2. Informal elite competition and media reports under authoritarianism

Popular descriptions of dictatorships have long suggested that an autocracy with non-monopolized power is a weak regime (O'donnell and Schmitter, Reference O'donnell and Schmitter1986; Przeworski, Reference Przeworski1991). However, the literature on authoritarian power-sharing challenges such perceptions, arguing that political power-sharing institutions, such as political parties, elections, and legislatures, help dictators enhance regime sustainability by successfully limiting their unilateral power (Geddes, Reference Geddes2003; Myerson, Reference Myerson2008; Gandhi, Reference Gandhi2008; Magaloni, Reference Magaloni2008; Svolik, Reference Svolik2009; Boix and Svolik, Reference Boix and Svolik2013; Svolik, Reference Svolik2013; Pepinsky, Reference Pepinsky2014). In particular, power-sharing studies link single-party or dominant-party dictatorships to regime durability, as those parties more effectively contain power struggles among ruling elites (Geddes, Reference Geddes2003; Magaloni, Reference Magaloni2008; Magaloni and Kricheli, Reference Magaloni and Kricheli2010).

In this paper, we examine how power-sharing in a party dictatorship works when the main channel of power distribution and competition consists of non-institutionalized and informal political networks, i.e., factions. As a single-party authoritarian regime, the Chinese political system does not adopt many democratic institutions that are intended to decentralize de facto political power; although a few political reforms have been attempted, the extent of institutionalized power sharing remains quite limited. Scholars have viewed the Politburo Standing Committee (PSC) as a key power-sharing institution in China as it provides a system of collective leadership (Lin, Reference Lin2004; Svolik, Reference Svolik2013). While this collective leadership is the fundamental basis of the Chinese political system, beyond a nominal division of labor among PSC members, the broad political process that influences the composition and management of the PSC is not formally institutionalized and thus hinges on an informal power struggle among past, current and potential national leaders.

Scholars of Chinese politics have discussed whether power distribution across informal political groups can induce a power-sharing or power-balance outcome. The seminal work by Nathan (Reference Nathan1973) and a subsequent study by Dittmer and Wu (Reference Dittmer and Wu1995) claim that policy and ideological struggles between factions create an inter-factional balance of power in which multiple factions compete and maintain balance among themselves. In contrast, Tsou (Tsou, Reference Tsou1976, Reference Tsou1995) argues that elite pluralism is an empirical exception in Chinese politics and that the power struggle among Chinese elites “always involves one side winning all and/or the other side losing all” (Tsou, Reference Tsou1995, p.97). Countering Tsou's point, Nathan and Tsai (Reference Nathan and Tsai1995) argue that an outcome with a balance of power is a not only frequent but also stable feature of Chinese politics. More recently, Li (Reference Li2012) characterizes the collective leadership of the CCP as a system of checks and balances through “one party, two coalitions,” where equally powerful factions representing different social and political groups in China hold each other in check.

Despite the long-standing theoretical debate, few studies have used empirical data to illustrate the competitive behaviors of authoritarian elites across factions. Our study contributes by providing an empirical examination of how members of strong factions and weak factions behave differently. Our period of research (2000–2014) ranges from the end of the Jiang Zemin administration (1989–2002) to the beginning of Xi Jinping's regime (2012–present). The Hu Jintao and Wen Jiabao administrations (2002–2012), which constitute the bulk of the period under analysis, are not typically framed as a period in which one faction or one top leader dominated others. Furthermore, during this period intellectuals within the party actively debated the possibility of intra-party democracy (dangneiminzhu) (Bing, Reference Bing2014).

If this seemingly balanced period is indeed characterized by power-sharing and power-balancing among political factions, our analyses will show that the relatively dominant factions are criticized or challenged by weaker counterparts alleging corruption equally or more frequently than the reverse. Previous faction politics theories support this pattern and potential balancing effects. Nathan (Reference Nathan1973) claims that “the flexibility of the weaker factions and their capability for intermittent functioning” enhance weak factions’ ability to defend. Furthermore, when a strong leader emerges, weaker factions perceive it as a common threat and band together against the strong faction (Nathan and Tsai, Reference Nathan and Tsai1995).

Hypothesis I: Powerful factions are more or equally likely to receive negative news coverage than are weaker factions.

If, on the contrary, the intrinsic nature of Chinese elite competition is to remove threats from one's political competitors in order to obtain more power as Tsou (Reference Tsou1976, Reference Tsou1995) theorizes, the analyses will reveal an imbalanced pattern of attacks in which stronger factions aggressively attack their weaker counterparts, using the opportunity and power to do so. However, empirical patterns may differ depending on how these political attacks are strategized and constrained at the client level in the factional structure. On the one hand, strong faction members may hold the members of the most challenging factions in check, as the rise of these factions could threaten the status and resources enjoyed by the most powerful faction. The recent literature of elite purges in autocracies provides insights supporting this claim. Several studies investigate the targets of purges under dictatorship and find that competent and experienced elites are more likely to be purged by a dictator as they are perceived as greater threats (Wong and Chan, Reference Wong and Chan2021; Goldring and Matthews, Reference Goldring and Matthews2021; Bokobza et al., Reference Bokobza, Krishnarajan, Nyrup, Sakstrup and Aaskoven2022).

Hypothesis II: Media reports by strong factions are more likely to target the second strongest faction than to target weaker factions.

Nonetheless, attacking the second strongest faction members may come with potential political costs, especially as a client, if the attacked faction's patron retaliates in the future. This risk may induce the members of strong factions to attack an easy target, such as members of the weaker or weakest faction. This strategization is similar to the logic that dictators purge the opposing elites when the elites’ capabilities to fight back are low (Sudduth, Reference Sudduth2017) or that coup-entry dictators are more likely to purge civilian elites because the cost of purging is lower than purging military elites (Goldring and Matthews., Reference Goldring and Matthews2022).

Hypothesis III: Strong factions are more likely to attack the weakest faction(s) rather than targeting other strong factions.

We measure the competitive behavior of Chinese political elites by observing regional news reports on corruption investigations in other provinces. One principal reason for relying on local news reports is the unavailability of information on central-level political competition. While autocracies are in general less transparent than democracies (Hollyer et al., Reference Hollyer, Rosendorff and Vreeland2015), political competition among top national leaders in non-electoral autocracies like China tends to be even more opaque. One way to overcome this data limitation is to incorporate behavioral data from lower-level political elites who are directly connected to the top national leaders. The justification is that, in a hierarchical political system such as China's, the behavior of lower-level elites ought to benefit their political patrons (Shih, Reference Shih2008b). In particular, this study investigates the possibility that strong factions may benefit by being able to be more vocal in criticizing the defects of other factions, as an alternative benefit of factional connection to a strong patron, creating greater legitimacy for the promotion of strong faction members not only within the party but also among the public.

Another reason for employing local media reports is the limited discretion that local governments and local media have on reporting negative political incidents such as corruption investigations. For this, it is worth discussing how news articles are published in Chinese regional newspapers. It is well-known that media freedom is restricted in China as in many other authoritarian countries (King et al., Reference King, Pan and Roberts2013; Stockmann, Reference Stockmann2013). However, total concealment of negative information on governance is not optimal to the regime as a media environment that only allows glorifying news reports fails to provide adequate incentives for bureaucrats to perform better (Lorentzen, Reference Lorentzen2014). Furthermore, the information provided by such a government lacks credibility among citizens, leading them to look for alternative channels (Zhu et al., Reference Zhu, Lu and Shi2013; Huang, Reference Huang2017). In particular, when the legitimacy of the regime depends on internal discipline and reform efforts, information on self-discipline and reform progress must be transmitted to the public to maximize regime support. Thus, allowing negative or critical reports on local affairs can be a useful strategy for the state to maximize regime stability (Lorentzen, Reference Lorentzen2014). As our data exemplify, it is not impossible to find news reports on corruption cases in local newspapers in China, and some reports discuss the details of local officials’ corruption.Footnote ².

This does not imply, however, that Chinese local media serve as a watchdog with regard to local government. Although there is much variation in local media in terms of the extent to which different outlets criticize the government or report on sensitive social issues (Lei, Reference Lei2016), surveillance or criticism by the media is still structurally unlikely and rare. This is because all local media are either operated or supervised by the local propaganda department, and the local propaganda department, in turn, is supervised by the party committee at the corresponding level, which is under the authority of the local party secretary.Footnote ³ Therefore, by design, there is little incentive for editors and managers of local media companies to criticize their own local government, as their career paths are largely decided via local leaders’ evaluations. However, media outlets have much more latitude to criticize other localities’ misgovernance. Criticizing other local governments does not interfere with the political fortunes of their own supervisor, and may actually contribute to the supervisor's success by disparaging local leaders who are in competition with the supervisor.

In summary, while the freedom of the Chinese media is among the most limited in the world, what circulates and resonates in media does matter in China (Distelhorst, Reference Distelhorst2012; Yang, Reference Yang2020). As corruption investigations are conducted by a CCP organ, news reports about ongoing corruption investigations do not provide new information to national leaders, but the leaders still cannot necessarily control which cases are circulated by the media or which become well-known to the public (Lorentzen, Reference Lorentzen2014; Lu and Ma, Reference Lu and Ma2019). At the same time, in a strictly hierarchical governance system like China, a corruption investigation of subordinates, especially one widely known of by the public, can damage one's career and thus one's promotion chances, as it signals incompetence, negligence of political responsibility, or potentially one's own corruption (Chen and Hong, Reference Chen and Hong2021).

3. Empirical strategy

3.1 Data and variables

Our key explanatory variable is the distribution of power among PSC members. The Chinese political system depends on collective leadership at the top level; seven to nine members constituting the PSC collectively hold central authority in the Chinese government. However, measuring the dynamic changes in relative political influence among these PSC members is not a straightforward process. The most accessible method is to employ the official rank of PSC members, but since the ranking of PSC members only renews every five years at the National Party Congress, it remains static during that period and is thus of limited use.

To measure the dynamic changes in the actual power distribution among PSC members, we generate alternative power indicators. The first alternative measure is based on national media exposure of the PSC members.Footnote ⁴ We take advantage of the Chinese media's close relationship with the central authority and their tendency to adapt to power shifts to measure the dynamic changes in power distribution among PSC members (Shirk, Reference Shirk2011; Jaros and Pan, Reference Jaros and Pan2018). We interpret the more frequent appearance of a PSC member's name in major national newspapers as a signal of his political importance and influence in China's politics relative to other members.

To construct this dynamic power measure, we collect all news reports that appeared in major news media from 2000 to 2014 that contained the names of PSC members at the time. We use the following four national newspapers in China: the Peoples’ Daily (Renmin Ribao), the Guangming Daily (Guangming Ribao), and China Youth Daily (Zhongguo Qingnian Bao). All of these newspapers are national, official party-line media, closely controlled by the CCPPD. To collect the news articles, we use the Hong-Kong based data vendor WiseNews, which provides the largest content database of mainland Chinese news media.Footnote ⁵

The time trends of newspaper articles mentioning the names of top national leaders are illustrated in the following figures. Figure 1 presents the annual trend of all PSC members. It is notable that the top two leaders, the president and the premier, are cited in the vast majority of our data. Figure A.1 shows the monthly trend of newspaper articles mentioning the top two leaders.

Fig. 1. Annual news reports on the Politburo Standing Committee members.

Finally, we build a yearly dynamic power indicator by calculating the share of news reports on a certain PSC member (p) out of the total number of news reports on all PSC members in each year (t):

(1)$${\rm Patron\ Power}_{\,pt} = {{\rm News}_{\,pt}\over \sum_{\,p}{\rm News}_{\,pt}}$$

Using this power measure, we also generate a variable measuring the power gap between the patron of the leader of the province producing the negative reporting (p _i) and the patron of the leader of the province being reported on (p _j). For province leaders with ties to more than one PSC member, we employ the sum of their patrons’ power, the highest-ranked patron's power, and the average power of their patrons in the analyses in order to rule out the possibility that the results are driven by a particular identification strategy.

(2)$${\rm Power\ {Difference}}_{\,p_i p_j t} = {\rm Patron\ Power}_{\,p_it}- {\rm Patron\ Power}_{\,p_jt} $$

In the empirical analyses, we employ these two dynamic power indicators as the main independent variables. To confirm that the empirical findings are not based on a specific measurement strategy, we use the various alternative measures, including one based on the official ranking and a power index based on the connectedness of the patron's political network. In addition, we generate another power index from national news media reports, in which we restrict the media exposure to those articles that mention a PSC member's name in the headline, rather than in the text of the article.

Our main dependent variable is political competition among provincial leaders, measured through inter-provincial “negative” news reports. We employ regional news media coverage of corruption investigations that took place in other provinces. To capture the competitive nature of news reporting on corruption, we create a province dyad pairing a news reporting province (news province, i) and a reported-on province (event province, j).

To collect the news reports from each province, we scraped the news contents of 143 local mainstream newspapers, i.e., provincial and prefectural newspapers, in China from WiseNews. Among these 143 newspapers, 49 are party-line newspapers directly controlled by the provincial or major prefectural party committees, and the remaining 94 are commercial newspapers operated by regional news corporations (Table A.1). After selecting newspapers, we searched for news reports on corruption investigations using the keyword “shuanggui.”Footnote ⁶

We then created direct data for each province dyad for each year. We include 30 provincial-level administrative divisions, excepting Beijing.Footnote ⁷ Each province is coded as a news province and an event province for each year. In other words, the unit of observation is a pair of provinces consisting of one news (reporting) province and one event (reported on) province, with a value that captures the number of news reports published by the news province on corruption investigations in the event province in a specific year. This design of the dependent variable results in 13,050 province dyads (30 news provinces × 29 event provinces × 15 years) consisting of 450 leader-years (30 provincial party secretaries × 15 years).Footnote ⁸

To identify the political network that exists between regional political leaders and top central leaders, we constructed a political elites dataset that captures details about the political careers of all provincial party secretaries and all CPS members during the period 2000–2014. We first incorporate the names of those provincial party secretaries from the China Communist Yearbook (zhonggongnianbao). We then extract biographical and career information from their personal biographies, including age, gender, place of birth, education, and work history, using the Chinese search engine Baidu Baike. Finally, we match the personal information and work histories of provincial leaders with information on the incumbent PSC members to construct faction networks for each provincial official. We assume a provincial party secretary is connected to a PSC member if they were born in the same province, graduated from the same school, or previously served in the same work unit for over a year, following Shih et al. (Reference Shih, Adolph and Liu2012). More recently, scholars have argued that workplace connections are a more accurate measure of factional relationships (Jia et al., Reference Jia, Kudamatsu and Seim2015; Keller, Reference Keller2016; Jiang, Reference Jiang2018). As the first robustness check, we employ the workplace connection exclusively.Footnote ⁹

3.2 Specification

The analyses are based on the following empirical specification.

(3)$$\eqalign{NegNewsCount_{ijt}& = \beta_1{Patron Power}_{it} + \beta_2{Patron Power}_{jt}\cr & \quad + \beta_3{{\rm \it {PS-GN\ SameFaction}}}_{it} + \beta_4{{\rm \it {PS-GN\ SameFaction}}}_{jt}\cr & \quad + \beta_5 AgeGap_{ijt} + \beta_6Tenure Gap_{ijt}\cr & \quad + X_{ijt}\mu + \eta_{ij} + \lambda_{t} + \varepsilon_{ijt}}$$

Our key dependent variable NegNewsCount represents the number of bilateral news reports published in province i (news province) on corruption investigations that occurred in province j (event province) in year t. To capture the dyadic reporting dynamics, we create two different news report variables: (1) the number of corruption investigation reports by i on corruption scandals in j and (2) the share of total news reports by i on event province j and j's prefectures that cover corruption investigations. Our independent variable is the power of the patrons connected to the leaders of the reporting and the reported-on provinces. We additionally examine the relative power of the patrons of the reporting and the reported-on provinces (Patron Power _it − Patron Power _jt).

We control for covariates affecting local leaders’ political incentives or competitive behavior. First, PS-GN Faction is an indicator variable for cases where the party secretary and the governor of a province belong to the same faction. The rationale behind this variable is that belonging to the same faction may encourage a province's leadership team to attack another province or may insulate a province from negative media reports. It is also necessary to consider other individual-level confounders that may lead certain leaders to compete more intensely with each other. For instance, all else being equal, public officials with similar ages or years of tenure are more likely to be in competition for promotion in the following evaluation cycle. To address these incentives, we control for the age gap and the years-in-current-office gap between the party secretaries in the two provinces that form each dyad. Our variables are summarized in Table A.2.

Furthermore, to control for any unobserved characteristics specific to individual provinces, inter-provincial relationships, or particular years, we employ two sets of fixed effects: for province dyad (η_ij) and for year (λ_t). Province dyad fixed effects address not only province-specific effects but also dyad-specific effects. Province-specific effects are particularly important to address because Chinese provinces enjoy considerably different levels of media availability and freedom (Lei, Reference Lei2016; Stockmann, Reference Stockmann2013). They also address the estimation bias caused by certain provinces being reported on more frequently because their officials tend to be more corrupt, as in the case of Shanxi province (Lorentzen and Lu., Reference Lorentzen and Lu.2018). Addressing dyad-specific effects is also crucial, as they will absorb the potential bias created from rivalry-prone provinces. Year fixed effects along with dyad fixed effects also prevent the possibility that certain events, i.e., a major event such as Bo Xilai's corruption case, drive our results. Finally, we use a linear model for panel data in the main analysis.Footnote ¹⁰ All errors are clustered at the dyad level to address the unique structure of dyadic data (Cameron and Miller, Reference Cameron and Miller.2014; Aronow et al., Reference Aronow, Samii and Assenova2015).