concealing corruption: how chinese officials …jenpan.com/jen_pan/sendup.pdfconcealingcorruption...
TRANSCRIPT
American Political Science Review, Page 1 of 19
doi:10.1017/S0003055418000205 © American Political Science Association 2018
Concealing Corruption: How Chinese Officials Distort UpwardReporting of Online GrievancesJENNIFER PAN Stanford UniversityKAIPING CHEN Stanford University
A prerequisite for the durability of authoritarian regimes as well as their effective governance is theregime’s ability to gather reliable information about the actions of lower-tier officials. Allowingpublic participation in the form of online complaints is one approach authoritarian regimes have
taken to improve monitoring of lower-tier officials. In this paper,we gain rare access to internal communi-cations between a monitoring agency and upper-level officials in China.We show that citizen grievancesposted publicly online that contain complaints of corruption are systematically concealed from upper-level authorities when they implicate lower-tier officials or associates connected to lower-tier officialsthrough patronage ties. Information manipulation occurs primarily through omission of wrongdoingrather than censorship or falsification, suggesting that even in the digital age, in a highly determined andcapable regime where reports of corruption are actively and publicly voiced, monitoring the behavior ofregime agents remains a challenge.
INTRODUCTION
A prerequisite for the durability of authoritarianregimes as well as their effective governanceis the ability to gather information about the
actions of government officials (Wintrobe 1998). How-ever, most mechanisms for monitoring governmentofficials—secret police, oversight committees—fail toprovide reliable information because officials haveincentives to collude with monitors to suppress infor-mation that would jeopardize their access to positionsof power and associated rents. To sidestep unreliablemonitoring agencies, authoritarian regimes have bor-rowed the strategy of bottom-up citizen participationused in democratic contexts to increase accountability(O’Donnell 1999; Olken 2007). By adopting channelswhere the public can lodge complaints and provideoversight over local officials, authoritarian regimes canin theory gather more reliable information becausecitizens have little incentive to collude with corruptofficials (Cai 2013; Dimitrov 2014a, 2014b). In thedigital age, these citizen complaints are often postedonline where they remain uncensored and publiclyaccessible (King, Pan, and Roberts 2013). Some arguethat autocrats can effectively glean information onofficials’ misconduct from posts shared on socialmedia and punish officials based on this information(Dimitrov 2014a; Nathan 2003; Qin, Stromberg, and
Jennifer Pan, Assistant Professor, Department of Communication,Building 120, Room 110 450 Serra Mall, Stanford University, Stan-ford, CA 94305-2050 (jenpan.com).
Kaiping Chen, Ph.D. Candidate, Department of Communication,Building 120, Room 110450 Serra Mall, Stanford University, Stan-ford, CA 94305-2050
Our thanks to Jianghong An, Rita Lu, Yanchen Song, Feiya Suo,Zhiheng Xu for excellent research assistance; to Steven Balla, PeterLorentzen, Guillermo Rosas, Lily Tsai, the SMAPP Global meet-ing, and Bay Area China Social Science Workshop participants fortheir extremely helpful comments and suggestions; and to the Stan-ford Asia-Pacific Scholars Fund, Stanford China Fund, and StanfordIRiSS Faculty Fellows Program for research support.
Received:December 23, 2016; revised: September 18, 2017; accepted:April 12, 2018.
Wu 2017a), and others suggests that these onlinechannels may increase authoritarian accountability(Gunitsky 2015; Yong 2005; Noesselt 2014; Qiang2011). Authoritarian regimes as diverse as China,Egypt, Pakistan, Russia, Saudi Arabia, Singapore, andVietnam have adopted online platforms where citizenscan express their grievances.1
In this paper,we gain rare access to internal commu-nications between amonitoring agency and upper-levelauthorities in China, and we show how citizen reportsof malfeasance implicating lower-level officials aresystematically concealed from upper-level authorities.Our evidence comes from an analysis of email archivesleaked from the Chinese Communist Party (CCP) Pro-paganda Department of a prefecture in south centralChina,which we call J. Prefecture.Although the leakedemails have received press coverage and remain pub-licly available, the archive is complex and large, andit has not been systematically analyzed. Using large-scale hand-coding and machine learning methods, weidentified 643 Online Sentiment Monitoring Reports(����) produced by the J. Prefecture PropagandaDepartment, which contain the details of 3,423 on-line complaints generating public anger and discontentfrom 2012 to 2014.Most importantly,we can differenti-ate between theOnline SentimentMonitoring Reportssent to upper-level provincial authorities and those thatare kept for internal circulation within the prefecture,
1 Citizen complaints in China will be discussed in the next Section. InEgypt, citizens can complain throughministry and other governmentwebsites (OECD 2013). Khyber Pakhtunkhwa province in Pakistancreated the KP Citizens Portal (http://smart.pmru.gkp.pk/) to gathercomplaints. In Russia, ministries have federal and regional websiteswhere citizens can submit complaints, for example, 36.mvd.ru/ forthe police department and 36.mchs.gov.ru/ for the ministry of emer-gency services in the Voronezh region. The Saudi Arabian websiteshakwa.net works with officials to gather citizen complaints. In Sin-gapore, citizen complaints are gathered through various online chan-nels (Rodan and Jayasuriya 2007). Various cities in Vietnam haveset up online complaint sites (e.g., http://egov.danang.gov.vn/gop-y).Note that online complaint forums may also serve other functionssuch as giving citizens an outlet to vent their anger or providing away to address grievances.
1
Dow
nloa
ded
from
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e. S
tanf
ord
Gra
duat
e Sc
hool
of B
usin
ess,
on
14 Ju
n 20
18 a
t 17:
19:3
4, s
ubje
ct to
the
Cam
brid
ge C
ore
term
s of
use
, ava
ilabl
e at
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e/te
rms.
htt
ps://
doi.o
rg/1
0.10
17/S
0003
0554
1800
0205
Jennifer Pan and Kaiping Chen
allowing us to compare online citizen complaints re-ported upward with complaints identified by the mon-itoring agency as important but not reported upward.This data shows that complaints implicating the pre-
fecture government in wrongdoing—including embez-zlement, graft, and violence—are less likely to be re-ported upward to provincial superiors compared tocomplaints related to governance issues such as edu-cation quality and pollution. This data also shows thatwrongdoing by subordinate counties with workplaceand birthplace ties to prefecture leaders are less likelyto be reported to provincial superiors, whereas citizencomplaints of wrongdoing implicating counties withoutpolitical connections to prefecture-level officials aremore likely to be reported up to the province. Finally,we find that citizen complaints of government wrong-doing are censored by J. Prefecture when complaintsare posted on websites managed by J. Prefecture; how-ever, the large majority of complaints remain publiclyavailable because they are posted to websites where J.Prefecture has no censorship authority.The information manipulation we identify is distinct
from the known problem of falsification of economicand fiscal data inChina (Cai 2000;Park andWang 2001;Tsai 2008; Wallace 2016). While falsification of eco-nomic statistics entails manipulation through falsifica-tion of performance targets, the information manipula-tion we identify is rooted in concealment. Some publiccomplaints are reported upward, and these complaintsreflect real grievances and public concerns; however,information distortion occurs because what is reportedupward simply does not reveal the entire picture, po-tentially increasing the difficulty of penalizing officialsfor engaging in this form of information manipulation.Furthermore,while economic falsification is focused ondistorting performance-based targets, manipulation ofpublic opinion directly distorts understandings of pub-lic satisfaction, contention, and regime support.Our re-sults show that even in a highly determined and capableauthoritarian regime where grievances are actively andpublicly voiced, lower-level officials continue to distortand manipulate information.These findings have implications for our understand-
ing of political communication and information con-flicts in authoritarian regimes. Authoritarian regimesface two types of conflicts over information. The firstconflict occurs between the regime and the public overwhat information is publicly available.The second con-flict occurs among regime elites, specifically, betweenupper-level authorities who want to monitor lower-level agents and lower-level agents who want to hidemalfeasance. Research in political communication hasemphasized the first conflict, with an emerging consen-sus that authoritarian regimes continue to control pub-licly available information in the age of social media(Kalathil and Boas 2010; MacKinnon 2012; Morozov2012). Our results reveal a different picture by focus-ing on the second conflict, joining a new strand of re-search that shows the constraints of autocrats’ abilityto gather reliable information about regime agents inthe digital age (Pan 2016). Even though authoritarianregimes may learn much by monitoring social media
and digital data as scholars have suggested,2 access tolarge-scale data by no means guarantees autocrats om-niscience over regime agents.These results enrich our understanding of the re-
lationship between nonelectoral forms of citizen par-ticipation and accountability in authoritarian contexts.Recent research shows how local governments in au-thoritarian regimes acknowledge and respond to citi-zen complaints at relatively high rates (Chen, Pan, andXu 2016; Distelhorst and Hou 2017; Meng, Pan, andYang 2017). There is a great deal of optimism sur-rounding the benefits of citizen participation in China,in nurturing an informed citizenship, in informing theregime of the public’s preferences, and even in facili-tating government accountability (Fishkin et al. 2010;He andWarren 2011;Ma 2012).However, governmentofficials can only be held accountable if information ismade available to those that have power to sanctionthese officials, and if those with sanctioning power havethe incentives and resources to punish offenders. Inauthoritarian contexts such as China, this sanctioningpower resides with upper-level authorities that controlthe promotion prospects of lower-tier officials, not thepublic. Therefore, our results, showing how the public’sgrievances and complaints do not always reliably maketheir way to upper-level authorities, reveal obstaclesto bottom-up accountability in authoritarian regimeseven when citizen participation increases. Our resultsdo not contradict findings that online and offline cit-izen complaints can uncover government corruptionand lead to the punishment of government officials;they simply suggest thatmany citizen reports of corrup-tionmay not lead to sanctions, and citizen participationis likely insufficient to root out corruption and to createthe conditions for full accountability.3
INFORMATION GATHERING ANDINFORMATION MANIPULATION
Authoritarian regimes are often administered throughmultiple layers of government, and central autocratsdelegate responsibilities for governance to multiplesubnational levels of government.4 In this section,we describe how dual incentives for informationgathering and information manipulation exist at each
2 See Qin, Stromberg, and Wu (2017a), Qin, Stromberg, and Wu(2017b), and MacKinnon (2012) for discussion of China; Pearce andKendzior (2012) of Azerbaijan; Gunitsky (2015) of Bahrain; House(2015) for Uzbekistan; andMacKinnon (2012),Morozov (2012), andGunitsky (2015) of Russia.3 As of December 2016, none of the prefecture-level officials of J.Prefecture, including those accused of corruption in citizen com-plaints,have been implicated inXi Jinping’s extensive anticorruptioncampaign.4 For example, Russia consists of federal subjects, which are dividedinto administrative districts. Administrative districts consist of cities,towns, and rural settlements. Cities are further subdivided into citydistricts. China is divided into 31 provincial-level administrative di-visions,which are further subdivided into prefectures;prefectures aresubdivided into counties (also called districts), and counties are sub-divided into townships (also call subdistricts). Similarly, Iran is di-vided into 31 provinces, which are further subdivided into counties;counties are subdivided into districts, and districts are subdividedinto cities and rural districts.
2
Dow
nloa
ded
from
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e. S
tanf
ord
Gra
duat
e Sc
hool
of B
usin
ess,
on
14 Ju
n 20
18 a
t 17:
19:3
4, s
ubje
ct to
the
Cam
brid
ge C
ore
term
s of
use
, ava
ilabl
e at
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e/te
rms.
htt
ps://
doi.o
rg/1
0.10
17/S
0003
0554
1800
0205
Concealing Corruption
subnational level, and we show how variation inincentives and resources between subnational levelsfacilitate or hinder surveillance capacities.
Theory of Information Gathering andManipulation
We assume the interest of political leaders at all subna-tional levels in an authoritarian regime is to maximizerents, and the primary way to maximize rents is to stayor advance in political office.5 The logic we describecan be applied to any superior-subordinate relation-shipwhere the superior controls the career prospects ofthe subordinate,andwhere the superior is also subordi-nate to some higher-level authority. For example, pre-fecture party secretaries in China (the superior) con-trol the career advancement of county party secretaries(the subordinate), but prefecture party secretaries arethemselves subordinate to provincial leaders. For ourset-up, there are two actors: (1) the superior who werefer to as the upper-level official and (2) the subordi-nate who we refer to as the lower-level official.6To remain or advance in political office, upper-level
officials delegate tasks to lower-level subordinates—for instance, attract investment, control crime, improveinfrastructure. However, in the course of carrying outthese tasks, lower-level officials engage in corrupt prac-tices to enrich themselves.7 Malfeasance by lower-levelofficials, and the by-products of corruption such as pub-lic dissatisfaction and unrest, could lead to scandalsthat impede or ruin the political careers of upper-levelofficials.Upper-level officials are interested in avoidingsuch scandals, and to do so, upper-level officials maywant to strengthen monitoring of lower-level officialsto determine whether the activities of subordinates willimpede their own political careers and to remove fromoffice subordinates whose actions jeopardize their owncareers. At the same time, since the rents of lower-tier officials are associated with political office, lower-tier officials have strong incentives to hide actions that
5 Research has established links between political office and mone-tary rewards (Truex 2014; Zhang, Giles, and Rozelle 2012). While itwould not be a stretch to imagine that the interest of central-levelleaders is also to maximize rents, we focus on subnational levels be-cause the way tomaximize rents for central-level leaders differs fromsubnational leaders. For example, central leaders might maximizerents by pursuing the survival of the regime since they can no longeradvance to higher political positions.6 We do not conceptualize monitors or monitoring agencies as in-dependent actors. We assume that monitors are either aligned withthe lower-level official they are supposed to monitor or aligned withthe upper-level official they are gathering information for. Monitorscan also be conceptualized as independent actors—secret police andother repressive apparatus can be powerful and can act indepen-dently. However, in our empirical context, monitoring agencies areseparate from the repressive apparatus and have limited resourcesrelative to lower-level and upper-level officials.7 Although officials have a choice in whether or not to engage in cor-ruption, we assume that corruption is endemic to officials in author-itarian regimes. Corruption is often defined as the abuse of publicoffice for private gain, but in many autocracies, access to the spoilsof political office is implicitly promised to political officeholders asa way of managing intra-elite conflict and coopting elites (Magaloni2008;Magaloni and Kricheli 2010).
would lead to scandals of their own and their removalfrom political office. This means that to survive oradvance in political office, political leaders at eachsubnational level of an authoritarian regime have dualincentives to (a) hide their own wrongdoing from su-periors who control their career prospects and (b) ob-tain enough information about the subordinates theycontrol so the activities of subordinates do not jeop-ardize their own career prospects. How these dualincentives play out in terms of the effectiveness ofinformation gathering vs. information manipulationdepends on the interaction between the incentives andresources of lower-level and upper-level officials. Be-low, we first examine each component, and then exam-ine their interactions.
Variation in Incentives. While these dual incentivesexist among all subnational officials, the strength ofboth the incentive to hide wrongdoing from superiorsand the incentive to obtain accurate information aboutsubordinates varies among officials depending on twomain factors: (1) the career ambition of the official and(2) the conditions of the region the official governs.Officials’ levels of concern with scandal varies with
their career ambitions—an extremely ambitious offi-cial actively seeking career advancement is likely moreconcerned with scandal than an official who is onlyinterested in preserving his or her current politicalposition. As a result, more ambitious officials havegreater incentives to hide wrongdoing from superiorsand greater incentives to monitor their subordinatesthan officials who only want to remain in their currentposition. Lower-level officials who simply want to re-main in office may only be interested in hiding wrong-doing that would result in major scandals, while lower-level officials who are actively pursuing advancementmay have incentives to suppress all forms of wrongdo-ing.This logic is similar to the finding ofKung andChen(2011), that provincial leaders with stronger incentivesfor career advancement weremore likely to falsify agri-cultural procurement data, which led to higher ratesof mortality during China’s Great Leap Famine.Alongthe same lines, upper-level leaders who want to remainin office may be only interested in monitoring to detectmajor scandals, while upper-level leaders actively pur-suing advancement may have incentives to detect andhead off all potential scandals.Second, in terms of regional conditions, officials who
govern regions whose social, economic, or political sit-uation is more precarious—for example, a region witha history of protest or corruption—may have strongerincentives to hide wrongdoing from superiors as wellas incentives to monitor the activities of their subor-dinates than officials in regions on more solid foot-ing. In regions with more precarious conditions, scan-dals are more likely to jeopardize officials’ chances ofadvancement.Given career ambition and regional conditions, we
expect some officials to be more determined to hidetheir wrongdoing than others, and some officials to bemore determined to monitor their subordinates thanothers. Those more determined to hide wrongdoing
3
Dow
nloa
ded
from
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e. S
tanf
ord
Gra
duat
e Sc
hool
of B
usin
ess,
on
14 Ju
n 20
18 a
t 17:
19:3
4, s
ubje
ct to
the
Cam
brid
ge C
ore
term
s of
use
, ava
ilabl
e at
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e/te
rms.
htt
ps://
doi.o
rg/1
0.10
17/S
0003
0554
1800
0205
Jennifer Pan and Kaiping Chen
from superiors have greater incentives to collude withmonitoring agencies in suppressing information, andthose more determined to monitor subordinates aremore likely to adopt mechanisms to counteract infor-mation manipulation by subordinates.
Variation in Resources. The resources available to of-ficials for hidingwrongdoing and obtaining accurate in-formation also varies. By resources, we mean assets atthe disposal of officials that can be used to exert con-trol over information and monitoring agencies. Theseassets include the authority to hire and promote, socialrelationship, and fiscal resources. Some officials will bebetter equipped to control monitoring agencies (andpreventmonitors from reportingmalfeasance) becauseinstitutional structures give officials control over thecareer prospects of monitors. Other officials can exertcontrol because social relationships such as kinship tiesallow officials to wield influence over monitors. Finally,officials with plentiful fiscal resources aremore likely toovercome information manipulation by lower levels—for example, they can pay monitors more so monitorsare less likely to be bribed or they can pay to havemoremonitors / monitoring systems.
Interaction between Incentives and Resources in Mon-itoring. Officials’ incentives to monitor interact withthe availability of resources for monitoring. For exam-ple, an ambitious upper-level official in a region withplentiful resources will likely put stronger monitoringmechanisms in place to overcome information manip-ulation by lower-level officials than an upper-level of-ficial in a region with plentiful resources who simplywants to remain in office. In general, if plentiful re-sources are available, monitoring may vary as a func-tion of incentives,with stronger monitoring when thereare stronger incentives.What happens if resources are limited or unavail-
able? If an upper-level official does not have the re-sources to strengthen monitoring, we might expectupper-level officials to take one of two strategies. First,the upper-level official may choose not to changeor bolster information collection procedures, knowingthat lower-level officials are likely hiding informationbut not knowingwhat is being hidden. If scandal breaksout, the upper-level official can plead innocence, de-flect blame onto the lower level for concealing infor-mation, and hope that these excuses can overcome thefallout of the scandal.8 Second, the upper-level officialmight seek to collude with lower-level officials in cor-rupt, rent-seeking activity, and collude with lower-levelofficials to suppress information.
Interaction between Superiors and Subordinates. Thedegree to which upper-level officials can monitor andthe degree to which lower-level officials can hidewrongdoing results from the interaction between theincentives and resources of upper-level officials and the
8 The upper-level official runs the risk that claims of innocence areinsufficient if the scandal is severe,which is why, if resources are avail-able, the official would choose to strengthen monitoring and avoidbeing blindsided.
incentives and resources of lower-level officials. Whenupper-level officials have plentiful resources for mon-itoring, whether we observe lower-level officials con-cealing wrongdoing depends on the relative incentivesand resources between superiors and subordinates. Forinstance, if lower-level officials have relatively strongerincentives and more resources than their superiors, weare more likely to see information manipulation andconcealment of wrongdoing by lower-level officials. Ifupper-level officials have relatively stronger incentivesand more resources than their subordinates, we areless likely to see concealment of wrongdoing by lower-level officials. When upper-level officials have limitedresources, we will likely observe lower-level officialsconcealing wrongdoing. In some of these cases, upper-level officials may suspect information manipulation istaking place but be unable to stop it, and in other cases,upper-level officials may be complicit in the informa-tion manipulation. Altogether, this suggests that un-less upper-level officials have stronger incentives andgreater resources than lower-level officials, informationmanipulation by lower levels will take place.
Information Gathering and Manipulation inChina
In this section, we describe how the Chinese regimeaims to use public participation via citizen complaintsto monitor lower-tier officials. We then show how thedynamics of obfuscation and monitoring we have de-scribed manifest in China and why information manip-ulation can persist despite publicly voiced grievances.
How China’s Propaganda Department Monitors Offi-cials with Public Complaints. China has devoted sub-stantial resources to monitoring the performance oflower-tier officials. The CCP has adopted limited ver-sions of elections and free media, while embracingcitizen complaints as an essential part of the CCP’sstrategy of monitoring lower-tier officials.9 There aremany channels for individuals to share their grievancesand complaints: the Bureau of Letters and Visits (���) where citizens can complain in person (Chen2009; Dimitrov 2014a), telephone hotlines (Economist2017), government-managed websites where citizenscan complain online (Chen, Pan, and Xu 2016; Distel-horst and Hou 2017), as well as web and mobile appsdesigned for individuals to complain to the govern-ment (China Daily 2015). The volume of citizen com-plaints, especially those submitted online, has skyrock-eted (Jiang,Meng, and Zhang 2016).The CCP propaganda department (���) is the
primary agency responsible for monitoring publicopinion and sentiment through citizen complaints(Central Propaganda Department 2009; National
9 The CCP has implemented elections, but only at the village andneighborhood levels (O’Brien and Li 2000). China has legislative in-stitutions, but delegates are not freely selected (Manion 2016; Truex2016). Chinese media is commercialized, and some forms of inves-tigative journalism are allowed (Lorentzen 2013; Stockmann 2013),but media outlets remain tightly state-controlled (Qin, Strömberg,and Wu 2016).
4
Dow
nloa
ded
from
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e. S
tanf
ord
Gra
duat
e Sc
hool
of B
usin
ess,
on
14 Ju
n 20
18 a
t 17:
19:3
4, s
ubje
ct to
the
Cam
brid
ge C
ore
term
s of
use
, ava
ilabl
e at
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e/te
rms.
htt
ps://
doi.o
rg/1
0.10
17/S
0003
0554
1800
0205
Concealing Corruption
Grassroots Party Work Key Textbook 2013).10 Pro-paganda departments exist at the central, provincial,prefecture, and county levels. Each level is responsiblefor monitoring public sentiment in its geographicjurisdiction, including public reports of corruption.11The propaganda department is also responsible forreporting this information upward to the propagandadepartments at the next level up in what is called“level-by-level” reporting so that the CCP and Chinesegovernment can use this information to help guide pol-icy and political decisions (Cai 2000; Central Commit-tee of theCommunist Party of China 2016;Huang 1995;Oi 1995). Evaluation metrics contained in internaldocuments such as the Propaganda and Thought WorkManagement Evaluation Form (�������������) reveal the importance of monitoring workfor the propaganda department.Approximately half ofthe areas on which CCP propaganda departments areevaluated relate to monitoring public sentiment andreporting this information to upper-level superiors.12Specific evaluation criteria include the number of citi-zen complaints collected and analyzed,13 the number ofbursts of positive and negative online discussions iden-tified, and whether there is regular reporting of infor-mation to upper levels of the government and party. Incontrast, it is not within the purview of the propagandadepartment to validate the veracity of citizen com-plaints, nor is it within its jurisdiction to discipline in-dividuals or organizations based on public complaints.The task of the propaganda department is to determinewhat key issues and/or events generate interest andattention among the public, and to report this informa-tion upward (Central Propaganda Department 2009).
10 Government bureaus and offices (e.g., bureau of letters and visits,public security bureau), and China’s Cyberspace Administration (���) also deal with citizen complaints.However, instead of monitor-ing,government bureaus are taskedwith collecting, investigating,andultimately resolving complaints related to their areas of work in theirgeographic jurisdictions.China’s CyberspaceAdministration (CAC),also known at the central level as the Office of the Central LeadingGroup for Cyberspace Affairs, is tasked with governing the internet.CAC, located at central and provincial levels of government, regu-lates the activities of internet content providers, from news portalsto social media platforms, and penalizes companies and individualsthat fail to complywith government regulations (CyberspaceAdmin-istration of China 2017).11 The central role of the propaganda department inmonitoring pub-lic opinion has been echoed by central as well as provincial lead-ers. Provincial party secretaries in provinces ranging from Henanto Guangdong have emphasized the central role of the propa-ganda department in monitoring public opinion (see http://bit.ly/2eJUPlw, http://bit.ly/2wM8W1U, accessed Sept. 5, 2017). Li Xi, for-mer deputy secretary of Shanghai and current party secretary ofLiaoning Province, wrote an editorial in People’s Daily, the officialnewspaper of theChineseCommunist Party, that provincial decisionsshould take into account public opinion information collected andanalyzed by the propaganda department (see http://bit.ly/2eK1M6f,accessed Sept. 5, 2017).12 The other half of the functions relate to cultural development andideological thought work.13 Documents in the leaked emails show that the provincial propa-ganda department explicitly tasks the prefecture propaganda depart-ment with monitoring complaints related to their jurisdiction acrosssocial media sites and internet platforms, including national-levelplatforms the prefecture does not manage.
Institutional Incentives for Concealing Corruption.China is a single-party authoritarian regime with fivelevels of state administration: central, provincial, pre-fectural, county, and township. We use the term “topleaders” to refer to leaders at each level of admin-istration who hold senior CCP positions—specificallythat of party secretary, vice party secretary, and polit-buro standing committee member.14 Individuals hold-ing these top party positions also typically occupykey government positions, so that crucial positions ofpower are all under CCP control.At each level of administration, top leaders de-
termine the advancement of top leaders at the nextlevel down, in what is called one-level-down manage-ment (����) (O’Brien and Li 1999), by evalu-ating them against performance targets (Edin 2003;Whiting 2004). Points are assigned to various dimen-sions of performance, but failure to achieve “veto” tar-gets, such as failure to ensure social stability—by pre-venting collective action or rooting out corruption—jeopardizes advancement prospects even if the perfor-mance is strong in other areas.15 This means that topleaders at each subnational level in China have incen-tives to hide their own wrongdoing from superiors onelevel upwho control their career prospects.At the sametime, top leaders at each level have incentives to ob-tain enough information about their subordinates onelevel down so the behavior or malfeasance of thesesubordinates does not jeopardize their own careerprospects.Opportunities for Information Manipulation. Topleaders have opportunities to conceal malfeasancefrom superiors one level up because they can exertcontrol over the propaganda department that moni-tors them and because of China’s level-by-level report-ing structure. Top leaders in China can exert controlthrough institutional incentives as well as social ties tomotivate collusion with the propaganda department.In terms of institutional incentives, the promotion ofheads of various bureaucracies at each level (e.g., theprefecture propaganda department) is determined bythe top leaders of that same level (e.g., prefecture partysecretary) rather than the bureau one level up (e.g.,provincial propaganda department) (Edin 2003; Kou
14 Note that when we refer to “top leaders,” we are referring to in-dividuals, at each level of administration, who hold these key partypositions. For the case of China, the upper-level and lower-level offi-cials we described in the theory section are all top leaders.15 Based on internal documents of criteria for political advancementfrom several central and eastern provinces, we find that, althoughthe overall weight given to economic development (e.g., 100 out of350 points) and social stability (75 out of 350 points) are highest,these broad categories are broken down into smaller subactivities—for example, social stability includes managing petitions (5 points)and strengthening public security (18 points) while economic devel-opment includes GDP growth (10 points) and service sector growth(10 points)—the subcategory with the highest point value is “con-struction of clean government” (������) worth 30 points.Thissubcategory refers to party discipline and the need for clean, non-corrupt government, which has been documented to be a veto target(see http://bit.ly/2xMvIXs and http://bit.ly/2j1IHl9 (accessed Sept. 3,2017).
5
Dow
nloa
ded
from
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e. S
tanf
ord
Gra
duat
e Sc
hool
of B
usin
ess,
on
14 Ju
n 20
18 a
t 17:
19:3
4, s
ubje
ct to
the
Cam
brid
ge C
ore
term
s of
use
, ava
ilabl
e at
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e/te
rms.
htt
ps://
doi.o
rg/1
0.10
17/S
0003
0554
1800
0205
Jennifer Pan and Kaiping Chen
and Tsai 2014;Whiting 2004; Zhan and Yu 2011).16 Asa result, although Chinese bureaucracies are part of amatrix organization (����) where bureaucraciesreport to top leaders at the same level and also reportto their functional equivalents one level up, satisfyingthe interests of top leaders at the same level takes pri-ority over the interests of functional superiors at thenext level up (Li 2012;Lieberthal andLampton 1992).17In addition, social relations between top leaders andpropaganda officials at the same level of administra-tion can motivate monitors to collude with top leaders.For example, the prefectural head of propaganda andthe prefecture politburo members are all elites withinthe prefecture andmay know each other through work,school, or even kinship ties.Information manipulation occurs when local propa-
ganda departments collude with top leaders becausethe local propaganda department can withhold infor-mation from superiors. For example, if a prefecturepropaganda department identifies online complaintsof corruption pertaining to prefecture officials, in orderfor provincial authorities to obtain this information,the information must be transmitted by the prefecturepropaganda department to the provincial propagandadepartment, and from the provincial propagandadepartment to provincial top leaders. In other words,since the propaganda department system for informa-tion gathering system relies on level-by-level reporting,there are opportunities for each subnational level toconceal information.Propaganda departments can be penalized for fail-
ing in their task of monitoring by CCP discipline in-spection commissions,which are the party organizationtaskedwith enforcing internal party rules.We can thinkof these discipline inspection commissions as a thirdparty operating outside the normal procedures for ad-vancement. However, collusion between top leadersand monitors may persist because the risk of penaltyis relatively lowwhen informationmanipulation of citi-zen complaints occurs through concealment rather thanfalsification. If propaganda departments regularly re-port public opinion upward but do not reveal the fullpicture, propaganda departments are fulfilling the let-ter, but not spirit, of monitoring regulations, and pun-ishment may be less likely relative to other forms ofinformation manipulation such as falsification of eco-nomic statistics.
Opportunities to Strengthen Monitoring. Since theshortcomings of level-by-level reporting are well
16 This differs from the promotion of top leaders at each level, whichis controlled by top leaders one level up.17 The only situation in which bureaucratic relationships to top lead-ers is subservient to the relationship with functional superiors onelevel up is when the bureaucracy is under vertical management (����). This arrangement applies to a handful of bureaucracies thatdeal with cross-regional issues, including the railway administration,civil aviation administration, customs administration, state security,land and resources administration, taxation, finance, and bureau ofstatistics. The propaganda department is not a bureaucracy undervertical management. The career advancement prospects of propa-ganda department officials are determined by top leaders at the samelevel, not by propaganda officials one level up.
known, China’s central authorities have worked toincrease the number of channels for citizen com-plaints and by directly gathering information in spe-cific situations—for example, making surprise visits tolower-level governments and creating mobile apps forcitizens to report corruption to the central level (Gao2016; Xinhua 2012). National-level social media plat-forms are a particularly fruitful channel for obtaininginformation because subnational levels of governmentdo not have the ability to censor these platforms di-rectly. As long as officials have access to human and/orcomputational resources, large-scale social media textdata can be mined for information on the activities oflower-level officials.Indeed, in recent years, China’s central authorities
have dramatically increased monitoring of the inter-net.Qin, Stromberg, andWu (2017a) find that the postson Sina Weibo mentioning the names of officials andcorruption predict charges of corruption by the cen-tral regime. However, this does not mean the centralregime, which has the most human and computationalresources of any level of government, is able to uncoverall complaints of lower-level corruption posted to so-cial media. Indeed, China’s central government is stillin the process of building a national-level surveillancesystem that can extract information from various socialmedia platforms, and this system is only intended toreach down to the prefecture level (leaving out county,township,and lower-level government offices and partyorganizations).18The capacity for monitoring social media using com-
putational methods is weaker among provincial andlower-level governments (Mai and Liber 2015). Whilesubnational governments can devote human resourcesto monitoring social media,19 whether they do so likelydepends on their incentives and availability of re-sources. At present, subnational governments remainprimarily reliant on internal, level-by-level reportingto identify lower-level corruption (Hsu, Zhao, and Wu2011). This suggests that while the Chinese regime hasinvested in public channels for expressing grievances toimprove monitoring of lower-tier officials, informationproblems persist, especially between subnational levelsof government.In sum, China’s central regime has embraced public
complaints as a core component of information gath-ering. At subnational levels, top leaders have oppor-tunities to manipulate information through control oflocal propaganda departments and they also have op-portunities to strengthen monitoring. In the followingsection, we describe our unique dataset of communi-cations between the J. Prefecture propaganda depart-ment and its provincial superiors, which allows us totest some aspects of our theory. An observable im-plication of the theory is that unless upper-level of-ficials have stronger incentives and greater resources
18 The central government aims to complete this system by 2020;information based on interviews with individuals working with thegovernment on computational surveillance.19 Qin, Stromberg, and Wu (2017b) estimate that it would take 2,080person-hours to identify the strikes over 3 years from the relatedposts on Sina Weibo.
6
Dow
nloa
ded
from
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e. S
tanf
ord
Gra
duat
e Sc
hool
of B
usin
ess,
on
14 Ju
n 20
18 a
t 17:
19:3
4, s
ubje
ct to
the
Cam
brid
ge C
ore
term
s of
use
, ava
ilabl
e at
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e/te
rms.
htt
ps://
doi.o
rg/1
0.10
17/S
0003
0554
1800
0205
Concealing Corruption
than lower-level officials, information manipulation bylower levels will take place. As we will see, in the caseof J. Prefecture, provincial and prefectural levels havesimilar levels of resources,but prefecture officials likelyhave stronger incentives for information manipulationthan provincial officials have for information gathering.We would thus expect to see information manipulationby J. Prefecture, and our data allows us to ascertainwhether this is the case.
LEAKED PROPAGANDA DEPARTMENTCOMMUNICATIONS
A problem with existing studies of the effectivenessof gathering information through citizen complaints isthat they rely on data found at specific levels of gov-ernment, for example, central government archives orcity government reports.As a result,we never see whatinformation is concealed by lower levels of governmentand not transmitted upward.This paper changes the situation. In December 2014,
an email archive from the Propaganda Department(���) of J. Prefecture was publicly leaked (Heno-chowicz 2014; Sonnad 2014). The email archive con-tains Online Sentiment Monitoring Reports producedby the J. Prefecture Propaganda Department, as wellas many other communications to and from the de-partment. The leak was reported and the archive ofemails remains publicly available (Henochowicz 2014;Sonnad 2014). The email archive is large and compli-cated by multiple email storage formats, diverse doc-ument types, numerous attachments, and many linksto outside information. Because of the complexity ofthis data, no systematic analysis has been conducted.To systematize this rich data source, we developed andapplied a variety of methods and procedures, fromlarge scale hand coding, to specially tuned and adaptedmethods of named entity recognition, and methods ofautomated text analysis and extraction. Because of theconsiderable effort entailed, the extracted data, as wellas replication materials for the analyses in this paperare available, see Pan and Chen (2018).We identified 2,768 emails sent to and from the J.
Prefecture Propaganda Department between May 14,2012 and December 6, 2014. Figure 1 shows the num-ber of emails in the archive by month between 2012and 2014. There are substantially more emails in thearchive in 2014 than in the previous two years.20 Wedo not know if this is because there are missing emailsfrom the previous years, or if it is because the emailaccount was used more actively in 2014. We focus ouranalysis on the time period when the data is mostplentiful, between January 1, 2014 and November 30,2014.21
20 The large number of emails in April 2014 is largely due to dupli-cation, which may be related to how the data was obtained. For thisreason, we remove exact duplicates in our analysis.21 We do not include December since the archive was leaked in earlyDecember.When the entire dataset is used, none of our substantiveresults change.
FIGURE 1. Number of emails in archive bymonth.
May Oct Mar Aug Jan Jun Nov
010
020
030
040
0
2012 2013 2014
FIGURE 2. Network structure of emails.Circles are email correspondents, linesindicate email correspondence.
The overall structure of communications capturedby these emails is shown with the network graph inFigure 2. Each circle is a specific email account andeach line denotes where one or more emails was sentfrom and to. The most central node is the J. Prefec-ture Propaganda Department, and the accounts it cor-responds most frequently with include the J. Prefecturegovernment (���) and the provincial propagandadepartment (����).
7
Dow
nloa
ded
from
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e. S
tanf
ord
Gra
duat
e Sc
hool
of B
usin
ess,
on
14 Ju
n 20
18 a
t 17:
19:3
4, s
ubje
ct to
the
Cam
brid
ge C
ore
term
s of
use
, ava
ilabl
e at
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e/te
rms.
htt
ps://
doi.o
rg/1
0.10
17/S
0003
0554
1800
0205
Jennifer Pan and Kaiping Chen
TABLE 1. Three Types of J. PrefectureOnline Sentiment Monitoring Reports
2012 to 2014 Jan. to Nov. 2014
Daily reports 514 249Weekly reports 114 55Bi-monthly reports 15 15Total 643 319
Among all emails, 643 contain Online SentimentMonitoring Reports produced by J. Prefecture.22 Be-tween January and November 2014, 319 emails containthese Online SentimentMonitoring Reports.There arethree types of Online Sentiment Monitoring Reports:Daily,Weekly, and Bi-monthly, with Bi-monthly reportonly appearing in 2014 (see Table 1). All reports con-tain three main sections: (a) review of current publicsentiment, (b) trending news items, and (c) nationalonline trends. Most relevant to our analysis is the firstsection, which details issues that have generated localonline discontent. Section (b) on trending news itemsincludes keywords of trending topics,often overlappingwith the more detailed analysis of section (a). Section(c) is a brief summary of national-level top news items,primarily from traditional news sources.Section (a) is further divided into four subsections:
(a1) the time period covered by the monitoring report(daily reports can cover a range of 1–4 days), (a2) thenumber of online posts analyzed for the report (e.g.,185 posts), (a3) an evaluation of the overall tendency ofpublic sentiment: negative (��), neutral (��), pos-itive (��), and (a4) a summary of key issues includ-ing descriptions of the topics that have generated neg-ative sentiment. The overall assessment of sentimentis almost always positive.23 Thus, we focus on the ac-tual posts classified as negative from section (a4) of thereport. Typically, the issue is described in a sentence ortwo, and a link to the original post and discussion issometimes included.In total, we extract 3,423 negative sentiment issues
from the Online Sentiment Monitoring Reports; af-ter removing exact duplicates, 2,879 remain.24 BetweenJanuary and November 2014, there are 1,925 negativesentiment issues, and after removing exact duplicates,1,412 issues remain. We examine the 1,412 nondupli-cate issues more closely and identify 1,038 unique com-plaints.25 In addition to using the Propaganda Depart-
22 There is a total of 653 Online Sentiment Monitoring Reports, tenof these reports are not produced by J. Prefecture, but instead pro-duced by subordinate counties.23 Only 20 sentiment monitoring reports among all reports classifythe overall tendency of public sentiment as negative, and in all ofthese cases, the negative assessment is qualified and described as“tending toward the negative” (����) or “neutral to negative”(�����).24 Some issues are highlighted in multiple reports because they per-sist over time.We eliminate exact duplicates for our main analysis. Ifduplicates are retained, the substantive results remain unchanged.25 The 1,412 issues do not contain exact duplicates but contain somecomplaints that discuss the same issue, perhaps using different words
ment’s terminology of “negative sentiment issues,” wealso refer to these posts as citizen complaints.In the process of extracting data from the email
archives, we noticed that a subset of Online SentimentMonitoring Reports was addressed to higher-level of-ficials, while others were kept for internal circulationwithin the prefecture propaganda department. We re-alized that we had data on all of the negative senti-ment issues the J. Prefecture Propaganda Departmentdeemed to be important, and the subset which they re-ported to upper-level officials.Reports sent to provincial superiors meet three con-
ditions, which are always met simultaneously. First, us-ing email metadata, we find that the report is emailedto the provincial propaganda department email ad-dress. Second, the subject of the email requests the at-tention of provincial officials, e.g., “Provincial LeadersPlease Inspect” (������). Third, the report itselfcontains the line “cc: City Party Secretary, City ViceParty Secretary,City People’s Congress Chairman,CityPeople’s Political Consultative Conference Chairman,City Politburo Standing Committee, City Vice Mayor,County Propaganda Departments, other relevant citydepartments.”Among the 1,412 unique negative senti-ment issues identified by the J. Prefecture PropagandaDepartment from January to November 2014, 590 aresent to upper-level leaders.
Observable Implications of Theory in J.Prefecture
This leaked data comes from one prefecture in China.Is this prefecture more or less likely to manipulateinformation than other prefectures? Is the provincewhere J. Prefecture is located more or less likely touncover manipulation by the prefecture? As discussedin the previous Section, information manipulation re-sults from the strategic interaction between superiorand subordinate levels of government, which in turndepends on the incentives and resources available toeach level of government. Here, we discuss the observ-able implications of the theory for J. Prefecture by de-scribing the incentives and resources of J. Prefectureleaders and their provincial superiors. In doing so, weset the scope conditions of our empirical analysis.Officials in J. Prefecture may have stronger incen-
tives for career advancement than their provincialcounterparts, which would result in stronger incen-tives for information manipulation at the prefecturelevel than incentives for information gathering at theprovincial level. There are two aspects that influenceincentives—regional conditions and career ambition.In terms of regional conditions, neither J. Prefecturenor its superior province has high levels of protest or a
or by citing difference sources.The 1,038 unique complaints are com-plaints that refer to different issues, in different locations, at differentpoints in time. We use 1,412 as the unit of analysis in the paper, butsubstantive results remain unchanged if we use the 1,038 unique com-plaints (see the Supplemental Appendix).
8
Dow
nloa
ded
from
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e. S
tanf
ord
Gra
duat
e Sc
hool
of B
usin
ess,
on
14 Ju
n 20
18 a
t 17:
19:3
4, s
ubje
ct to
the
Cam
brid
ge C
ore
term
s of
use
, ava
ilabl
e at
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e/te
rms.
htt
ps://
doi.o
rg/1
0.10
17/S
0003
0554
1800
0205
Concealing Corruption
recent history of corruption scandals relative to otherprefectures and provinces in China.26In terms of career ambition, we examine the party
position of J. Prefecture and provincial leaders follow-ing Kung and Chen (2011). Provincial leaders holdmore senior party positions than prefecture leaders sowe cannot directly compare their positions. Rather, wecompare the actual party position of the leaders rel-ative to positions leaders at comparable levels couldhold in the CCP. For provincial leaders, we examinewhether the party secretary and governor are alternateor full members of the CCP Central Committee—ifboth are full members, we label the province as “lessambitious” because leaders have already attained aCCP position that is senior for their level. If either isan alternate Central Committee member, we label theprovince as “more ambitious” because one of its topleaders could attain a party position of greater senior-ity.For prefecture leaders,we examinewhether the pre-fecture party secretary and mayor are in the provincialparty committee or provincial politburo—if either arein the provincial politburo, we label the prefecture as“less ambitious” because a prefecture leader has al-ready attained a party position that is senior for offi-cials at the prefecture level. If either the party secretaryor mayor are in the provincial party committee but nei-ther have attained the provincial politburo,we label theprefecture as “more ambitious” because one of its topleaders could attain a party position with greater se-niority. Both the provincial party secretary and gover-nor of J. Prefecture’s superior province were full mem-bers of the CCP Central Committee as of 2014. In con-trast, the prefecture party secretary and mayor of J.Prefecture were only members of the provincial partycommittee as of 2014. This suggests that J. Prefecturetop leaders faced stronger career ambitions than theirprovincial superiors. In 2016, the J. Prefecture partysecretary was elevated to the provincial politburo,which further suggests that, as of 2014, activities in J.Prefecture were motivated by incentives for politicaladvancement.How does the arrangement of career ambition we
observe in J. Prefecture compare to other Chinese pre-fectures? We randomly sampled two prefectures fromevery province in China,27 and we collected the bi-ographies of the provincial party secretary, provincialgovernor, prefecture party secretary, and prefecturemayor from these prefectures in 2014. Among 54 pre-fectures, 45 (83 percent) had the same arrangement as
26 For information on protest, we used data from http://bit.ly/1MOQzdz and http://bit.ly/1nk2Azf to calculate protests at theprefectural and provincial levels (accessed Dec 15, 2016). For infor-mation on corruption, we used data from Procuratorial Daily (����) between January 1, 2013 and October 11, 2014. Among 825reports of officials convicted of corruption, only one official from J.Prefecture was mentioned, and only 43 officials from the provincewere named, among them none were top provincial leaders.Relativeto other prefectures and provinces, the rate of conviction for corrup-tion is low in J. Prefecture and the upper-level province.27 We do not include the municipalities of Beijing, Shanghai, Tianjin,and Chongqing since their levels of administration differ from otherprovinces. Special administrative regions of China such as HongKong are also excluded.
J. Prefecture—provincial top leaders were full CentralCommittee members, and prefecture leaders were onlyprovincial party committee members.As of early 2018,in 11 prefectures (20 percent), at least one top pre-fectural leader has attained the provincial politburo.We do not know how many other prefectural leaderstried but failed to reach the provincial politburo, butthese estimates suggest that somewhere between 20 to83 percent of Chinese prefectures have a similar ar-rangement of career incentives as J. Prefecture and itsprovince.We do not observe substantial difference between J.
Prefecture and its superior province in resources. Topleaders in J. Prefecture, as in all other prefectures inChina, exert institutional control over the prefecturepropaganda department. In addition, J. Prefecture isone where resources are relatively plentiful.Comparedwith the other 10 prefectures in its province, J. Prefec-ture was third in terms of economic production in 2013and was in the middle of the pack in terms of fiscalrevenue. Compared to all other Chinese prefectures,J. Prefecture appears in the middle of the pack acrossa number of indicators, including GDP, foreign trade,and consumer demand (for details see SupplementalAppendix). The province to which J. Prefecturebelongs also has resources to strengthen monitoring.The province has been consistently ranked in themiddle of China’s provinces in terms of fiscal revenue.In sum, when we compare J. Prefecture and its
province, while there are no clear differences in termsof resource capacity or regional conditions, the incen-tives of J. Prefecture officials for career advancementsare likely stronger than those of provincial superiors.As a result, we would expect to observe informationmanipulation by J. Prefecture based on our theory. It isimportant to note that neither the province nor J. Pre-fecture represents an extreme case in terms of incen-tives or resources. Many other prefectures exhibit thesame pattern of career incentives as J. Prefecture, andJ. Prefecture’s control of the prefecture propagandadepartment is a feature of China’s administrativestructure.
Limitations
This dataset is unique because it provides a view of thetransfer of information between different levels of gov-ernment; however, it faces two main limitations. Thefirst limitation relates to the veracity of the data. Theinference we make in this paper depends on the verac-ity of the leaked email archive we analyzed. While wecannot know for certain whether the data are genuine,the size and extraordinary complexity of this archivemake it highly unlikely to be fake. In addition, thereare no signs the archive was generated by automatedmeans.The second limitation of this dataset is that it does
not represent the full set of communications betweenJ. Prefecture and its superiors. Although a wide rangeof issues, including topics marked for internal con-sumption, appear in this email archive, we do not have
9
Dow
nloa
ded
from
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e. S
tanf
ord
Gra
duat
e Sc
hool
of B
usin
ess,
on
14 Ju
n 20
18 a
t 17:
19:3
4, s
ubje
ct to
the
Cam
brid
ge C
ore
term
s of
use
, ava
ilabl
e at
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e/te
rms.
htt
ps://
doi.o
rg/1
0.10
17/S
0003
0554
1800
0205
Jennifer Pan and Kaiping Chen
records of in-person conversations, phone calls, textmessages, or the transmission of information in otherbureaucracies between the prefecture and upper lev-els of government. This raises the concern that, even ifinformation transmitted by email between the prefec-ture propaganda department and upper levels of gov-ernment is biased, upper-level superiors could have anaccurate picture of the activities and performance of J.Prefecture.Upon more careful consideration, however, the in-
completeness of our data presents a hard test for find-ing systematic concealment of wrongdoing. If upper-level authorities have a full and accurate picture of theperformance of J. Prefecture, then it would be riskierfor the propaganda department of J. Prefecture to ma-nipulate and suppress information because the threatof discovery increases. Because there are other chan-nels of communication between the prefecture and up-per levels of government, we should be less likely tofind information manipulation in one channel. If wedo observe information manipulation in these propa-ganda department emails, it suggests provincial author-ities lack a full picture of the activities of lower-levelofficials.
Legal and Ethical Considerations
The ethics of conducting research using leaked datahas been a subject of intense discussion, especiallyas leaked data has become ever more important andprevalent in the digital era (Michael 2014). Here, wediscuss the legal and ethical implications of workingwith this data.From a legal perspective, we do not know how the
data was originally obtained. However, the method bywhich we are obtaining the data—downloading froma publicly available website after learning of the datafrom journalists—is legal.Ethically, the phenomenon we are studying with this
data pertains to government institutions. The emailsthat we analyze belong to government offices andbureaucracies where more than one individual maybe sending and receiving correspondence. We do notknow the identities of any individual involved, and wedo not collect any personal, identifying informationfrom the leaked email archive. Nevertheless, we ob-tained approval from our university Institutional Re-view Board for analyzing this data.Finally, although the names of localities are available
in the publicly available email archive, we do not in-clude any place names in this paper since names arenot germane to our central arguments. Our paper isnot intended as a report on the activities of any par-ticular local government in China but rather a studyto improve our understanding of information controlsunder authoritarian rule.
BIASED UPWARD REPORTING
In this section, we show how citizen complaints re-ported upward by J. Prefecture is biased to con-
ceal complaints that implicate the prefecture ofwrongdoing—of breaking the law and engaging incorrupt practices—as well as to conceal complaintsof wrongdoing, implicating counties where prefectureleaders have birthplace or workplace ties. We first usean unsupervised method of text analysis and closereading of the text to explore the citizen complaintsdescribed in the J. Prefecture Online Sentiment Moni-toring Reports. Second, we hand code all unique com-plaints identified in the Online Sentiment MonitoringReports to determine whether complaints of wrongdo-ing by the government are less likely to be reportedto upper-level superiors, while accounting for alter-native explanations. Finally, we explore the role ofcensorship in concealing complaints from upper-levelauthorities.
Content of Citizen Complaints
We use a structural topic model (STM) to gain a betterunderstanding of the types of citizen complaints high-lighted by the J. Prefecture Propaganda Department inits Online SentimentMonitoring Reports, and whetherthere are differences among topics in terms of upwardreporting (Lucas et al. 2015; Roberts et al. 2013, 2014).We are interested in examining whether a complaintwas reported to provincial superiors, and we want tounderstand the relationship between this covariate andthe content of complaints. By using an STM, we allowwhether a complaint is reported upward to affect theproportion of issues focused on a topic and the distri-bution of words that characterize a topic.To determine the number of topics, we compare the
held-out likelihood, residual, semantic coherence ofmodels with 25 to 80 topics. The model with 40 top-ics yielded the most intuitive results and forms thefocus of subsequent analysis (Chang et al. 2009). Wehand labeled each topic by reading documents asso-ciated with the topic, by examining the words thatappear with highest probability in that topic, and byexamining the words that are frequent and exclusiveto that topic. We were able to hand label 30 of thetopics.28Among the 30 topics we could label, half deal
with complaints of government corruption andmalfea-sance.29 Complaints accuse government of embezzle-ment and misuse of public funds, for example:
28 A coherent topic was not easily discernible for the remaining 10topics.29 These topics include Y. county gov’t misconduct, P. county gov’tmalfeasance in land development, D. county gov’t malfeasance inland development, Complaints about J. Prefecture governance, Pre-fecture gov’t malfeasance in public welfare, J. county gov’t malfea-sance in land seizures, Complaints about gov’t misconduct and cor-ruption,Embezzlement by civil servants,Gov’t malfeasance in publicgoods provision, Collection of illegal fees, D. county gov’t malfea-sance, County gov’t violence and misconduct, County gov’t embez-zlement in land, D. county gov’t corruption and malfeasance, X.county gov’t illegal land development and violence. The estimatedtopic proportion of all labeled corruption-related topics totals 39.05percent.
10
Dow
nloa
ded
from
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e. S
tanf
ord
Gra
duat
e Sc
hool
of B
usin
ess,
on
14 Ju
n 20
18 a
t 17:
19:3
4, s
ubje
ct to
the
Cam
brid
ge C
ore
term
s of
use
, ava
ilabl
e at
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e/te
rms.
htt
ps://
doi.o
rg/1
0.10
17/S
0003
0554
1800
0205
Concealing Corruption
“Complaint from village doctors: relevant laws allow 50 to60 year old village doctors to participate in skill-based se-lection examination free of charge.However, theX.countyhealth department took hard earned money from 98 ruraldoctors by charging them 250 yuan per person for takingthe exam.” (X. county ������,����������������������������������,� X. county��������������������,����250�)
Several topics deal with rent seeking in the form of il-legal land development and land taking, for example:
“Online complaints claim that the P. county H. village gov-ernment spent four years using various methods such asdeception and force to expropriate tens of acres of forestand farm land from villagers. And for four years, villagerssaw that these forcibly taken lands were unused. Believ-ing the land was greatly wasted, villagers planted crops onthem. However, the seedlings were secretly destroyed byvillage officials based on instructions from the townshipgovernment officials.”(��, P. county H. village�������,������������,����������� ������������������������ ���������,�����,�����,��������������������)
The remaining half of topics we could label dealwith governance issues, such as complaints of pollution,complaints about inadequate public goods provision(roads, schools, health care access, utilities), complaintsabout student safety and welfare issues, as well as eco-nomic disputes between the public and businesses.30An example of a governance-related complaintwrites:
“The garbage transfer station in P. county YM road is ex-tremely smelly and nearby residents are miserable.Onlinecomplaints reported that although P. county is building ahealth-friendly city, the garbage transfer station near YMlake park is noxious all day long...internet users have com-plained repeatedly to the relevant departments, but thesmell persists...hope that the responsible agency can dealwith the problem soon and move the garbage transfer sta-tion elsewhere.” (P. county YM road��������,�������������, P. county��������,�� YM������������...�������������...� �����,���������,����...���������,��������)
Figure 3 shows the effect of upward reporting ontopic prevalence for each of the 30 labeled topics. Thepoint estimate is the mean effect of upward reporting,and the lines are 95% confidence intervals. If the esti-mate and its confidence interval cross the vertical zeroline, thenwhether or not the topicwas reported upwarddoes not affect the proportion of complaints focusedon this particular topic.Estimates above zero are topicsthat are more likely to be reported upward. Estimatesbelow zero are topics that are less likely to be reportedupward.
30 The estimated topic proportion of all labeled governance-relatedtopics totals 33.78 percent.
From Figure 3, we can see that seven topics, mostlyrelated to corruption, are less likely to be reportedupward: Y. county gov’t misconduct, P. county gov’tmalfeasance in land development, “X. county taxidriver strike, D. county gov’t malfeasance in land de-velopment,Complaints about J. Prefecture governance1, Prefecture gov’t malfeasance in public welfare, andJ. county gov’t malfeasance in land seizures. There arefour topics that are more likely to be reported upward,and one relates to corruption: Complaints about taxiservice, Student safety and welfare issues, X. countygov’t illegal land development and violence, and Com-plaints about public goods provision 4. For the remain-ing labeled topics, which deal with a mix of governanceand corruption issues, upward reporting does not havea statistically significant effect on the prevalence of thetopic.These descriptive results suggest that upward report-
ing may be motivated by a desire to protect the in-terests of prefecture-level officials. Two topics criticiz-ing prefecture government performance—Complaintsabout J. Prefecture governance 1 and Prefecture gov-ernment malfeasance in public welfare—are less likelyto be reported upward, while no topics critical of theprefecture government are more likely to be reportedupward.A hazier picture emerges regarding the corruption
of county officials. Figure 3 shows that some topicspertaining to county-level corruption are more likelyreported upward (e.g., X. county gov’t illegal land de-velopment and violence) while other county-relatedtopics with similar content are less likely reported up-ward (e.g., P. county gov’t malfeasance in land develop-ment and J. county gov’t malfeasance in land seizures).When we examine grievances related to county cor-ruption more closely, and we find that complaints ofwrongdoing in some counties are sent up, while com-plaints of wrongdoing in others are not. Below are twogrievances related to X. county, which are reported up-ward,and two grievances related toY.county,which arenot reported upward. In each pair of complaints, onerelates to financial corruption and the other to violenceperpetuated by county officials. These two posts aboutX. county are reported upward:
“Villagers of X. county, S. village reported that the formerparty secretary HCY of X. county, C. township illegallysold acres of reservoir land in the village.” (X. county S.village �������������� X.county, C. township��� HCY������� �)
“X. county’s police station chief ZMHwas reported in on-line complaints to have led station staff and police in usingtorture to extort confessions; victims were not given waterfor around 30 hours and were coerced into confessing thecrimes they were accused of.” (� �� X. county��� ZMH�����������,����������30����,��� �)
These two posts about Y.county are not reported up-ward:
11
Dow
nloa
ded
from
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e. S
tanf
ord
Gra
duat
e Sc
hool
of B
usin
ess,
on
14 Ju
n 20
18 a
t 17:
19:3
4, s
ubje
ct to
the
Cam
brid
ge C
ore
term
s of
use
, ava
ilabl
e at
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e/te
rms.
htt
ps://
doi.o
rg/1
0.10
17/S
0003
0554
1800
0205
Jennifer Pan and Kaiping Chen
FIGURE 3. Effect of upward reporting on topic prevalence, with mean and 95% confidenceintervals.
“Online complaints question whether the health insurancebureau in Y. county misused public funds. When payingfor health insurance at the beginning of 2012, online com-plaints report that the bureau forced them to pay for twoadditional years’ fees, totaling more than 2,700 yuan [perperson].But, the additional fees were not put into the peo-ple’s health spending accounts nor given to employers assubsidies, so health accounts had no funds, and the bureauprovided no receipts for the fees.” (� �� Y. county����, 2012�������, Y. county�������������������, �2700��,�������������������,� �������,�������,�����,��������)
“Online complaints say that around ten city managementpolice in Y. county severely beat citizens. On Feb 27th,while the county development and reform committee washolding a provincial conference on petitions andmaintain-ing social stability, the city management police were beat-ing citizens for their amusement outside the building.” (�� Y. county�����������2�27�,����������������,�����������)
Accusations targeting Y. county are similar to thoseleveled at X. county, yet only the complaints related toX. county are reported by J. Prefecture to the province.Our examination of the content of citizen complaints
reveals that complaints of government corruptionappear frequently among the complaints collected byJ. Prefecture’s Propaganda Department. However, cit-izen complaints reported up to provincial authorities
do not reflect the distribution of topics that J. Prefec-ture has identified by monitoring citizen complaints.Certain topics are reported upward with greater fre-quency, and others are reported upward with lesserfrequency. This means the subset of information re-ported to provincial officials does not simply reduce thescale of information, and the complaints shared withthe upper level do not reflect the overall prevalenceof topics gathered by the J. Prefecture PropagandaDepartment.
Concealing Complaints of Wrongdoing
To systematically test whether upward reporting isskewed to protect the interests of prefecture-level of-ficials, we use logistic regression to estimate the effectof wrongdoing by J. Prefecture and the effect of wrong-doing by counties with patronage ties to J. Prefectureofficials on upward reporting while controlling for al-ternative explanations.Following the theory described in the theory Sec-
tion, we assume that top leaders in J. Prefecture wantto maximize rents, which may include engaging in cor-rupt practices, and want to minimize the risk of beingdisciplined and removed from office by their provincialsuperiors. Prefecture officials can access rents and im-prove their hold on political power by protecting theinterests of those within their patronage networks. AsHillman (2010) notes in his study of local-level factionsin China, patronage networks between local politicalleaders and their subordinates help patrons advance
12
Dow
nloa
ded
from
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e. S
tanf
ord
Gra
duat
e Sc
hool
of B
usin
ess,
on
14 Ju
n 20
18 a
t 17:
19:3
4, s
ubje
ct to
the
Cam
brid
ge C
ore
term
s of
use
, ava
ilabl
e at
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e/te
rms.
htt
ps://
doi.o
rg/1
0.10
17/S
0003
0554
1800
0205
Concealing Corruption
TABLE 2. Upward Reporting for Prefectureand County Wrongdoing
NotReported reportedupward upward Total
Prefecture wrongdoing 17 61 78Wrongdoing in
patronage counties71 116 187
Wrongdoing in non-patronage counties
68 68 136
politically and increase rent-seeking opportunities forpatrons. In terms of observable implications, we expectcomplaints that could lead to the punishment of J. Pre-fecture officials to be less likely reported upward, andcomplaints that incriminate those in the patronage net-work of J. Prefecture leaders to be less likely reportedupward.31
Key Variables. Our unit of analysis is each citizencomplaint aggregated by the J. Prefecture PropagandaDepartment.Our dependent variable is whether or notthat post is reported upward to provincial-level offi-cials. Our main independent variable is whether thepost accuses J. Prefecture government officials or gov-ernment agencies of wrongdoing (PrefectureWrongdo-ing), which includes accusations of corruption and vio-lence, as well as violations of laws and regulations.32 Wealso create a variable that determines whether a postaccuses any of J. Prefecture’s subordinate counties ofwrongdoing (County Wrongdoing).33Among the 1,412 unique complaints identified in
the Online Sentiment Monitoring Reports, 28 percentrelate to government wrongdoing: 78 complaints tar-get the prefecture level, and 323 the county level (seeTable 2). Among the 78 reports of prefecture-levelwrongdoing, 31% (17) are reported upward to provin-cial superiors. Among the 323 reports of county-levelwrongdoing, 43% (139) are reported upward.To differentiate between counties that are in or out
of the patronage network of J. Prefecture officials,we examine the biographies of J. Prefecture politburomembers to identify county birthplace and workplace
31 Note that patronage may explain information manipulation bythe prefecture on behalf of its client counties, but patronage is lesslikely to explain information manipulation between the prefectureand province. Even if J. Prefecture officials were in the patronagenetwork of provincial leaders, we would expect J. Prefecture to re-port its wrongdoing to provincial superiors, but we would expect theprovince not to report J. Prefecture wrong-doing to the center.32 We hand code posts because it offers us the most precise estimateof our quantity of interest.We do not use the results of the STM be-cause STM, as with topic models more generally, allows any particu-lar document (complaint) to containmultiple topics, and interpretingtopics generated by the model is more ambiguous than using explicitcoding rules.33 If a complaint accuses both the prefecture and county governmentof wrongdoing, the complaint is coded as 1 for Prefecture Wrongdo-ing and for County Wrongdoing.
ties. Among the 14 members of the J. Prefecture polit-buro, three were born outside of the province, six wereborn in another prefecture of the same province, andfive were born in J. Prefecture or its subsidiary coun-ties. All of the five politburo members born in J. Pre-fecture come from one of two counties. If we alsoconsider counties where current J. Prefecture polit-buro members have worked, there are six counties34out of twelve where two or more politburo membershave worked or were born. We code the Connectionvariable as 1 if the complaint pertains to any one ofthese six county-level units, and 0 otherwise.35 Table 2shows upward reports of county-level wrongdoing forcounties with and without birthplace and workplaceties to J. Prefecture officials. For counties that haveconnections, 38% (71) complaints of wrongdoing arereported upward, and for counties that do not haveconnections to the prefecture, 50% (68) complaints ofwrongdoing are reported upward. The difference be-tween upward reports of wrongdoing in counties withand without connection is statistically significant with ap-value <0.01.36
Alternative Explanations. We account for four alter-native explanations that could explain upward report-ing of complaints. The first relates to the importance ofthe issue; the second to the reliability of the complaint;the third to alternative sources of information; and thefourth to the division of responsibilities between pre-fecture and provincial levels of government.It may be that the subset of information reported
upward represents the content provincial officialshave deemed to be most important. In other words,prefecture propaganda departments remove contentrelated to less important issues so provincial superiorscan focus on the topics of greatest concern. Since theCCP’s goals for monitoring citizen complaints is inlarge part related to identifying poorly performinggovernment officials, if importance motivated upwardreporting, we should see more upward reporting ofgovernment wrongdoing.However, other characteristics of citizen complaints
might make complaints more or less important toupper-level officials. We include three variables toproxy different dimensions of importance. The firstvariable (Prevalence) denotes whether the complaintrelates to an issue area that has garnered the greatestpublic attention in J. Prefecture. If there is a recurringtheme to citizen complaints, perhaps this is what pre-fecture official would highlight as important for theirsuperiors.37 Prevalence is a binary variable that takeson the value of 1 if the topic with the highest topic
34 Specifically, there are five counties and one county-level district.35 Approximately 40 percent of complaints related to counties withinthe patronage network of prefecture leaders (see Supplemental Ap-pendix for summary statistics).36 The difference between upward reports of prefecture wrongdoingand reports of wrongdoing in counties without connection is also sta-tistically significant with a p-value <0.01.37 We also examine the correlation between topic proportion fromthe STM model and upward reporting. Topics that the J. PrefecturePropaganda Department has highlighted more frequently are notthose that the department reports up to provincial-level officials.The
13
Dow
nloa
ded
from
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e. S
tanf
ord
Gra
duat
e Sc
hool
of B
usin
ess,
on
14 Ju
n 20
18 a
t 17:
19:3
4, s
ubje
ct to
the
Cam
brid
ge C
ore
term
s of
use
, ava
ilabl
e at
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e/te
rms.
htt
ps://
doi.o
rg/1
0.10
17/S
0003
0554
1800
0205
Jennifer Pan and Kaiping Chen
proportion from the STM model for that post is oneof four topics most prevalent across all posts, and 0otherwise.38 The second variable (Group Issue) mea-sures the number of people impacted by the com-plaints, since complaints that affect more people maybe more important than complaints affecting one per-son or one household.For each complaint,Group Issuetakes on the value of 1 if the complaint is shared bymore than one individual or household, and 0 other-wise. For example, a complaint that a drunken policeofficer crashed his vehicle into a pedestrian would becoded as an individual issue (0), while a complaint thatvillage cadres illegally seized land from villagers wouldbe coded as an issue pertaining to multiple households(1).39 The third variable, the (Sentiment) of complaintsmay be another dimension of importance to measurethe intensity of the complaint.We determine the senti-ment of posts using dictionary-based and probabilisticmethods,which produce similar results.40 Larger valuesdenote more positive sentiment while smaller valuesdenote more negative sentiment.The second alternative explanation relates to
whether the complaint is based on personal, directexperience, or whether the complaint is based onindirect observations of incidents or events that theperson writing the complaints has not experienceddirectly. When complaints are not based in personalexperience, they may be regarded as more speculativeand less worthy of upward reporting. We create avariable (Personal Experience), which takes on thevalue of 1 when the post is based on direct, personalexperience, and 0 otherwise.41The third alternative explanation relates to alterna-
tive sources of information. Lower-tier officials may bemore likely to report information to superiors if theirsuperiors receive the same type of information throughother sources, such that information manipulationis easier to detect. We included two variables—Collective Action and Petition—to denote informationprovincial-level superiors may also obtain throughother channels. Upper levels may obtain informationabout collective action events from the public securitybureau and information about petitions from theBureau of Letters and Visits. For Collective Action,posts discussing real-world collective action events arecoded as 1. For Petition, posts related to real-worldpetitions are coded as 1.42 Note that like the prefecture
correlation between expected topic proportion and upward report-ing is 0.12.38 Approximately 18 percent of complaints take on the value of 1 forprevalence (see summary statistics in Supplemental Appendix).39 The majority of complaints, 95 percent, deal with group-based is-sues (see summary statistics in Supplemental Appendix).40 We measure sentiment using the National Taiwan University Sen-timentDictionary as well as amultinominal logistic regressionmodeltrained on the sentiment of Chinese language movie reviews.Resultspresented in the paper are based on the second measure.41 Slightly less than 20 percent of complaints are based on personalor direct experience (see summary statistics in Supplemental Ap-pendix).42 Only 4 percent of complaints relate to collective action, and 1 per-cent of complaints relate to petitions (see summary statistics in Sup-plemental Appendix).
propaganda department, public security and lettersand visits offices at the prefecture level are primarilyaccountable to the prefecture government. As a result,if J. Prefecture suppresses information across differentinformation gathering channels, complaints relatedto Collective Action and Petition would not be morelikely to be reported upward.The last alternative explanation relates to the divi-
sion of responsibilities between prefecture and provin-cial governments. Perhaps certain citizen complaintsare not reported upward because the responsibilityfor dealing with these complaints falls to the prefec-ture government rather than the provincial govern-ment. In other words, the prefecture government onlyreports upward complaints that require provincial gov-ernments to take action, and do not report upwardcomplaints that pertain to issues the prefecture shouldmanage itself.We include Provincial Jurisdiction to de-note complaints that likely necessitate provincial in-volvement, which are issues that require interventionoutside of the prefecture.43 For example, citizens inJ. Prefecture complain about water pollution in a lo-cal river; however, the contaminants come from an up-stream lake bordering several prefectures. This issuerequires provincial intervention so the relevant com-plaints would be coded as 1. Note that the division oflabor should not, in theory, prevent upward reportingbecausemonitoring is concernedwith providing upper-level superiors an understanding of public opinion. Forexample, if a prefecture government engages in corruptpractices in an area under its jurisdiction, upper-levelsuperiors should be interested in learning this informa-tion since they are responsible for evaluating the per-formance of lower-tier officials.
Regression Results. Table 3 shows coefficient esti-mates and standard errors of logistic regression withthree specifications.44 In column (1), we estimate theeffect of Prefecture Wrongdoing alone on upwardreporting. The result indicates that complaints re-lated to J. Prefecture wrongdoing are less likely tobe reported upward, and this result is statisticallysignificant. In column (2), we examine the effectof Prefecture Wrongdoing as well as the interactionbetween County Wrongdoing and patronage connec-tions on upward reporting. Complaints related to J.Prefecture wrongdoing remain less likely to be re-ported upward, but in addition, complaints relatedto wrongdoing by counties where prefecture-levelleaders have birthplace or workplace ties are lesslikely to be reported upward while complaints re-lated to wrongdoing in the remaining, politically un-connected counties are more likely to be reportedupward.
43 Only 1 percent of complaints pertain fall under provincial juris-diction (see summary statistics in Supplemental Appendix).44 We do not cluster standard errors because the vast majority ofcomplaints identified in the monitoring reports are about differentissues, individuals, and incidents. Out of 1,412 complaints, there are1,038 unique issues. We conduct an additional analysis at the is-sue level, and our results remain unchanged (see Supplemental Ap-pendix for regression results at the issue level).
14
Dow
nloa
ded
from
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e. S
tanf
ord
Gra
duat
e Sc
hool
of B
usin
ess,
on
14 Ju
n 20
18 a
t 17:
19:3
4, s
ubje
ct to
the
Cam
brid
ge C
ore
term
s of
use
, ava
ilabl
e at
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e/te
rms.
htt
ps://
doi.o
rg/1
0.10
17/S
0003
0554
1800
0205
Concealing Corruption
TABLE 3. Predictors of Upward Reporting
Upward Reporting
(1) (2) (3)
Prefecture Wrongdoing − 0.994∗∗∗ − 0.997∗∗∗ − 0.990∗∗∗
(0.280) (0.286) (0.297)County Wrongdoing 0.289 0.273
(0.190) (0.199)Connections − 0.012 0.002
(0.130) (0.138)County Wrongdoing × Connections − 0.482∗ − 0.461∗
(0.263) (0.275)Prevalence − 0.244
(0.153)Group Issue − 2.328∗∗∗
(0.368)Sentiment 2.304∗∗∗
(0.291)Personal Experience − 0.016
(0.153)Collective Action 0.245
(0.286)Petitions − 0.800
(0.594)Provincial Jurisdiction − 0.199
(0.466)Intercept − 0.284∗∗∗ − 0.282∗∗∗ 1.330∗∗∗
(0.055) (0.081) (0.381)Observations 1,412 1,412 1,412
Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01
Column (3) includes controls for alternative expla-nations. The main results remain robust. Wrongdoingby J. Prefecture is less likely to be reported upwardand wrongdoing by politically connected counties isless likely to be reported upward. From column (3),we can also see that upward reporting is not related tothe prevalence of the issue, and issues which pertain togroups rather than individuals are less likely to be sentup. Instead of extremely negative complaints beingmore likely to be reported upward, we find that com-plaints with more positive sentiment are more likelyto be reported upward. Complaints based on personalexperience are not more likely to be reported upward.There are no statistically significant relationships be-tween collective action and upward reporting or pe-titions and upward reporting. Finally, issues that arelikely to require regional coordination and fall underthe jurisdiction of the provincial government are notmore likely to be sent up.Figure 4 plots first differences from the logistic re-
gression specified in column (3) of Table 3 so we canbetter interpret these results.We can see from Figure 4that complaints related to prefecture wrongdoing are21% less likely to be reported upward than complaintsthat do not mention prefecture wrongdoing. For com-plaints related to county-level wrongdoing, contentpertaining to counties that have ties to prefecture polit-buro members are 11% less likely than content im-plicating other counties to be reported upward. There
FIGURE 4. First differences.
is little difference in upward reporting (6%) betweenposts that pertain to prevalent topics and those thatdo not. Issues that encompass greater numbers of peo-ple are 46% less likely to be reported upward than is-sues related to individuals. Comparing posts with themost positive sentiment and most negative sentiment,content with the most positive sentiment is 51% more
15
Dow
nloa
ded
from
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e. S
tanf
ord
Gra
duat
e Sc
hool
of B
usin
ess,
on
14 Ju
n 20
18 a
t 17:
19:3
4, s
ubje
ct to
the
Cam
brid
ge C
ore
term
s of
use
, ava
ilabl
e at
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e/te
rms.
htt
ps://
doi.o
rg/1
0.10
17/S
0003
0554
1800
0205
Jennifer Pan and Kaiping Chen
likely to be reported upward. There is virtually no dif-ference (<1%) in posts based on personal, direct expe-rience and those based on indirect information. Com-plaints that discuss real-world collection action are 6%more likely to be reported upward, and complaints per-taining to petitions are 17% less likely to be reportedupward, but neither result is statistically significant. Fi-nally, when a complaint requires the intervention ofprovincial authorities, it is 4% less likely to be reportedupward, but again, the difference is not statistically sig-nificant.These results clearly show that upward reporting is
driven by the incentives of prefecture-level officials toprotect their interests, to conceal online complaints ac-cusing the prefecture government of wrongdoing aswell as complaints of wrongdoing by counties whereprefecture officials have birthplace and workplaceties.
Wrongdoing Not Censored
Finally, we are interested in whether J. Prefecture con-ceals wrongdoing solely by not reporting the infor-mation upward, or whether they also censor the con-tent they do not report to upper levels. Examiningwhether the content that is not reported upward iscensored provides us with information about the lo-cal governments’ tactics for information manipula-tion, but perhaps more importantly, sheds light onwhether upper levels of government are also moni-toring online complaints directed at subordinate pre-fectures and counties. Presumably, if upper levels ofgovernment are also monitoring complaints, upper lev-els are more likely to detect information manipula-tion if the concealed information were still availableonline.We create two new variables to facilitate this anal-
ysis: the dependent variable Censorship, based onwhether the complaint has been censored or if it re-mains publicly viewble, and an independent variablePrefecture Censorship Authority denoting whether theprefecture has the authority to censor a complaint.To create the Censorship variable, we use a three-
step process to check whether the content is stillpublicly available online. First, we use the Googlecustom search API to search the text of each com-plaint and collect the first ten URLs of search re-sults.45 Second, we load the content of each returnedURL, and we use a variety of automated text matchingmethods—e.g., cosine similarity, sub-string matching—to determine whether the content of the complaintmatches the content of the returned URL. Third, weuse extensive human validation to ensure the valid-ity of our text matching, hence the censorship mea-sure. If the complaint remains publicly viewable onany platform, we code Censorship as 0, and if it is nolonger available, then this variable takes on the valueof 1.
45 If we are not able to find the complaints using the Google customsearch API, we conduct an additional manual search using a varietyof search engines such as Baidu.cn.
Based on interviews,we find local-level officials havecensorship authority only for websites they manage.For example, J. Prefecture operates a prefecture gov-ernment website with a public forum, and the J. Pre-fecture government can censor content on this forum.However, for most other websites (e.g., national-levelplatforms such as Sina Weibo or Tianya), the J. Prefec-ture government has no censorship authority.To createthe Prefecture Censorship Authority variable, we codea complaint as 1 if it was posted to websites ownedor operated by J. Prefecture, and 0 otherwise. We findthat 16% of complaints are posted to websites run byJ.Prefecture,which the prefecture can censor,while theremaining bulk of complaints are posted to sites whereJ. Prefecture has no censorship authority.Table 4 shows the result of logistic regressions where
the dependent variable is Censorship. Column (1)shows a model where prefecture wrongdoing is not in-teracted with censorship authority, while column (2)contains this interaction. In column (1), we see thatposts related to prefecture wrongdoing are more likelyto be censored and this effect is exacerbated whenthe prefecture has censorship authority. In column (2),when these variables are interacted, we see that postsrelated to J. Prefecture wrongdoing are more likely tobe censored, but only when the prefecture has cen-sorship authority over the site where the complainthas been made. Since 84 percent of complaints are notposted to sites where J. Prefecture has censorship au-thority, this does not apply to the bulk of complaints.We also find that complaints that containmore positivesentiment are less likely to be censored,and complaintsbased on personal experience are less likely to be cen-sored.46
These results show that, when possible, local of-ficials use all tools at their disposal to manipulateinformation—by distorting what is reported upwardand by censoring complaints of prefecture wrongdo-ing onwebsites they control.However, sincemost com-plaints are posted to websites that J. Prefecture cannotcensor, information manipulation likely occurs primar-ily through the upward reporting process rather thancensorship.Incomplete censorship of J. Prefecture wrongdoing
leaves open the possibility that provincial and centralleaders could find and punish malefactors based oncomplaints posted to sites where J. Prefecture cannotcensor. However, our finding that prefecture officialspersist in concealing wrongdoing despite this possibil-ity strengthens the conclusion that upper-level lead-ers are not investing the time or resources to evaluateall information pertaining to lower-level governments.As a result, lower-level officials can engage in infor-mation manipulation despite incomplete censorship.If provincial officials were able to detect prefecture-
46 Since most complaints are posted to websites that J. Prefecturecannot censor, we do not think the negative effect of personal expe-rience on censorship means that more reliable content is less likelyto be censored. A more likely explanation is that posts based onpersonal experience may get more attention or are less likely to beflagged as spam content, and hence less likely to be deleted.
16
Dow
nloa
ded
from
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e. S
tanf
ord
Gra
duat
e Sc
hool
of B
usin
ess,
on
14 Ju
n 20
18 a
t 17:
19:3
4, s
ubje
ct to
the
Cam
brid
ge C
ore
term
s of
use
, ava
ilabl
e at
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e/te
rms.
htt
ps://
doi.o
rg/1
0.10
17/S
0003
0554
1800
0205
Concealing Corruption
TABLE 4. Predictors of Censorship
Censorship
(1) (2)
Prefecture Wrongdoing 0.435 0.160(0.272) (0.313)
Prefecture Censorship Authority 0.392∗∗ 0.290(0.177) (0.186)
Prefecture Wrongdoing × Prefecture Censorship Authority 1.601∗∗
(0.761)Prevalence − 0.878∗∗∗ − 0.878∗∗∗
(0.220) (0.220)Group Issue − 0.270 − 0.269
(0.332) (0.332)Sentiment − 0.653∗ − 0.654∗
(0.359) (0.360)Personal experience − 1.648∗∗∗ − 1.654∗∗∗
(0.278) (0.279)Collective Action 0.242 0.245
(0.343) (0.343)Petition − 0.813 − 0.794
(0.785) (0.785)Provincial Jurisdiction − 0.936 − 0.956
(0.754) (0.754)Intercept − 0.687∗∗ − 0.668∗
(0.347) (0.348)Observations 1,337 1,337
Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01
level corruption (from the complaints posted to web-sites the prefecture cannot censor), the province wouldmore likely be able to identify discrepancies betweenprefecture propaganda reports and provincial informa-tion. Then, if the province so desires, it could sanc-tion,or punish,prefecture propaganda departments formanipulating information. The strong empirical resultthat the prefecture propaganda department systemat-ically manipulates information along with our existingknowledge of China’s information-monitoring systemsuggests provincial superiors in this region are not di-rectly expending resources to mine and analyze datafrom online complaints for identifying lower-level cor-ruption.
CONCLUSION
China is often characterized as a regime that has pre-vailed against the information problems that plagueautocrats, in large part because it has built many chan-nels to gather information through citizen complaintsand grievances. Our evidence indicates the opposite—that there are systematic shortcomings in China’s abil-ity to gather reliable and accurate information aboutthe actions of regime agents through citizen participa-tion. China remains reliant on monitoring agencies togather, distill, and verify the large quantities of infor-mation voiced by the public.
Our data, based on rarely seen internal communi-cations between a monitoring agency and upper-levelauthorities in J. Prefecture, reveals information manip-ulation by the prefecture government. These resultscorrespond with our theoretical expectations given therelative incentives and resources of top leaders in J.Prefecture to hide information relative to the incen-tives and resources of top provincial leaders to over-come information manipulation. However, in other lo-calities where the relative incentives and resourcesof upper-level and lower-level officials differ fromthose in J. Prefecture, the strategic interaction betweenupper-level and lower-level officials over informationmay yield different outcomes.We hope future researchwill examine this dynamic in other regions of China,and in other authoritarian contexts.We might also expect deviations from the findings in
this analysis when it comes to monitoring by the cen-tral government.China’s central authorities have moreresources and stronger capabilities to directly monitorregime agents than subnational levels of government.However, the center has made clear that it is inter-ested primarily in monitoring national- and regional-level activities.47 The purpose of China’s hierarchicaladministrative structure is to delegate, and likewise,
47 As discussed previously, we know that central authorities are cur-rently developing systems that would directly gather and analyze cit-izen complaint data from social media platforms down to the prefec-ture level.
17
Dow
nloa
ded
from
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e. S
tanf
ord
Gra
duat
e Sc
hool
of B
usin
ess,
on
14 Ju
n 20
18 a
t 17:
19:3
4, s
ubje
ct to
the
Cam
brid
ge C
ore
term
s of
use
, ava
ilabl
e at
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e/te
rms.
htt
ps://
doi.o
rg/1
0.10
17/S
0003
0554
1800
0205
Jennifer Pan and Kaiping Chen
responsibility for regularly monitoring low-level offi-cials has been delegated to subnational leaders. Offi-cials at these lower levels—the county and below—arethose primarily responsible for policy implementationand those who most frequently interact with the pub-lic.Thus, informationmanipulation and concealment ofcorruption at county and lower levels will likely per-sist despite increased central monitoring. This is con-sequential for the regime since public trust and satis-faction with county and lower levels of government inChina is low, hindering the country’s ability to govern,to carry out policies, and to prevent mass incidents (Li2004;Whyte 2010; Yan and Peng 2010).
These results demonstrate an alternative way inwhich information manipulation occurs—not primar-ily through censorship or the deletion of undesirableinformation and not through falsification or fabrica-tion, but through partial concealment. Because upper-level authorities rely onmonitoring agencies to synthe-size large quantities of online data, monitoring agen-cies can satisfy upper-level demands for informationwith truthful but incomplete information that system-atically hides corruption.Our findings bring additional nuance to our un-
derstanding of the relationship between nonelectoralforms of citizen participation and accountability in au-thoritarian regimes. Although individuals living underauthoritarian rule are in some ways free to publiclyexpress their grievances, whether this information canlead to accountability depends on whether this infor-mation can make its way to those with sanctioningpower, and whether those who can sanction actuallydo so.Our results, showing how the public’s complaintsof lower-level malfeasance do not reliably make theirway to upper-level authorities, reveal one way in whichpublic participation is insufficient for accountability innondemocratic contexts.Finally, these results show that conflicts over infor-
mation remain a challenge for authoritarian regimes inthe digital age. Even in a highly determined and ca-pable authoritarian regime where grievances are ac-tively and publicly voiced, information manipulationby lower-level officials persists.Even in the era of large-scale data and increasingly sophisticated methods foranalyzing large quantities of data, political incentivescontinue to motivate information manipulation. Eventhough authoritarian regimes across the world areadopting online systems for public complaints, and au-tocrats may learn a great deal from monitoring onlinecontent and social media, this proliferation of recordedinformation by no means guarantees autocrats omni-science.We hope our results will generate greater focuson political communication and information conflictsamong elites and regime insiders to complement exist-ing work on information conflicts between the regimeand society.
SUPPLEMENTARY MATERIAL
To view supplementary material for this article, pleasevisit https://doi.org/10.1017/S0003055418000205.
REFERENCES
Cai,Yongshun. 2000. “Between State and Peasant: Local Cadres andStatistical Reporting in Rural China.” The China Quarterly 163:783–805.
Cai,Yongshun. 2013. “Information Collection and Anticorruption inChina.” InElites and Governance in China, eds.Xiaowei Zang andChien wen Kou. New York: Routledge, 136–52.
Central Committee of the Communist Party of China. 2016. “Zhong-guo gongchandang dangnei jiandu tiaoli[China Communist PartyInternal Supervision Regulations].”.
Central Propaganda Department. 2009. Wangluo Yuqing XinxiGongzuo Lilun yu Shiwu [Theories and Guidance on Online Pub-lic Sentiment]. Xuexi chuban she.
Chang, Jonathan, Jordan L. Boyd-Graber, Sean Gerrish,Chong Wang, and David M. Blei. 2009. “Reading Tea Leaves:How Humans Interpret Topic Models.” In Nips, Vol. 31, pp. 1–9.
Chen, Jidong, Jennifer Pan, andYiqing Xu. 2016. “Sources of Author-itarian Responsiveness: A Field Experiment in China.”AmericanJournal of Political Science 60 (2): 383–400.
Chen, Xi. 2009. “Power of Troublemaking: Chinese Petitioners’ Tac-tics and Their Efficacy.”Comparative Politics 41 (4): 451–71.
China Daily. 2015. “NewAnti-Corruption App Sparks Instant Floodof Public Reports.” Accessed December 13, 2016. http://bit.ly/1GhcZQl.
Cyberspace Administration of China. 2017. “Hulianwang XinwenXinxi Fuwu Guanli Guiding [Internet News and Information Ser-vice Management Regulations].” Beijing, China.
Dimitrov, Martin K. 2014a. “Internal Government Assessments ofthe Quality of Governance in China.” Studies in Comparative In-ternational Development, 1–23.
Dimitrov, Martin K. 2014b. “What the Party Wanted to Know Citi-zen Complaints as a “Barometer of Public Opinion” in CommunistBulgaria.”East European Politics & Societies 28 (2): 271–95.
Distelhorst, Greg, and Yue Hou. 2017. “Constituency Service UnderNondemocratic Rule:Evidence fromChina.” Journal of Politics 79(3).
Economist. 2017. “Chinese Officials Use Hotlines to Take the Pub-lic’s Pulse.” Accessed August 20, 2017. http://whr.tn/1imqqlx.
Edin,Maria. 2003. “State Capacity and Local agent Control in China:CPP Cadre Management from a Township Perspective.” ChinaQuarterly 173: 35–52.
Fishkin, James S., Baogang He, Robert C. Luskin, and Alice Siu.2010. “Deliberative Democracy in an Unlikely Place:DeliberativePolling in China.”British Journal of Political Science 40 (2): 435–48.
Freedom House. 2015. Freedom in the World 2015: The Annual Sur-vey of Political Rights and Civil Liberties. Lanham,MD: Rowman& Littlefield.
Gao, Jie. 2016. “‘Bypass the Lying Mouths’: How Does the CCPTackle Information Distortion at Local Levels?”The China Quar-terly. p. doi:10.1017/S0305741016001491.
Gunitsky, Seva. 2015. “Corrupting the Cyber-Commons: Social Me-dia as a Tool of Autocratic Stability.” Perspectives on Politics 13(01): 42–54.
He, Baogang, and Mark Warren. 2011. “Authoritarian Deliberation:The Deliberative Turn in Chinese Political Development.” Per-spectives on Politics 9 (2): 269–89.
Henochowicz, Anne. 2014. “Thousands of Local Internet Pro-paganda Emails Leaked.” China Digital Times. http://j.mp/leakedEmails.
Hillman,Ben. 2010. “Factions and Spoils:Examining Political Behav-ior within the Local State in China.”The China Journal 64: 1–18.
Hsu, S. Philip, Suisheng Zhao, and Yushan Wu. 2011. In Searchof China’s Development Model: Beyond the Beijing Consensus.Abingdon, UK: Routledge.
Huang, Yasheng. 1995. “Administrative Monitoring in China.” TheChina Quarterly 143: 828–43.
Jiang, Junyan, Tianguang Meng, and Qing Zhang. 2016. “Closing theInfluence Gap: The Policy Consequences of Online Participationin China.”APSA Annual Meeting.
Kalathil, Shanthi, and Taylor C. Boas. 2010.Open Networks, ClosedRegimes: the Impact of the Internet on Authoritarian Rule. Wash-ington D.C.: Carnegie Endowment.
King, Gary, Jennifer Pan, and Margaret E. Roberts. 2013. “HowCensorship in China Allows Government Criticism but Silences
18
Dow
nloa
ded
from
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e. S
tanf
ord
Gra
duat
e Sc
hool
of B
usin
ess,
on
14 Ju
n 20
18 a
t 17:
19:3
4, s
ubje
ct to
the
Cam
brid
ge C
ore
term
s of
use
, ava
ilabl
e at
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e/te
rms.
htt
ps://
doi.o
rg/1
0.10
17/S
0003
0554
1800
0205
Concealing Corruption
Collective Expression.”American Political Science Review 107 (2):1–18.
Kou, Chien-wen, and Wen-Hsuan Tsai. 2014. “‘Sprinting with SmallSteps’ Towards Promotion: Solutions for the Age Dilemma in theCCP Cadre Appointment System.”China Journal 71: 153–71.
Kung, James Kai-Sing, and Shou Chen. 2011. “The Tragedy of theNomenklatura: Career Incentives and Political Radicalism duringChina’s Great Leap Famine.” American Political Science Review105 (1): 27–45.
Li, Lianjiang. 2004. “Political Trust in Rural China.”Modern China30 (2): 228–58.
Li, Ruichang. 2012.Zhengfujian Wangluo Zhili: Chuizhi Guanli Bu-menYuDifangZhengfujianGuanxi Yanjiu [NetworkGovernance:Top-downManagement of Departments and Inter-GovernmentalRelationships]. Shanghai, China: Fudan University Press.
Lieberthal, Kenneth, and David Lampton. 1992. Bureaucracy, Poli-tics, and Decision Making in Post-Mao China. Berkeley, CA: Uni-versity of California Press.
Lorentzen, Peter. 2013. “Regularizing Rioting: Permitting Protest inan Authoritarian Regime.”Quarterly Journal of Political Science 8(2): 127–58.
Lucas, Christopher, Richard A. Nielsen,Margaret E. Roberts, Bran-donM.Stewart,Alex Storer, andDustin Tingley.2015.“Computer-Assisted Text Analysis for Comparative Politics.” Political Analy-sis p. mpu019.
Ma, Jun. 2012. “The Rise of Social Accountability in China.” Aus-tralian Journal of Public Administration 71 (2): 111–21.
MacKinnon, Rebecca. 2012. Consent of the Networked: The World-wide Struggle For Internet Freedom. New York, NY: Basic Books.
Magaloni,Beatriz. 2008. “Credible Power-Sharing and the LongevityofAuthoritarianRule.”Comparative Political Studies 41 (4/5):715–41.
Magaloni, Beatriz, and Ruth Kricheli. 2010. “Political Order andOne-Party Rule.”Annual Review of Political Science 13: 123–43.
Mai, Bo, and Tim Liber. 2015. “Web Tracking with Chinese Charac-teristics.” 65th Annual International Communication AssociationConference.
Manion,Melanie. 2016. Information for Autocrats: Representation inChinese Local Congresses.Cambridge,UK:Cambridge UniversityPress.
Meng, Tianguang, Jennifer Pan, and Ping Yang. 2017. “ConditionalReceptivity to Citizen Participation: Evidence from a Survey Ex-periment in China.”Comparative Political Studies 50 (4): 399–433.
Michael, Gabriel J. Forthcoming. “Who’s Afraid of Wikileaks?Missed Opportunities in Political Science Research.” Review ofPolicy Research.
Morozov, Evgeny. 2012. The Net Delusion: the Dark Side of InternetFreedom. New York, NY: Public Affairs.
Nathan, Andrew. 2003. “Authoritarian Resilience.” Journal ofDemocracy 14 (1): 6–17.
National Grassroots Party Work Key Textbook. 2013. Zuixin Dangde Xuanchuan Gongzuo Caozuo Fangfa yu Chuangxin Shiwu[LatestGuidelines andProcedures onHow toPerformPropagandaWork]. Beijing, China: Hongqi Press.
Noesselt,Nele. 2014. “Microblogs and the Adaptation of the ChineseParty-State’s Governance Strategy.”Governance 27 (3): 449–68.
O’Brien, Kevin, and Lianjiang Li. 1999. “Selective Policy Implemen-tation in Rural China.”Comparative Politics 31 (2): 167–86.
O’Brien,Kevin J., and Lianjiang Li. 2000. “Accommodating “democ-racy” in a one-party state: Introducing village elections in China.”The China Quarterly 162: 465–89.
O’Donnell, Guillermo. 1999. Horizontal Accountability in NewDemocracies. In The Self-Restraining State: Power and Ac-countability in New Democracies, eds. Larry Diamond, An-dreas Schedler, and Marc F. Plattner. Boulder, CO:Lynne RiennerPublishers, Inc., pp. 29–51.
OECD. 2013. OECD e-Government Studies: Egypt 2013. Paris,France: OECD Publishing.
Oi, Jean C. 1995. “The Role of the Local State in China’s TransitionalEconomy.”The China Quarterly 144: 1132–49.
Olken,Benjamin A. 2007. “Monitoring Corruption:Evidence from aField Experiment in Indonesia.” Journal of Political Economy 115(2): 200–49.
Pan, Jennifer. Forthcoming. “HowChinese Officials Use the Internetto Construct their Public Image.” Political Science Research andMethods.
Pan, Jennifer, and Kaiping Chen. 2018. “Replication Data forConcealing Corruption: How Chinese Officials Distort UpwardReporting of Online Grievances.” https://doi.org/10.7910/DVN/2HRRC7.
Park, Albert, and Sangui Wang. 2001. “China’s Poverty Statistics.”China Economic Review 12 (4): 384–98.
Pearce, Katy E., and Sarah Kendzior. 2012. “Networked Authoritar-ianism and Social Media in Azerbaijan.” Journal of Communica-tion 62 (2): 283–98.
Qiang, Xiao. 2011. “The Battle for the Chinese Internet.” Journal ofDemocracy 22 (2): 47–61.
Qin, Bei, David Strömberg, and Yanhui Wu. 2016. “Media Bias inChina.”Working paper.
Qin, Bei, David Stromberg, and Yanhui Wu. 2017a. “Why DoesChina Allow Freer Social Media? Protests versus Surveillance andPropaganda.” CEPR working paper DP11778.
Qin, Bei, David Stromberg, and Yanhui Wu. 2017b. “Why DoesChina Allow Freer Social Media? Protests versus Surveillanceand Propaganda.” Journal of Economic Perspectives 31 (1):117–40.
Roberts,Margaret E.,BrandonM.Stewart,Dustin Tingley,Chris Lu-cas, Jetson Leder-Luis, Bethany Albertson, Shana Gadarian, andDavid Rand. 2014. “Topic Models for Open Ended Survey Re-sponses with Applications to Experiments.” American Journal ofPolitical Science 58: 1064–82.
Roberts, Margaret E., Brandon M. Stewart, Dustin Tingley, andEdoardo M. Airoldi. 2013. “The Structural Topic Model and Ap-plied Social Science.” In Advances in Neural Information Process-ing SystemsWorkshop onTopicModels:Computation,Application,and Evaluation.
Rodan, Garry, and Kanishka Jayasuriya. 2007. “The TechnocraticPolitics of Administrative Participation:Case Studies of Singaporeand Vietnam.”Democratization 14 (5): 795–815.
Sonnad, Nikhil. 2014. “Hacked emails reveal China’s elaborate andabsurd internet propagandamachine.”Quartz. http://j.mp/Sonnad.
Stockmann, Daniela. 2013. Media Commercialization and Authori-tarian Rule in China. New York: Cambridge University Press.
Truex, Rory. 2014. “The Returns to Office in a ‘Rubber Stamp’Parliament.” American Political Science Review 108 (02): 235–51.
Truex, Rory. 2016.Making Autocracy Work. Cambridge, UK: Cam-bridge University Press.
Tsai, Lily. 2008. “Understanding the Falsification of Village IncomeStatistics.”The China Quarterly 196: 805–26.
Wallace, Jeremy L. 2016. “Juking the Stats? Authoritarian Informa-tion Problems in China.” British Journal of Political Science 46(01): 11–29.
Whiting, Susan. 2004. “The Cadre Evaluation System at the GrassRoots: The Paradox of Party Rule.” In Holding China Together:Diversity and National Integration in the post-Deng Era, eds.Dali Yang and Barry Naughton. Cambridge,UK:Cambridge Uni-versity Press, 101–19.
Whyte, Martin. 2010.Myth of the Social Volcano: Perceptions of In-equality and Distributive Injustice in Contemporary China. Stan-ford, CA: Stanford University Press.
Wintrobe, Ronald. 1998. The Political Economy of Dictatorship.Cambridge, UK: Cambridge University Press.
Xinhua. 2012. “China’s craze for online anti-corruption.” AccessedSeptember 10, 2016. http://bit.ly/2blgIYz.
Yan, Ye, and Min Peng. 2010. “The Unbalance Political Trust Distri-bution between Central Government and Local Government inChina.”The Journal of Gansu Administration Institute 3 (008).
Yong, Hu. 2005. Blogs in China: China Media Project Case Study.Technical report, Peking University.
Zhan, Weiping, and Yongda Yu. 2011. “Study of the RelationshipBetween Cadre Promotion and Career Path.” Human ResourcesDevelopment of China 7: 75–8.
Zhang, Jian, John Giles, and Scott Rozelle. 2012. “Does it Pay to be aCadre? Estimating the Returns to being a Local Official in RuralChina.” Journal of Comparative Economics 40 (3): 337–56.
19
Dow
nloa
ded
from
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e. S
tanf
ord
Gra
duat
e Sc
hool
of B
usin
ess,
on
14 Ju
n 20
18 a
t 17:
19:3
4, s
ubje
ct to
the
Cam
brid
ge C
ore
term
s of
use
, ava
ilabl
e at
htt
ps://
ww
w.c
ambr
idge
.org
/cor
e/te
rms.
htt
ps://
doi.o
rg/1
0.10
17/S
0003
0554
1800
0205