safety. Among provincial party bosses, Xi's clients are the most connected of all, however—and Xi's clients have not been purged. At the same time, in the fast-changing environment of the broad purge, building a local network of loyal clients is not without risk, especially given the anti-factional rhetoric of Xi's anti-corruption campaign. An anti-client bias is still a safe strategy because it clearly conforms to the organizational discipline monitored by the CDIC. In sum, we see an empirical investigation of anti-client bias among Xi's clients as a strict test of our theory of risk reduction in political selection in the broad purge environment.Footnote 18
To test this implication of our theory, we analyze how (if at all) Xi's clients differ from their non-client peers in their appointment decisions. For consistency, we use the same measure of patronage connection as in our main models, borrowed from Jiang (Reference Jiang2018), that is, we define provincial party bosses as Xi's clients if they were initially promoted to provincial party boss under Xi's party leadership and were not purged under his leadership. By this definition, twenty-six of forty-eight provincial party bosses in our dataset are Xi's clients. Table H.1 in the Online Appendix lists them and the provinces they manage.
We estimate the heterogeneity of political selection decision making by Xi's clients and their peers with a triple interaction analysis, controlling for individual, prefectural, and provincial covariates. We visualize the marginal effect of connections on political turnover by factional connection with Xi in Figure 3.Footnote 19 It shows anti-client bias among both Xi's clients and non-clients. Although the estimate for Xi's clients has a steeper slope than that for non-clients, the 95 per cent confidence interval of marginal effects for the two groups overlaps over the full range.Footnote 20 As purge intensity is smaller for Xi's clients than for non-clients (with mean CM officials purged equal to 0.88 and 1.4, respectively), it is more meaningful to test the difference in anti-client bias between the two groups at the same level of purge intensity. We estimate the difference between Xi's clients and non-clients in the weight of connections in promotion at different levels of purge intensity. Table H.3 in the online Appendix shows the estimation of marginal difference. We find the difference is not statistically significant at the 5 per cent level at any level of purge intensity.Footnote 21 To sum up, all evidence combined, we find that the level of anti-client bias is not significantly different between Xi's clients and non-clients.
Figure 3. Conditional effect of Xi's faction.
These findings support the salience of the anti-client bias. Moreover, they have institutional implications for a distinction between clientelistic connections and organizational loyalty in Xi's broad purge. Connections with Xi are more like an endowment; organizational loyalty is a decision. Xi's clients have connections with the supreme leader, but in their political selection decisions, they nonetheless demonstrate organizational loyalty.
Pre-broad Purge Period
Our theory that the broad purge environment encourages risk reduction in political appointments can be further tested in a low-intensity corruption crackdown. As we theorized earlier, the scale and other features of Xi's crackdown create a distinctive purge environment in which the safe response for party bosses is to promote subordinates unconnected to them. As purge intensity is the key moderator that affects the strength and direction of connections in appointments, we expect party bosses to adopt a less risk-reducing strategy in an environment with fewer officials purged. To empirically examine this key conjecture, we analyze how the number of CM officials felled on corruption charges affects political appointment decision making during 2008–12, before Xi's broad purge. In these years, allegations of corruption against CM officials were less common: only fourteen CM officials lost office as a result of a corruption investigation. Using the same specification as our baseline model, we estimate the effect of these smaller purges on anti-client bias by provincial party bosses.
Results are in Table 2. Following our baseline specification, we first regress political turnover on patronage connections and performance indicators in Columns 1 and 2. By contrast to the result for the purge period, the coefficients on connections are positive but statistically insignificant in the pre-purge period. The coefficients on fiscal and economic performance are close to 0, which is consistent with previous research that suggests a weak effect of performance on political selection among Chinese prefectures (see, for example, Landry, Lü, and Duan Reference Landry, Lü and Duan2018). Turning to the interaction analysis in Columns 3 to 5, in all specifications, the estimate of the interaction of connection and purge intensity is close to 0 and statistically insignificant. These results suggest that anti-client bias is not a relevant political appointment response in an environment where losing office for alleged corruption is rare. In addition to this analysis for the pre-purge period, we also conduct a marginal-effect estimation using the entire 2008–17 period.Footnote 22 We use the kernel-estimation strategy developed by Hainmueller, Mummolo, and Xu (Reference Hainmueller, Mummolo and Xu2019), which allows flexible estimation of the interaction Purge Intensity × Connection in different years. Figure H.1 in the Online Appendix shows the result, which is generally consistent with our theoretical prediction. The marginal effect of the interaction term is close to 0 and statistically insignificant in the years before the campaign starts and begins to decline to negative values from 2013. We acknowledge that the confidence interval is large because of limited variation in purge intensity to compute the marginal effect of connections on promotion in each year, especially before and in the early years of the campaign. However, we are more interested in the trend that reflects the dynamics of officials' risk aversion in political selection. Overall, we see the full-sample analysis as additional support for our theory of exacerbated risk reduction in the broad purge environment.
Table 2. Effect of purge on political selection in pre-campaign period
Notes: Dependent variable is political turnover (promotion = 1; otherwise = 0). FE = fixed effects. Standard errors clustered at prefectural level are reported in parentheses. Individual-level controls are: female; college; years of work experience; years of party membership; age; age squared; years in office; years in office squared; and work experience in discipline inspection, courts and law enforcement, organization, and propaganda. * p < 0.1; ** p < 0.05; *** p < 0.01.
Conclusion
This article contributes to our knowledge about the configuration of elite politics in contemporary China. Chinese elite politics has long been studied as factionalized (see, for example, Nathan Reference Nathan1973; Pye Reference Pye1981). The literature on corruption in China mostly argues that crackdowns consolidate factions (see, for example, Pei Reference Pei2016). Chinese factions are studied not as clusters of common policy preferences, but as informal networks built on ties that carry a presumption of personal loyalty. We show that Xi's broad purge has weakened informal networks in the sense that party bosses reduce their risk of purge by not further promoting their clients.
The anti-client bias we theorize and document here partially resembles the “connection penalty” in Fisman et al.'s (Reference Fisman2020) analysis of China's politburo. The differences are important, however. First, the extreme political environment is crucial to the story we tell here: the circumstances of broad purge make all players insecure; and their perceived risk of purge drives political selection decision making. Secondly, instead of focusing on selection of leaders at the party center, we investigate how political leaders in the localities use their selection decision-making power as a signal to demonstrate loyalty to the party center. The third difference has to do with how ties that carry a presumption of personal loyalty may originate. Fisman et al. (Reference Fisman2020) focus on social ties established through common hometown origins and overlap in college experience. These social ties are static endowments, which individuals cannot change through choices they make after joining the government. We adopt Jiang's (Reference Jiang2018) notion, focusing on patronage connections that originate in political selection decisions. These connections are dynamic, changing over time with the political turnover of both clients and patrons.
To be sure, if patrons build networks of clients through promotions, then the provincial party bosses we study here are, by definition, regularly building networks because, purge or no purge, they must promote some officials. They are expanding them with new clients, rather than deepening them by rewarding existing clients. We could speculate that these new ties may not be a solid-enough foundation for factional loyalty. We prefer to think of it somewhat differently. We focus in this article on anti-client bias as a signal to Beijing of conformity to organizational discipline over faction building in political selection. It is not only a signal to Beijing, however. It is at the same time a signal to existing and prospective clients: party bosses who abandon their clients at promotion time are ungenerous or unreliable patrons. The decline of factions we theorize and document here is a counterintuitive effect of substantive importance for Chinese politics, which should be revisited analytically in the post-purge era.
Supplementary Material
Online appendices are available at: https://doi.org/10.1017/S000712342200062X
Data Availability Statement
Replication data for this article can be found at: https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/ZN9U5I
Acknowledgments
We thank BJPS editor René Lindstädt and three anonymous referees for their helpful comments. We thank Pierre Landry and Tianyang Xi for generously sharing with us their biographical data on Chinese local chief executives. We thank Linda Zhang, Cecilia Wu, and William Tong for research assistance. We thank discussants and other participants in presentations and workshops at the American Political Science Association 2018 Annual Meeting, Duke University, Harvard University, and Peking University. We are especially grateful to Haohan Chen, Xiaoshu Gui, Junyan Jiang, Ning Leng, Peng Peng, Victor Shih, Ezra Vogel, Yuhua Wang, David Weimer, Xiaomeng Wu, and Tianyang Xi for helpful comments on earlier drafts.
Financial Support
None.
Competing Interests
None.
Open access
