Social Ties and the Selection of China's Political Elite

Transcription

1 Social Ties and the Selection of China's Political Elite Raymond Fisman, Jing Shi, Yongxiang Wang, and Weixing Wu Abstract We examine the role of social ties in the selection of China's Politburo, the country's highest governing body. While there is no correlation between social connections (as measured by hometown or college ties) and selection probability in an unconditional regression, in analyses that include xed eects to absorb quality dierences across groups we nd that hometown and college ties are each associated with 6-10 percentage point reductions in selection probability. The connections penalty is stronger under Mao, suggesting that anti-factionalist sentiment plays a role, and weaker for ties to Standing Committee members, consistent with intra-factional competition. JEL classication: D72; P26 Keywords: Social Networks, Political Connections, Political Elite, Politburo, China 1

2 1 Introduction We study the selection of ocials into the Central Politburo (hereafter Politburo), the most powerful body in the Chinese government. Beyond the direct importance of understanding what determines the top leadership of the world's most populous nation (and second largest economy), studying elite political selection in China may provide insights into the selection of high-level ocials in non-democracies more generally. The Politburo's members are selected every ve years from among the members of the Central Committee of Chinese Communist Party (hereafter the Central Committee), whose membership in turn is drawn from the top ranks of provincial ocers, top military leaders, and central government ministers. While the Central Committee is nominally responsible for electing the Politburo (much as individual citizens are nominally responsible for electing Chinese ocials at lower levels), in practice the Politburo itself is thought to have a decisive role in selecting new members (see, for example, Li (2008)). In our paper, we examine whether Central Committee members with connections to incumbent Politburo members are more likely to be elected to the next Politburo, using data from the post-war period. There is, ex ante, reason to expect that such ties may provide a leg up in the Politburo selection process. For example, in writing about selection of the 17th Politburo, Shirk (2012) observes that it was commonly perceived that Politburo selection, revolve[s] around the distribution of seats among personalistic factions the networks of loyalty between senior political gures and the ocials who have worked with them, are from the same region or studied at the same university and who have risen through the ranks with their patrons. (Such connections may also lead to higher selection rates because social ties facilitate the transmission of soft information on candidate quality; see, for example, Fisman et al. (2017).) We focus on several forms of connections, alluded to in the preceding quote, that have wellestablished precedence in earlier work: hometown (i.e., prefecture) ties, college ties, and past employment relationships. 1 We begin with specications that parallel those employed in earlier papers that study the relationship between social ties and promotion among low-tier government ocials in China, which include controls for attributes such as education and age. In these initial analyses, we nd no correlation between connections and Politburo selection for hometown and college ties, and a strong positive correlation for workplace ties. These ndings plausibly suer from an upward bias in the 1 Recent studies that examine the benets of these types of connections in China include Cai (2014), Heidenheimer and Johnston (2011), Shih et al. (2012), Jia et al. (2015), Wang (2016), and Shih and Lee (2017) who explore their role in promotions in the Chinese bureaucracy, and Fisman et al., (forthcoming) who study their role in election to the Chinese Academies of Science and Engineering. 2

3 eect of connections, however, as there may be quality dierences across groups that are not captured by our controls. For example, by far the most commonly represented college among Politburo members is Tsinghua University, which is also China's most prestigious school. Simply controlling for higher educational attainment does not account for the dierence between Tsinghua versus lower-tier institutions. 2 (Similarly, more able ocials may follow common trajectories up through the bureaucracy, leading to a spurious positive correlation between workplace overlap and Politburo selection.) We thus introduce xed eects for hometown, college, and work history. In these specications, which control for quality dierences across groups, we nd that both hometown and college ties are associated with a lower probability of Politburo selection. For hometown ties, in our favored specication which includes hometown xed eects and a range of individual controls a Politburo connection reduces the likelihood that a Central Committee member is elected by 6.1 percentage points, a 52 percent decline relative to the baseline election rate. For college ties, the comparable gure is an 8.9 percentage point reduction in election probability. Once we account for workplace xed eects, we observe no detectable relationship between workplace ties and Politburo selection. We next explore possible reasons for the connections penalty in Politburo selection: (1) antifactionalism; (2) competition within and between factions; (3) quotas; and (4) a mechanical explanation resulting from the removal of better-qualied ocials from a group of candidates when one is elected to the Politburo. We begin with the readiest explanation for the connections penalty, one which turns the more standard favoritism intuition on its head: given concerns of favoring one's own group, Communist Party leaders have, since the revolution, inveighed against the dangers of factionalism, and might have used Politburo selection as a highly visible means of setting an example. The anti-factionalist rhetoretic was perhaps most fervent under China's post-war leader, Mao Zedong, who argued that it was harmful to both the collective and the individual if one chose to support another simply because he is an old acquaintance, a fellow townsman, a schoolmate, a close friend, a loved one, an old colleague or old subordinate. 3 In addition to its prominence in Mao's rhetoric, anti-factionalism was written into the Communist Party's constitution during the Seventh National Congress in June 11, When we divide the sample into the Mao and post-mao eras, we nd that the negative relationship between hometown connections and selection is signicantly stronger under 2 To draw a comparison to the U.S. setting, many law schools are represented among the judges on the various circuit and state supreme courts, yet only Harvard and Yale Law Schools are represented on the U.S. Supreme Court. One would not wish to conclude that appointments to the country's top court are the result of connections indeed, incumbent justices have no role in selecting new members. 3 From The Collected Works of Mao Zedong, Volume II, translation obtained from 4 See in particular the General Principles, and also Article 23 of Section 1. 3

4 Mao, consistent with the anti-factionalist rhetoric in the earlier part of our sample being at least partly responsible for the overall connections penalty we observe (the college tie eect also stronger under Mao, but the dierence between the two periods is small and insignicant). There are two distinct aspects of factional competition, each of which might lead to an observed connections penalty. First, as suggested by earlier research on factional power sharing (Shih et al. (2012)), if incumbent Politburo members wish to ensure that no single faction grows too strong, opposing coalitions may block powerful hometown or college groups from gaining additional members (intergroup competition). We do not, however, nd any dierential eect of connections for the hometown or college with the largest representation among incumbent Politburo members. Second, as emphasized by Francois et al. (2016), incumbent Politburo members may wish to maintain their dominant position within a group network (in our setting, the hometown or college network) and thereby block potential challengers from within their own group (intragroup competition). This explanation would apply less to Politburo Standing Committee (henceforth PSC) members, who sit higher up in the hierarchy than other Politburo members. Consistent with intragroup competition, the negative eect of hometown ties comes only from connections to non-psc members (we observe a more negative eect of college ties to non-psc members relative to PSC members, though the dierence between the two is small in magnitude and statistically insignicant). We next explore whether the negative correlation between connections and selection results in part from quotas for the Politburo, which aim to ensure that no group benets from disproportionate representation. However, we nd no eect of ties based on birth province, which would be the most natural demarcation for geography-based quotas, since the central government primarily controls province-level (rather than prefectural) appointments and hence has greater need to show impartiality at the province level. 5 (Quotas are less plausibly responsible for the negative college tie eect for which, to our knowledge, there has never been any discussion of quota allocation, even informally.) Finally, we consider whether our xed-eects result could be generated mechanically as a result of quality-based selection. This may occur if, when a candidate from a given group is elected, he leaves behind a pool of lower-quality candidates. These (low-quality) leftover candidates will be connected in the following term, thus leading to a negative correlation between connections and selection. While we cannot fully rule out this possibility, we argue that it is unlikely to be the main factor explaining our results, since we nd a connections penalty exists even for the subset of rst-time Central Committee members, which alleviates the concern that leftover candidates are of lower quality. 5 In the Chinese administrative hierarchy, provinces are under the direct control of the central government, while municipalities are under the direct control of provincial leaders. See, for example, Chung (1995). 4

5 Overall, our results indicate that, at least for the highest and most visible levels of the Chinese polity, connections may reduce the chances of promotion. Collectively, these results stand in contrast to the positive role of connections documented in earlier work (see Cai (2014) for a book-length treatment), which focused largely on promotion at lower levels of the Chinese bureaucracy, indicating that dierent possibly more stringent standards govern selection at the upper echelons of government. We contribute to the literature that aims to understand the selection of ocials in China specifically, and in non-democracies generally. In prior work on China, evidence on the role of favoritism has been mixed. Most recently, Francois et al. (2016) study the benets of factional ties for promotion within the Communist Party. They nd a benet from factional connections to the General Secretary, the highest-ranking position in the Chinese government, but also a tendency for factional power sharing at lower levels. This work builds on related earlier work by Shih et al. (2012), who similarly nd that individuals with factional ties to top ocials via hometown, college, or workplace ties have an advantage in promotion. We view our work as complementary to these earlier studies which, owing to their data and emphasis, tend to focus more on lower-level promotions (in particular from the rank of Alternative Central Committee to Central Committee membership) rather than accession to the Politburo. This dierence, along with our more inclusive measure of connections to any Politburo member (rather than the highest-ranked individuals), may account for our apparently contradictory ndings. Furthermore, as will become apparent in our analysis, accounting for hometown and college xed eects is crucial to examining the eects of (time-varying) connections, since some hometowns and educational institutions tend to produce many more high-ranked ocials than others. We view this additional aspect to our empirical analysis to represent an advance on prior work, enabled by the richness of our connections data and the length of our panel. Our work also links to a larger body of research on the determinants and consequences of promotional structures throughout the Chinese hierarchy. Jia et al. (2015), for example, report a complementary eect of connections and performance in determining provincial leaders' promotions, 6 while Persson and Zhuravskaya (2015) explore the role of promotions and thus career concerns in governing the policy choices of provincial leaders (Kung (2014), in his analysis of grain distribution during the Great Famine, shows in particular how such promotional concerns can misre). Our work also contributes to our understanding of the role of connections in China more broadly, linking to the vast literature on guanxi ties (for recent empirical examples see Fisman et al., (forthcoming) on the role of connections in election to the Chinese Academies of Science and Engineering, and Kung and Ma (2016) on the value of connections for small business growth). 6 We do not observe any eect of performance whether directly or conditional on connections in our own data, but provincial leaders represent only about a fth of our sample. 5

6 2 Background and data 2.1 The organization of the Chinese polity The Central Committee of the Communist Party of China (Central Committee) is a political body that comprises the top leaders of the Communist Party. Its members are elected at the convening of the National Congress of the Communist Party of China, under the guidance of the Politburo. 7 While the number of Central Committee members uctuates from term to term (and has grown over time), it has had approximately 200 members in each term since the early 1970s. The Central Committee's membership is comprised of national leaders, chief ocers at institutions that are under the direct control of the Central Committee (e.g., the Organization Department and the Propaganda Department), heads of ministries under the control of the State Council (China's chief administrative body), provincial governors and party secretaries, chief military ocers, and leaders from eight People's Organizations (e.g., the All-China Federation of Trade Unions and the Communist Youth League) who also hold the rank of minister. The Central Committee meets at least annually, to discuss and rene formal government policies. The de facto leadership of the government resides within the Politburo, a collection of approximately 25 top leaders selected from the membership of the Central Committee at its rst convening, which takes place immediately following the National Congress. In most terms, a small number of replacement members were also elected during later Central Committee meetings to replace Politburo members lost to death, removed due to corruption, or purged for political reasons (especially during the Cultural Revolution). 8 Other than the twelfth term ( ), during which ten members retired and were replaced by six new members, the number of mid-term replacements is generally very small. Throughout, we will include all Politburo members selected at any point during a term as new members, and will code their connections based on the composition of the Politburo at the time of selection. While the Central Committee is nominally responsible for electing the Politburo, according to Li (2008), the notion that the Central Committee elects the Politburo is something of a ction. (In fact, the Politburo selection follows a single candidate election rule, whereby there are no more candidates than available seats, so once the slate is chosen, the Central Committee acts as a rubber stamp.) Li goes on to suggest that Politburo members themselves, particularly those on the Standing Committee, hold sway in the selection process. 7 Starting with the Central Committee's eleventh term, which began in 1977, the National Congress has been held every ve years. Prior to that, the Congress was held at less regular intervals. 8 It may be argued that Politburo members who die while in oce may still inuence the selection of their successors. There are 15 such cases in our data; our results are virtually unchanged if we assume that candidates who share a hometown, college, or workplace with recently deceased Politburto members are connected. 6

7 The power of the Politburo manifests itself in a number of ways. Most importantly, higher levels of government generally direct the selection of candidates at lower levels, giving the Politburo control over the choice of provincial leaders. Additionally, given the relative infrequency of Central Committee meetings, the direction of China's central government is largely left up the Politburo and in particular members of its Standing Committee, who are selected from among the Politburo's membership. 2.2 The role of social ties in Chinese politics and society Guanxi is the network-based system of favor exchange that plays a central role in political, economic, and social life in China (Chen and Chao (2004)). Often, guanxi is built on social connections, including hometown ties, work ties, college ties, family ties, and mentor-mentee relationships. We focus on hometown, college, and workplace ties both because of their prominence in establishing guanxi and also because they are observable albeit subject to measurement error based on readily obtainable biographical information. Perhaps the most widespread source of guanxi is hometown connections. As expressed by anthropologist Leo Douw in his introduction to a book-length treatment of the topic, [t]he cultivation of hometown ties is part and parcel of the Chinese culture of establishing guanxi, or relationships of mutual obligation between individuals, and is therefore also an inherent part of the social structure in which doing business in China is embedded at present. Moreover, ethnic Chinese communities abroad have usually preserved a distinctly Chinese cultural identity which is centered on the sharing of roots in the hometown (Douw (2013)). Social organizations based on place of origin are very common, and are used to facilitate communication, strengthen within-group networks, enlarge the group's political power in the new location, and also to form coalitions to better compete in commercial enterprises. 9 Scholars of guanxi have also emphasized the role of college ties as a source of connections. The relationship does not require that two individuals overlap in their time in college (see, for example, Yang (2001)), much as research using college ties in the U.S. as a proxy for social ties does not emphasize concurrent enrollment for a connection to exist (see, in particular, Cohen et al. (2010) and Cohen and Malloy (2014)). Finally, we follow prior research in using workplace ties as a proxy for the social connections that may facilitate guanxi. In this case, we do require that two individuals overlap in their period of employment in order to qualify as connected (the approach taken by earlier researchers as well). In 9 See, for example, Fisman et al., (forthcoming) for an empirical analysis on the importance of hometown ties in election to the Chinese Academies of Science and Engineering, as well as further references on the formation and function of hometown ties. 7

8 part, this reects the fact that there are some very common career trajectories for those selected as Central Committee members (and hence eligible for Politburo selection) even with our requirement that Politburo candidates overlap in their employment histories, well over half of candidates are coded as workplace-connected; if no temporal overlap were required, the gure would rise to above 90 percent Social ties and the connections penalty While favor exchange under guanxi most intuitively predicts a positive association betwen Politburo ties and the election of new Politburo members, there are several plausible reasons that one may observe the opposite pattern; in the analysis that follows we will provide suggestive evidence to adjudicate amongst these various explanations. The rst explanation stems from the backlash against guanxi, associated most prominently with Mao Zedong, who inveighed against factionalism as a threat to the Chinese Communist Party and to Chinese society more broadly. Mao's anti-factionalist writings and rhetoric date at least to the late 1920s, when he led a regiment of the Red Army. For example, in a 1929 article Mao mentions Xiao Tuanti Zhuyi (small group mentality) as a form of individualism that was corrosive to the interests of the collective. 11 Mao's strong anti-factionalist stance set the tone for future leaders. Mao's successor, Deng Xiaoping, carried the torch of anti-factionalism forward, vociferously denying that he or Mao were ever associated with any faction and, like Mao, Deng spoke out against factions as impediments to party unity. 12 Concerns of in-group favoritism led the government to impose rules, dating back to at least the early 1990s, that aim to prevent local ocials from favoring those from their home regions. Given Mao's forceful anti-factionalist rhetoric and perhaps the resulting desire to set an example (despite the absence of any formal restrictions on Politburo selection), connections may plausibly have been a liability rather than an advantage in Politburo selection. A connections penalty may also arise from competition both across and within factions. The role of intergroup competition is suggested by earlier research on factional power sharing (e.g., Shih et al. (2012)), which argues that dierent factions within the Chinese polity aim to balance the power held by individual groups or networks. Interfactional competition could motivate coalitions of incumbent 10 While we follow earlier work in utilizing workplace ties to measure connections, as Francois et al. (2016) emphasize, it may be a problematic to do so because of the endogeneity of promotion of faction members through the hierarchy. We view this particular concern as less likely to apply to college ties and especially hometown ties. 11 See On correcting mistaken ideas in the party, contained in The Collected Works of Mao Zedong, Volume I. Mao continued to speak against factionalism through the 1930s. See, for example, the speech we quote in the introduction, which was made in See, for example, Deng's 1989 speech, We must form a promising collective leadership that will carry out reform, reprinted in The Collected Works of Deng Xiaoping, Volume III. 8

9 Politburo members to block the candidacies of already-powerful hometown or college groups to keep them from gaining yet more Politburo members. This would lead to a negative association between connections and selection. Norms of favor exchange may also, somewhat counterintuitively, lead to discrimination against cofactionals by incumbent Politburo members who may block potential challengers from within their own network in order to maintain their position of domaninance (as model by Francois et al. (2016)). Such competition within factions (intra-factional competition) would also lead to a negative correlation between connections and selection. A variant on intergroup competition is that there may exist informal quotas to ensure that no single group benets from disproportionate inuence. The government's top leadership may thus avoid selecting candidates from regions already represented in the Politburo, leading to to a connections penalty. Finally, selection on the highest-quality member within a group may mechanically lead to the appearance of a connections penalty. Intuitively, this is because a group becomes connected at t + 1 precisely because a well-qualied candidate from the group at time t was elected, leading to the creation of a connection. At t + 1, the average quality of remaining candidates (and thus selection probability) will be lower as a result. In contrast to the preceding theories predicting a negative correlation between social ties and selection, this selecting out explanation does not imply that a given candidate is less likely to be selected if he is connected rather, the negative correlation is the result of a decline in candidate quality that is associated with the formation of a connection. We will be particularly attentive to assessing the plausibility of the selection out hypothesis in the next section. In this section we have aimed to explain why, ex ante, the overall relationship between social ties and selection is ambiguous. On the one hand, the central government itself has acknowledged the continued role of social ties as a source of favoritism (indeed despite his vehement denials, Deng himself was accused by some as being aligned with a Mao faction, a point we return to below). At the same time, the need to make a show of anti-factionalism, as well as competition both within and between factions, provide a counterweight to the eects of favoritism. The aim of our data analysis is rst and foremost to document that the negative eect dominates. We also endeavor to provide further evidence albeit suggestive on which of the above factors are the main drivers of the overall negative relationship between connections and Politburo selection. 2.4 Data Our analysis requires background information on the full set of Central Committee members (including the small subset that are Politburo members). Our starting point for developing this database 9

10 is the website maintained by the Communist Party of China, which includes Central Committee lists going back to its seventh term ( ). 13 Background information on these individuals including place of birth, year of birth, and detailed education and work history may be found via the Political Elites of the Communist Part of China database maintained by the National Chengchi University in Taiwan. Only a few candidates from the ninth and tenth term election cycles ( and ) are not contained in the database, since they are not minister-level ocials. They are instead lower-level ocials elected to the Central Committee during the Cultural Revolution who, by virtue of their celebrity status as working class heroes, are easily tracked down via Baidu Baike, the Chinese equivalent of Google. Our outcome measure throughout is Elected it, an indicator variable denoting that candidate i was selected for term t of the Politburo. As noted in Section 2.1, while almost all new Politburo members are selected at the Central Committee's rst meeting, replacement members may also be chosen at mid-term meetings. We set Elected it = 1 for all individuals elected during term t regardless of when during the term they are selected. While Politburo members at term t 1 are eligible for membership also at term t, we omit them from our analysis, as they are generally reelected unless of retirement age. We also use these data to generate measures of connectedness of Central Committee members (who comprise the full set of eligible Politburo candidates) to incumbent Politburo members. Consider rst our measure of connectedness based on hometown ties. The candidates for term t of the Politburo include members of the Central Committee during term t 1. We dene candidate i for Politburo term t to be hometown-connected (HometownT ie = 1) if there exists at least one Politburo member at term t 1 (and hence in the Politburo when selection of the term t Politburo takes place) who is from the same prefecture as i. HometownT ie can be measured from the eighth term ( ) onward, since we require lagged observations of the Politburo to calculate connections of candidates to incumbent Politburo members. Our data end with the eighteenth term ( ). We similarly construct CollegeTie based on Central Committee and Politburo members' undergraduate institutions, for the eighth through eighteenth terms. When analyzing the eects of college ties, we include only the 1361 candidate-term observations for individuals with a college degree (68.9 percent of the full sample) to avoid confounding the eect of college ties on selection with the eect of having a college degree See last accessed March 28, We also begin our data in the post-war period because it is when Mao came to power. In the previous term, which stretched from 1928 to 1947, the Chinese central government was also structured quite dierently. For example, the Central Committee had only 23 members, as compared to the approximately 200 members it has had for most of the post-war period. 14 The estimated eect of college ties is reduced by about a third if we include non-college graduates and set CollegeT ie = 0 for these observations. 10

11 For workplace ties, we require that Politburo candidates and Politburo incumbents have a period of overlap in their work histories, more specically a period of time in which both worked in the same organization/department in the same prefecture. While no single position within the Central Committee guarantees Politburo membership, some positions tend to be elected at much higher rates than others. We therefore include controls for whether a Central Committee member is a military ocer (Military); an indicator denoting that an individual is the party secretary of one of the directly-controlled municipalities of Beijing, Shanghai, and Tianjin, or is the party secretary of Guangdong (4_Leader s) since these are positions that have most commonly (but by no means always) seen representation in the Politburo; an indicator variable for provincial leaders (Province); and to account for political dynasties we include the variable Princeling, which captures whether any of the candidate's parents or parents-in-law ever served in the Politburo. We also include hometown, workplace, and college xed eects to capture average dierences in the rate of Politburo selection as a function of these background characteristics. Our data include 1143 distinct candidates, 565 of whom appear only once in our data. A substantial number also appear as candidates twice (382 individuals) and three times (147 individuals). We dene PriorCandidacies as the number of previous terms an individual appeared as a (non- Politburo) member of the Central Committee. We control for prior candidacies throughout, given the higher likelihood of success for longer tenured Central Committee members. Table 1 provides summary statistics on the main variables we employ in our main analysis. Observe that workplace ties are by far the most common form of connection, despite our requirement that individuals overlap both in department and prefecture. This statistic emphasizes the fact that political elites often come up through similar career channels, with many spending time at the Secretariat of the Central Committee (55 candidates) and the Organization Department of the Central Committee (39 candidates), both located in Beijing. It is this type of concern that leads us to analyze the eect of connections after accounting for city-, college-, and workplace xed eects, since we need to account for the fact that some regions, schools, and workplaces tend to produce more high-level ocials than others. College ties are the least prevalent form of social connection. This arises, at least in part, because nearly a third of candidates (concentrated in the earlier part of our sample) did not complete a college degree and hence have no college tie. 3 Results Before turning to regression analyses, we present a series of gures to explore the patterns in the raw data. In Figure 1, we show the fraction of Central Committee members that are elected to the Politburo as a function of their college connections. We include only college graduates, to avoid 11

12 merely picking up the eect of having a college degree. We present this comparison in the left panel, which shows a higher rate of Politburo election among college-connected candidates (9.6 percent versus 6.8 percent for unconnected candidates). However, a large fraction of the sample nearly two-thirds of college-educated candidates graduated from schools with no connections throughout the entire period. And as we noted in the introduction, this leads to diculties in interpreting the full sample results: while it is possible that the higher rate of selection among connected candidates indicates that colleges with more frequent representation on the Politburo are able to exploit those connections to maintain relatively steady representation, it may also be the result of unobserved quality dierences across colleges. The concern over quality dierences is underscored by a comparison of colleges with frequent Politburo ties versus those with no Politburo representation at all. For example, by far the most common college of attendance in the post-mao era is Tsinghua University (24 members, or 16.7 percent of the sample), also China's most prestigious university. 15 Peking University the country's second-ranked school produced the second-most Politburo members (8, or 5.6 percent) since The pool of Central Committee candidates is also dominated by individuals from elite schools, though less so than the Politburo 6.3 percent of Central Committee members attended Tsinghua, 4.7 percent attended Peking University, and more broadly elite universities are over-represented. Overall, the patterns in the data suggest that there is positive selection on education as one rises through the bureaucracy, and hence a need to try to control for it. To mitigate concerns of unobserved quality dierences across colleges, we omit the neverconnected college graduates in the right panel of Figure 1, leading to a lower selection rate for connected candidates: 9.6 percent, versus 13.1 percent for unconnected candidates. By including college (or hometown or workplace) xed eects, we will identify the eect of ties from the rotation of individuals on and o the Politburo, which we argue deals eectively with concerns of unobserved quality. Figure 2 shows the analogous patterns for hometown ties. The left histogram, which compares the election rates of candidates with and without hometown ties to the incumbent Politburo members for the full sample, shows that the selection probability is near-identical for candidates with and without hometown ties (7.4 versus 6.9 percent respectively). When we omit candidates from hometowns that have zero hometown ties throughout the entire sample period (48 percent of all observations) in the right panel, hometown connections are associated with a lower chance of Politburo selection: 7.4 percent, versus 11.8 percent for those without hometown connections. Finally, we present the comparable gure for workplace ties in Figure 3. In the left histogram 15 Far fewer Politburo members were college educated prior to Tsinghua is still the dominant college of Politburo members if we use the entire sample. 12

13 we observe consistent with prior research that workplace ties are very strongly correlated with election: Politburo election rates are 9.4 percent versus 3.0 percent for those with and without workplace ties respectively. To focus on workplaces with a reasonable amount of within-workplace variation in connections, we omit workplaces with W orkt ie = 0 throughout the sample and also those that are almost always connected (W orkt ie = 1 for at least 85 percent of observations). This reduces the sample by 44.5 percent. When we compare election rates in this subsample, the gap between connected and unconnected candidates narrows to 6.3 versus 2.9 percent. This narrowing indicates that some workplaces may simply produce individuals destined for high-level leadership positions, rather than serving as a source of favoritism between incumbent and prospective Politburo members. Again, we will use xed eects to account for such dierences in the regression results that follow. 3.1 Main regression results Our main analyses explore the relationship between connections and Politburo selection, including a range of controls. Our specications all take the following form: Elected it = β Connection c it + γ c + ω t + ɛ it (1) where Elected it is an indicator variable denoting that Central Committee member i was elected to the Politburo for term t, and Connection c it denotes that candidate i was connected to at least one incumbent Politburo member via connection type c {HometownT ie, CollegeT ie, W orkt ie}. For each type of connection, we include a full set of xed eects for the source of the tie. So when we measure connections by hometown ties we include 264 hometown xed eects; similarly, we have 391 college xed eects for the college tie specication, and 289 workplace xed eects for the workplace tie specication. 16 It is because of the large number of xed eects in each case that we consider each type of connection separately (the correlation among connection types is low, so that examining each connection type separately is less of a concern than it otherwise might be: the correlation between HometownTie and WorkTie is 0.056, while ρ(hometownt ie, CollegeT ie) = and ρ(w orkt ie, CollegeT ie) = 0.074)). ω t is a term xed eect, and ɛ it is an error term clustered at the candidate-level. We begin by presenting results based on specications that do not include connection type xed eects, and thus incorporate both the within- and between-group variation in connections. noted above, this may create an upward bias in ˆβ, because of variation in candidate quality across 16 We include xed eects for all city-department combinations for which there exists at least one overlap in the workplace histories of a Politburo member and a Central Committee member. As 13

14 groups. In columns (1) (3) of Table 2, we show results using hometown, college, and workplace ties as measures of connections. In line with the patterns presented in Figures 1 and 2, there is a modest positive correlation between connections and Politburo selection, using hometown ties (column (1)) or college ties (column (2)) as the measure of connections, though this relationship does not approach statistical signicance in either case. Using workplace ties in column (3), we nd a strong positive association between connections and Politburo selection. In columns (4) (6) we include a set of controls, including log(age), PriorCandidacies, 4_Leader, Provincial, Military, and Princeling, as well as a set of education dummies indicating completion of a college, master's, and doctoral degree. These quality measures lead to a modest reduction the estimated eect of connections. We next turn to our main results in Table 3, which augments each column of Table 2 with connection type xed eects. In column (1), in which we use HometownTie as our connection measure and include hometown xed eects, we generate results that are in line with those shown in the right half of Figure 1: hometown-connected candidates are 7.7 percentage points less likely to be selected as Politburo members (p-value < 0.001). In column (2) we use CollegeTie as our measure of connections, and include undergraduate institution xed eects; again, we nd a negative impact on Politburo election, of 10.1 percentage points (signicant at the 5 percent level). In column (3), with WorkTie as the connections measure and workplace xed eects, we nd a precisely estimated near-zero eect, and can reject at a 95 percent condence level a positive work tie eect of greater than 3 percentage points. This may be because of the relative unimportance of work ties on average, or because our proxy for workplace-based networks is too coarse. (Francois et al. (2016), for example, look at the Shanghai Gang of ocials who worked in the Shanghai municipal bureaucracy in some capacity, highlighting the exceptionality of the Shanghai political machine. 17 ) We include additional candidate-level controls in columns (4) - (6), which leads to a small reduction in our estimates of the eect of hometown and college connections on Politburo selection. The coecient on HometownTie in column (4) implies a 6.1 percentage point reduction in the probability of Politburo selection (signicant at the 1 percent level). Relative to the selection base rate of 11.8 percent for HometownT ie = 0 candidates (from hometowns with variation in this variable), our estimate implies that a hometown tie reduces a candidate's election probability by 52 percent. The coecient on CollegeTie in column (5) implies an 8.9 percentage point reduction in the probability of Politburo selection, a very large impact given the base rate of election of 13.1 percent for CollegeT ie = 0 candidates (who graduated from colleges with some variation in CollegeTie). 17 When we look at the eect of Shanghai Gang connections using the denition of Francois et al. (2016), we estimate a negative eect, though very imprecisely measured. 14

15 As noted in Section 2.3, the inclusion of group xed eects may create a mechanical negative relationship between connections and selection, because a group becomes connected at t+1 precisely because a well-qualied candidate from the group at time t was elected in order to create the connection. This bias may be exacerbated by the xed eects, which emphasize the within-group variation in connections. To assess the extent to which this is likely a rst-order concern, we analyze at a subsample of the data that includes only the candidate-term observations when an individual rst appears in the Central Committee (and hence as a candidate for the Politburo). This removes from the sample the leftover candidates who are passed over (and thus remain in the candidate pool for the next term) when a group members is elected. Assuming that the quality of new arrivals at the Central Committee is independent across terms, the selecting out of high-quality candidates should be less of a concern in this subsample. We present these results in Table 4, which estimates equation 1 on the subsample of rst-time candidates. The point estimates reecting the connections penalty for hometown and college ties are diminished by about a third, though both are still large in magnitude, and in both specications the hometown connection penalty remains highly signicant. While this test does not provide a decisive rejection of the selecting-out hypothesis (for example, the selecting out of higher quality candidates could take place below the Central Committee level), it does provide some suggestive evidence to the contrary. Finally, we note that there is a small positive and borderline signicant correlation between workplace ties and selection in column (3), which is reduced in magnitude and signicance with the addition of controls in column (6). This might oer some tentative evidence of workplace-based favoritism or soft information (as a result of the insider knowledge that a workplace tie might facilitate). 3.2 Interpreting the connections penalty While our main objective in this paper is assessing the overall correlation between connections and Politburo selection, in this section we oer some tentative results to adjudicate amongst the various explanations for the connections penalty set forth in Section 2.3: (1) anti-factionalism; (2) competition within and between factions; and (3) quotas. Throughout this section, we focus on hometown and college ties, given the lack of any discernable eect of workplace ties on selection. We begin by examining the role of anti-factionalism. While, as noted earlier, leaders since the revolution have spoken against factionalism, this rhetoric was particularly emphatic under Mao. We conjecture, therefore, that if the need to take a visible stand against factionalism is behind the connections penalty, the eect will be particularly pronounced under Mao's leadership. We therefore dene an indicator variable Mao to denote elections that took place during the Mao 15

16 era (terms 8-11, which comprise elections up to and including 1977). We present in Table 5 specications that parallel those in our main results, with the interaction M ao Connection added for each of hometown and college ties. In column (1), which includes only hometown xed eects as controls, the direct eect of HometownTie, which captures the eect of hometown ties in the post-mao era, drops by a third relative to its size in Table 3 (signicant at the 5 percent level). The coecient on the interaction term Mao HometownT ie is negative and nearly twice as large as the direct hometown eect, implying that the eect of hometown ties on selection was three times larger under Mao. The outsized eect of hometown ties under Mao is further amplied when we add individual-level controls in column (3). (If we include a full set of interaction controls, i.e., Mao times each of the control variables, none of the interactions is signicant, and an F-test for their joint signicance has a p-value of 0.22.) While there are potentially multiple explanations for the stronger negative eect of hometown ties in the Mao = 1 period, we argue that it is broadly consistent with the view that under Mao, with his very public anti-factionalist stance, the political elite may have been particularly attuned to concerns of in-group favoritism. In columns (2) and (4) we show the results for college ties. While the coecient on Mao CollegeT ie is negative in both cases, the estimates are small relative to the direct eect, and imprecisely estimated. Table 6 disaggregates hometown and college ties to provide some tentative evidence on the roles of inter- and intragroup competition. As noted in the introduction, the role of intergroup competition is suggested by research on factional power sharing to balance the inuence of individual groups or networks. Interfactional competition may imply that coalitions of incumbent Politburo members would block the candidacies of already-powerful hometown or college groups, to keep them from gaining yet more Politburo members. We operationalize this prediction by comparing the eect of connections to members of the most-represented group among incumbent Politburo members to the eect of connections to other Politburo incumbents. We dene the indicator variable Largest City Tie to denote candidates who share their hometown with the most commonly represented hometown among Politburo incumbents. We analogously dene Largest College Tie for ties to the most prevalent college among Politburo incumbents. In the rst two columns of Table 6, we present results that allow the eect of hometown and college ties to dier for the largest group of each type. In both cases, the coecients are very similar for the variables capturing ties to the largest group and all other groups, which appears inconsistent with the intergroup competition view. We next explore whether the negative eect of connections results from Politburo members working to maintain their dominant position within a group network by blocking potential challengers from within their own network (intragroup competition). If this is a primary reason for our empirical result, we argue that it should be attenuated among more senior members of the 16

17 Politburo, who will be less threatened by new (more junior) members. To capture the dierential eect of seniority, we disaggregate our connections measure to look at ties to Politburo Standing Committee (PSC) members versus other (non-psc) Politburo members. 18 We present these results in the second pair of columns in Table 6. Looking rst at hometown ties, in column (3) the coecient on PSC hometown ties close to zero, whereas the coecient on non-psc ties is large (-0.072) and signicant at the 1 percent level. We can reject equality of the PSC and non-psc hometown tie coecients at the 10 percent level. For college ties (column (4)), while the PSC tie coecient is again closer to zero than the non-psc tie coecient, the dierence is small and the two are statistically indistinguishable from one another. Overall, Table 6 provides no evidence in favor of intergroup competition and mixed evidence on intragroup competition. Finally, we consider whether the negative impact of connections on election at least for hometown ties derives from geographic quotas. This is related to the intergroup competition hypothesis above, in that it implies a more equitable distribution of Politburo members across groups, as top leaders avoid selecting candidates from regions already represented in the Politburo to ensure that no single area benets from disproportionate inuence. The lack of any evidence of a largest group eect already cases doubt on this possibility. To further probe whether this is a plausible explanation for the negative hometown tie result, in Table 7 we examine the relationship between home province ties and Politburo selection. In China, the central government has direct control over provincial leadership positions, while provincial leaders control municipal appointments. Given this hierarchy, if there were reason for the central government to demonstrate equitable treatment regionally it would most likely operate at the provincial level. Table 7 presents variants on specication (1), using HomeprovinceTie as our measure of connections. In columns (1) and (2) we present specications that use only HomeprovinceTie as a measure of connections, and include home province xed eects. The point estimate on HomeprovinceTie, while negative, is small in both cases and not statistically signicant. In columns (3) and (4) we include HometownTie as a covariate. Once we account for birth city, the coecient on HomeprovinceTie is positive, though small and not statistically signicant. These results cast doubt on the geographic quotas explanation for our ndings, because historically such quotas formal or otherwise within the Chinese central government bureaucracy operate at the province level. Overall we take the results in this section as providing tentative support for two possible explanations for the connections penalty the stronger eect during the Mao years is consistent with 18 If we distinguish between connections to older versus younger Politburo members, split based on the ocial retirement age of 67, we similarly nd that the negative correlation between hometown ties and selection is driven by ties to younger members. The interpretations associated with this nding are similar older members may feel less threatened by competition, particularly if they are on the cusp of retirement and hence may even wish to elect a within-group successor. Interestingly however, even for retirement age Politburo members, the eect of hometown ties is still non-positive. 17