how the chinese government fabricates social media posts for … · 2017-08-28 · how the chinese...

American Political Science Review (2017) 111 3 484ndash501

doi101017S0003055417000144 ccopy American Political Science Association 2017

How the Chinese Government Fabricates Social Media Posts forStrategic Distraction Not Engaged ArgumentGARY KING Harvard UniversityJENNIFER PAN Stanford UniversityMARGARET E ROBERTS University of California San Diego

T he Chinese government has long been suspected of hiring as many as 2 million people to surrep-titiously insert huge numbers of pseudonymous and other deceptive writings into the stream ofreal social media posts as if they were the genuine opinions of ordinary people Many academics

and most journalists and activists claim that these so-called 50c party posts vociferously argue for thegovernmentrsquos side in political and policy debates As we show this is also true of most posts openlyaccused on social media of being 50c Yet almost no systematic empirical evidence exists for this claimor more importantly for the Chinese regimersquos strategic objective in pursuing this activity In the firstlarge-scale empirical analysis of this operation we show how to identify the secretive authors of theseposts the posts written by them and their content We estimate that the government fabricates and postsabout 448 million social media comments a year In contrast to prior claims we show that the Chineseregimersquos strategy is to avoid arguing with skeptics of the party and the government and to not even discusscontroversial issues We show that the goal of this massive secretive operation is instead to distract thepublic and change the subject as most of these posts involve cheerleading for China the revolutionaryhistory of the Communist Party or other symbols of the regime We discuss how these results fit withwhat is known about the Chinese censorship program and suggest how they may change our broadertheoretical understanding of ldquocommon knowledgerdquo and information control in authoritarian regimes

INTRODUCTION

Social media in China appears as vibrant and ex-tensive as in any Western country with more than1300 social media companies and websites and

millions of posts authored every day by people all overthe country At the same time the Chinese regime im-poses extensive and varied controls over of the entiresystem (Brady 2009 Cairns and Carlson 2016 Knockelet al 2015 MacKinnon 2012 Ng 2015 Shirk 2011Stockmann 2013 Stockmann and Gallagher 2011 Yang2009) Which social media companies are preventedfrom operating in China is easy to see (the so-calledGreat Firewall of China) and the scholarly literaturenow offers considerable evidence on how and why

Gary King is Albert J Weatherhead III University Profes-sor Institute for Quantitative Social Science 1737 CambridgeSt Harvard University Cambridge MA 02138 GaryKingorg(KingHarvardedu)

Jennifer Pan is Assistant Professor Department of Communica-tion 450 Serra Mall Building 120 Stanford University Stanford CA94304 jenpancom (jp1stanfordedu)

Margaret E Roberts is Assistant Professor Department of Po-litical Science University of California San Diego Social SciencesBuilding 301 9500 Gilman Dr 0521 La Jolla CA 92093-0521MargaretRobertsnet (merobertsucsdedu)

Our thanks to Neel Guha Peter Dyrud Yingjie Fan and many oth-ers for superb research assistance Danielle Allen Peter Bol BeckyFair Chase Harrison Iain Johnston Franziska Keller Blake MillerJean Oi Samantha Ravich Brandon Stewart Daniela StockmannAndy Walder Yuhua Wang Chaodan Zheng and Yun Zhu for help-ful comments and DARPA (contract W31P4Q-13-C-0055983-3)and the National Science Foundation (grant 1500086) for researchsupport Authors are listed alphabetically Data and informationnecessary to replicate the results in this article appear in King et al(2017)

Received August 26 2016 revised November 6 2016 acceptedApril 11 2017

they censor certain individual social media posts thathave appeared on the web or filter them out beforeappearing In both cases the censorship apparatus al-lows a great deal of criticism of the regime its officialsand their policies (which can be useful information forthe central government in managing local leaders) butstops discussions that can generate collective action onthe ground (King et al 2013 2014)1

According to numerous speculations by scholars ac-tivists journalists officials in other governments andparticipants in social media the Chinese regime alsoconducts ldquoastroturfingrdquo or what we might call reversecensorship surreptitiously posting large numbers offabricated social media comments as if they were thegenuine opinions of ordinary Chinese people The peo-ple hired for this purpose are known formally as Inter-net commentators () although more widelyas 50c party members () so called because theyare rumored to be paid 50 cents (5 jiao or about$008) to write and post each comment (Tong and Lei2013) We show that this rumor turns out to be in-correct however we adopt this widely used term todenote social media comments posted at the directionor behest of the regime as if they were the opinions ofordinary people2

1 Although we make general statements about Chinese censorship itis important to note that censorship in China is by no means mono-lithic in its operations or outcomes Censorship is largely carriedout by Internet content providers and regulations over the flow ofinformation continues to evolve As a result there is variation in theprecise details of censorship in China by platform geography andover time2 Thus 50c party members are distinct from ldquovolunteer 50c mem-bersrdquo ( known as ldquobring your own grainersrdquo) who expressproregime or anti-Western sentiment online without being paid by

484

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44

How the Chinese Government Fabricates Social Media Posts for Strategic Distraction Not Engaged Argument

The nearly unanimous view of journalists (and hu-man rights activists) is that 50c party posts stronglyargue with and debate against those who criticize thegovernment its leaders and their policies This is alsothe view of many scholarly publications discussing thisactivity (and was our view as well prior to the re-search reported here) We systematically summarizethese views including a quantitative analysis of socialmedia posts openly accused of being written by 50cparty members Unfortunately until now no methodhas existed for detecting 50c party members whichposts they write their content the size of the opera-tion or why they write them and so the best anyonecould do was base claims on intuition logic occasionalanecdotes rumors and leaked government directives

In this article we offer the first systematic empiricalevidence for the content of 50c party posts their au-thors and the governmentrsquos strategic objectives3 Webegin by analyzing an archive of emails leaked from theInternet Propaganda Office of Zhanggong a district ofGanzhou City in Jiangxi Province These emails giveexplicit details of the work of numerous 50c accountsin this district Although in the public domain and re-ported in the press (eg Henochowicz 2014 Sonnad2014) the structure of the archive is complicated toolarge to understand by traditional qualitative meth-ods and in formats (and attachments) far too diverseto make standard methods of automation feasible Assuch it has never before been systematically analyzedand little of it has been explored We have developedan approach to analyze this dataset and have extractedmore than 43000 known 50c party posts and their au-thors from it

We first characterize the patterns in these data viatheir network and time series structures Then we sys-tematically analyze the content of the 50c posts in ourleaked archive and extrapolate to the rest of Chinain stages We then use this methodology to study thecontent of the posts and finally infer the goals behindthis massive government program and how it may re-veal broader government strategies We validate our50c party member predictions with a novel sample sur-vey of predicted 50c party members as well as severalunusual gold standard evaluations that we develop tovalidate our validation We estimate and reveal the sizeof what turns out to be a massive government operationthat writes approximately 448 million 50c posts a yearWe also discuss our assumptions interpretations andwhat might go wrong with our evidence and inferences

At every stage our results indicate that prevailingviews about the 50c party are largely incorrect We

the government the ldquolittle red flowersrdquo () unpaid red guardswho also attack opponents of the regime online the ldquoAmerican CentPartyrdquo () who express Western democratic values and criti-cize the Chinese communist regime online and the ldquoInternet waterarmyrdquo () which refers to for-hire astroturfers working forand advancing the interests of companies and other actors willing topay their fees None are known to be organized as an entire collectiveOf course political parties do not exist in China and so despite thename the 50c party is not a political party For an excellent overviewof the purpose and tactics of the volunteer 50c party see Han (2015a)3 Because China is a single-party regime we use the terms govern-ment and regime interchangeably to refer to those in power

show that almost none of the Chinese governmentrsquos50c party posts engage in debate or argument of anykind They do not step up to defend the governmentits leaders and their policies from criticism no matterhow vitriolic indeed they seem to avoid controversialissues entirely Instead most 50c posts are about cheer-leading and positive discussions of valence issues Wealso detect a high level of coordination in the timing andcontent in these posts A theory consistent with thesepatterns is that the strategic objective of the regime is todistract and redirect public attention from discussionsor events with collective action potential

The theoretical implications of our findings are pre-sented later in the article We give a unified parsimo-nious summary of Chinese government internal infor-mation control efforts and show how these findingsmay cause scholars to rethink the notion of ldquocommonknowledgerdquo in theories of authoritarian politics moregenerally Finally we conclude and then give a sum-mary of what we might have missed and how scholarscan follow up on this work4

WHAT WE THINK WE KNOW

We summarize here views about the 50c party of (1)journalists (2) academics and (3) social media par-ticipants accusing others of being 50c party membersThe dominant view of most is that 50c party membersengage in ldquohand-to-handrdquo verbal combat making spe-cific directed arguments that support the governmentits leaders and their policies and opposing argumentsto the contrary they do this by engaging in debate withand criticism of Chinarsquos enemies including those whooppose it inside the country and from abroad For (1)and (2) we offer brief literature reviews for (3) we findand analyze posts accused by others of being 50c Inthat section we also introduce and validate a schemeto categorize 50c posts on the web we use it in thissection to understand the posts accused of being writ-ten by 50c members and then for many other purposesthroughout this article

Although the difficulties of collecting data on aninherently secret operation means that most priorliterature includes ldquono successful attempts to quan-tify regime-sponsored commentary in Chinardquo (Miller2016) the work cited in this section involves consider-able effort and creativity and even a few clever effortsto guess what might be 50c posts before turning totry to explain or predict the guesses (Miller 2016 Han2015b) For example Han (2015b) uses informationfrom leaked censorship directives and local media re-ports of the training of online commentators in an on-line ldquoguerrilla ethnographyrdquo Still the lack of ground

4 Appendix B briefly describes the (unintended) events and implica-tions of the Chinese governmentrsquos unexpected decision to respondto an early version of this article through an editorial in a state-owned newspaper In so doing they appear to give a rare admissionof the existence of the 50c party and confirmation of our empiricalresults In doing this they also provide important information aboutgovernment strategy that we discuss (Our Online SupplementaryAppendix provides additional details and a translation of the originaleditorial)

485

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44

Gary King Jennifer Pan and Margaret E Roberts

truth means that for identification of 50c posts Hanhas to rely on anecdotal evidence and intuition (egwhether posts ldquosmell strongly of official propagandardquo)In other works sophisticated unsupervised statisticaltechniques have been used but still generated ldquono ev-idence of large-scale Wumao [50c] activity on Weibordquo(Yang et al 2015) As these authors make clear littlesolid empirical evidence exists about the content andextent of 50c party posts

Journalists The popular press describes 50c membersas ldquoundercover progovernment Internet commentersrdquo(Keating 2011) who ldquoset out to neutralize undesirablepublic opinion by pushing pro-Party views throughchat rooms and web forumsrdquo (Bandurski 2008 41)They ldquoshape online public opinionrdquo by labeling ldquocrit-ical opinion leaders as traitors of the countryrdquo (Lam2012) Prominent dissident Ai Weiwei said ldquoIf you op-pose the US and Japan [online] you are a member ofthe 50 cents armyrdquo (Strafella and Berg 2015 154) The50c party members ldquocombat hostile energyrdquo defined asposts that ldquogo against socialist core valuesrdquo or ldquoare notamenable to the unity of the peoplerdquo Such informationshould be ldquoresolutely resisted proactively refuted andeagerly reported to Internet authoritiesrdquo (Haley 2010)Through active engagement of opposition views theytry to ldquosway public opinionrdquo (China Digital Space 2016Ng 2011) ldquoinfluence public opinion pretending tobe ordinary citizens and defending or promoting thegovernmentrsquos point of viewrdquo (Lam 2013) or ldquosteerconversations in the right directionrdquo (Economist 2013)Estimates by journalists of the size of the 50c party isbetween 500000 and 2 million (Philipp 2015)

Academics Academics have indicated that between250000 and 300000 paid 50c party members writepseudonymous posts directed by the Chinese govern-ment (Freedom House 2009 Barr 2012 Greitens 2013)Because of the absence of systematic scholarly re-search on the subject academics express a wider rangeof possibilities (and uncertainties) for what 50c partymembers write about However in most cases theirconclusions mirror those of the journalists that 50cparty members generate proregime commentary andargue with its critics Deibert and Rohozinski (201054) describe 50c party members who ldquopatrol chatroomsand online forums posting information favorable to theregime and chastising its criticsrdquo They ldquomix control andactivism on line making favorable comments andgenerally pushing discussion toward pro-Party linesrdquo(Greitens 2013 265) They are an ldquoarmy of onlinecommentators promoting the Chinese CommunistPartyrsquos line on sensitive subjectsrdquo (Bremmer 2010 seealso Hassid 2012) They ldquofacilitate state propagandaand defuse crisesrdquo (Han 2015b) ldquopost comments fa-vorable towards the government policiesrdquo (Tang et al2012 299)ldquodefending the governmentrdquo and ldquofightingrdquothose who ldquocriticize the governmentrdquo (Zhang et al2014 1889) and for example ldquoattack calls for thecountry to launch a lsquojasmine revolutionrdquorsquo (Bambauer2013 29)

Social Media Participants Participants in social mediaregularly characterize 50c party members by openly ac-cusing others of being members themselves To system-atically characterize their views we obtained a randomsample of 9911 social media posts from 2010 to 2015that contain the word ldquordquo (ldquo50c partyrdquo) Fromthese data we drew a sample of 128 posts writtenby people accused in other posts of being 50c partymembers

We then sorted these ldquoaccused 50c postsrdquo into one ofsix categories using a categorization scheme we will usethroughout this article With two independent Chineselanguage coders and 200 randomly selected posts fromthe 9911 posts we measured the intercoder reliabil-ity of the categorization scheme at 93 agreement(see Appendix A for details) Two of the categoriescomprising 65 of the accused 50c posts representthe views of academics and journalists and include (1)taunting of foreign countries (which is 29 of this sam-ple) and (2) argumentative praise or criticism (36of the sample) Taunting includes denigrating favor-able comparisons of China compared to other usuallyWestern countries and taunting of prodemocracy orpro-West values or opinions Argumentative praise orcriticism involves engaged argument and debate aboutcontroversial (nonvalence) issues criticism of oppo-nents of the government or praise of the leaders

The categorization scheme also includes (3) nonar-gumentative praise or suggestions (22 of the sam-ple) and two categories that everyone agrees are notwhat 50c party members are writing about (4) factualreporting (8) and (5) cheerleading (at 5) Nonar-gumentative praise or suggestions includes discussionof noncontroversial valence issues such as improvinghousing or public welfare or praise of government offi-cials but does not debate or take opposing viewpointsCategory (3) does not threaten the regime in any wayand indeed Chen et al (2016) show that local govern-ments openly discuss nonargumentative valence issueswith others on government websites)

Factual reporting involves descriptions of govern-ment programs events initiatives or plans Cheerlead-ing includes expressions of patriotism encouragementand motivation inspirational slogans or quotes grate-fulness discussions of aspirational figures cultural ref-erences or celebrations (Appendix A also includes asixth ldquootherrdquo or ldquoirrelevantrdquo category but we removethis so that the percentages from the first five categoriesadd to 100)

Thus social media participants accusing others ofbeing 50c party members agree with journalists andmost scholars that the content of 50c posts is basicallyantidisestablishmentarianismmdasharguing with those whooppose with the regime its leaders or their policies

We now go a final step and study the identities ofaccused (as distinct from actual) 50c party memberswhich can be difficult because such accusations occuron comment or discussion threads where participantsare anonymous However by careful and extensivecross referencing of profile information across multi-ple platforms we were able to unearth personal detailsfor a handful of these individuals Their backgrounds

486

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44


vary greatly but in each case it seems obvious thatthey are highly unlikely to be real 50c party membersFor example among those accused of being 50c partymembers include Zhou Xiaoping () a bloggerwell known for his anti-West and nationalist sentimentand He Jiawei () a blogger known for critiquesof the Chinese government who posts on Boxun a sitehosted outside of China devoted to covering topics suchas Chinese government human rights abuses Otherwell-known figures accused of being 50c include LinYifu () a Peking University professor who waschief economist and senior vice president of the WorldBank from 2008 to 2012 In none of these cases arethese people likely to be 50c party members Howeverthose accused of being 50c party members also includefigures not connected to politics such as (in our data)a comedian a lawyer and a marketing executive

It appears that the evidence base of those accusingothers of being 50c party members is no better thanthat of academics or journalists Although the priorbeliefs of all three groups about the content of 50c partyposts are almost the same little evidence supports theirclaims

LEAKED INTERNET PROPAGANDA OFFICECOMMUNICATIONS

1 Data and Methods The problem in the literaturehas been that ldquodetecting the Wumao [50c party] isdifficult because there is no ground truth informationabout themrdquo (Yang et al 2015) We are fortunate to beable to change this situation In December 2014 anony-mous blogger ldquoXiaolanrdquo (httpsxiaolanme) releasedan archive of all 2013 and 2014 emails to and somefrom the account of the Internet Propaganda Office() a branch of the propaganda department ofZhanggong District Zhanggong District is a country-level administrative unit (with a population in 2013 of468461) that is part of the moderate-sized GanzhouCity located in Jiangxi Province The emails reportedactivities of Internet commentators including numer-ous 50c posts from workers claiming credit for complet-ing their assignments and many other communicationsThe hack was widely reported and the archive of emailshas been publicly available since (Henochowicz 2014)

The archiversquos large size complicated structure nu-merous attachments diverse document formats (screenshots Word Excel PowerPoint raw text text as part ofother emails etc) multiple email storage formats andmany links to outside information has made digestingmuch of it impossible either for individuals reading andcoding by hand or for existing methods of automatedtext analysis Journalists managed to pull out a fewexamples to write newspaper articles but no systematicanalysis has been conducted of these data

To systematize this richly informative (and essen-tially qualitative) data source we developed and ap-plied a variety of methods and procedures from large-scale hand coding to specially tuned and adaptedmethods of named entity recognition to methods ofautomated text analysis and extraction Because ofthe considerable effort and resources necessary we

FIGURE 1 Network Structure of LeakedEmail Correspondents

Note Circles are email correspondents and edges (lines) in-dicate email correspondence Most of the correspondence istoward the center of the flower-like structure (to the ZhanggongInternet Propaganda Office and then out from that office tohigher-level offices

have made structured and easy-to-access forms of thesedata along with other replication information publiclyavailable in Dataverse so that others may follow up (seeKing et al 2017)

From this work we identified 2341 emails sent fromFebruary 11 2013 to November 28 2014 Of these1208 contained the text of one or (usually many) more50c posts In all from these emails and their attach-ments we harvested 43757 known 50c posts that forma basis for our subsequent analyses and as a trainingset help identify other 50c posts (Although we havethe name direct contact information and often pho-tographs of many of the people discussed in this articlewe have no academic reason to make this informationmore public than it already is and therefore do not doso Other data and replication information is availablein our Dataverse archive see King et al 2017) Weconduct rigorous evaluations of our claims in subse-quent sections For now we characterize the contentwith several separate descriptive analyses

2 Structure We portray the overall structure of com-munications in these emails with the network diagramin Figure 1 Each circle is a specific email account andeach line denotes where one or more emails was sentfrom and to The large flower-like shape at the bottomrepresents 50c party members sending in copies of theirposts to the Zhanggong District Internet PropagandaOffice () claiming credit for completingtheir assignments This office then reports up to otheroffices (see the lines out from the center of the flowershape) including the speaker of Zhanggong PeoplersquosCourt News office (

487

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44


FIGURE 2 Time Series of 43757 Known 50c Social Media Posts with Qualitative Summaries of theContent of Volume Bursts

010

0020

0030

0040

00

Date (Jan 2013 minus Dec 2014)

Cou

nt o

f Pos

ts

Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov

1 Qingmingfestival(April)

2 ChinaDream(May)

3 Shanshanriots (July)

4 3rd plenumCCP 18th

Congress (Nov)

5 Two meetings(Feb)

6 Urumqi railexplosion (May)

7 Govtforumpraisecentralsubsidy

(JulminusAug)

8 MartyrsDay

(Oct)

) and the District Party Office InformationDepartment ()

3 Identifying 50c Party Members Next most of thescholarly literature describes 50c party members as or-dinary citizens hired for very low piecemeal wages Wefound instead that almost all 50c workers in our sam-ple are government employees (consistent with somearguments by Han [2015b]) Of the 43757 posts only281 were made by individuals or groups that we couldnot identify (the content of these posts were very sim-ilar to those we could identify) The remaining 993were contributed by one of more than 200 governmentagencies throughout the Chinese regimersquos matrix or-ganizational structure (of geographic representation byfunctional area) in Zhanggong District including 9159posts (209 of the 43757 total) made directly by theZhanggong Internet Propaganda Office 2343 (54)by the Zhanggong District Bureau of Commerce() 1672 by Shuixi Township ( oneof several townships in Zhanggong) and 1620 by Nan-wai Subdistrict ( one of several subdistrictsin Zhanggong) Others come from functional bureausin Zhanggong District (eg Sports Bureau Bureau of Human Resources and Social Se-curity Bureau of Taxation ZhanggongDistrict court) the government offices of Zhanggongrsquossubdistricts and townships (eg Shahe Town-ship Ganjiang Subdistrict) functional de-partments in each subdistrict or township ( Shuixi Township Party Office) and administrativeoffices of neighborhoods and villages in Zhanggongrsquos

townships and subdistricts (eg Dongyang Shan neighborhood of the Nanwai Subdis-trict Hele village of the Shuixi Subdis-trict)

Of the 50c posts in this archive 2998 did notcontain a URL or a description of the site wherethe content was posted Of the remainder 5338of the 50c posts were comments on governmentsites (GanzhouWeb Newskj DajiangWeb JidanWebJiangxiWeb CCTVWeb RenminWeb JiujiangWebQiangGouWeb) and 4662 were on commercial sitesOf the 50c posts on commercial sites 5398 wereon Sina Weibo 3210 on Tencent Weibo 1075 onBaidu Tieba and 269 on Tencent QZone with therest in the long tail receiving less than 1 each

We also found no evidence that 50c party memberswere actually paid 50 cents or any other piecemealamount Indeed no evidence exists that the authors of50c posts are even paid extra for this work We cannotbe sure of current practices in the absence of evidencebut given that they already hold government and Chi-nese Communist Party jobs we would guess that thisactivity is a requirement of their existing job or at leastrewarded in performance reviews

4 Coordination and Content We now offer a first lookat the 43757 posts from the 50c party we unearthedWe do this by plotting a daily time series of countsof these posts in Figure 2 The most important findingin this graph is that the posts are far from randomlyor uniformly distributed instead being highly focusedinto distinct volume bursts This suggests a high level

488

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44


of coordination on the part of the government Indeedoften the most influential patterns in most social mediaare the bursts that occur naturally when discussions goviral The governmentrsquos manufactured bursts mirrorthese naturally occurring influential patterns but attimes of the governmentrsquos choosing Bursts are alsomuch more likely to be effective at accomplishing spe-cific goals than a strategy of randomly scattering gov-ernment posts in the ocean of real social media (Wealso looked extensively for evidence that 50c posts werecreated by automated means such as bots but the evi-dence strongly indicates to the contrary that each waswritten by a specific often identifiable human beingunder direction from the government)

Although we conduct rigorous quantitative analysesof the content of 50c posts in the sections to follow herewe provide a feel for the content of the posts by labelingthe largest volume bursts in this set (with numbers cor-responding to those in the figure) The labels are briefsummaries we chose from reading numerous posts aprocess we found easy and unambiguous The followinglist gives the first indication that the focus of these postsis on cheerleading possibly for purposes of distractionrather than engaged argumentation and debate

1 Qingming (Tomb Sweeping Day) More than 18000posts about veterans martyrs how glorious orheroic they are and how they sacrificed for China

2 China Dream More than 1800 posts about Pres-ident Xi Jinpingrsquos ldquoChina Dreamrdquo Potentiallya reaction to the April 2013 Peoplersquos Dailypiece instructing municipal governments to carryout China Dream propaganda campaigns (seehttpjmpchinadream)

3 Shanshan Riots 1100 posts immediately followingShanshan riots in Xinjiang At 530 pm Zhang-gong County sent an email to itself (probably BCC-ing many others) highlighting three popular postsabout Xinjiang and identifying this as a terroristincident At 800 pm on the same day ZhanggongCounty sent an email to Ganzhou City to which itreports having created hundreds of 50c posts seem-ingly to distract from the riots about China Dreamlocal economic development and so forth

4 18th Party Congress 3rd Plenum More than 3400posts related to the 3rd plenary session of the Chi-nese Communist Partyrsquos 18th Congress which dis-cussed plans for deepening structural reform

5 ldquoTwo Meetingsrdquo More than 1200 posts aboutGanzhoursquos Peoplersquos Congress and Political Consul-tative Committee meetings and policies to be dis-cussed at the two meetings including factual report-ing of environmental issues one child policy ruralissues as well as growth and development

6 Early May Burst 3500 posts about a variety of top-ics such as mass line two meetings peoplersquos liveli-hood and good governance Immediately followedthe Urumqi railway explosion

7 Praise for Central Subsidy More than 2600 postscelebrating the second anniversary of ldquoCentral So-viet Areas Development policyrdquo () sub-sidies from the central government to promote the

development of region where the original ChineseCommunist Party bases were located (including theregion where Zhanggong is located) at the sametime the local government held an online QampAsession for citizens

8 Martyrrsquos Day 3500 posts about martyrs and thenew Martyrrsquos Day holiday celebrating heroes of thestate

Although we cannot know for certain the exact causeor intended purpose of each burst of 50c party postsFigure 2 is consistent with a strategy of distractionFor example several bursts follow events with ldquocol-lective action potentialrdquo (ie actual or potential real-world crowd formation and related activities see p 6of King et al [2013] for a precise definition) Theseevents include the Shanshan riots and the early Mayburst following the Urumqi railway explosion Otherbursts occur during national holidays when peopleare not working which tend to be prime time peri-ods of political unrest Indeed the Qingming festi-val or Tomb Sweeping Day has historically been afocal point of protests in China and for this reasonwas largely banned during the Maoist era In recentyears Qingming a day on which people pay respectsto the dead has drawn attention to sensitive eventssuch as the deaths of those in the 1989 Tiananmencrisis (Johnson 2016) The central regime and Jiangxiprovinces have both issued notices about the Qingmingfestival as a period when local governments need toincrease their vigilance to prevent protest (see httpjmpjiangxi and httpjmpMinistryCivil) Similarlypolitical meetings are periods when government andparty officials believe that protests are more likely totake place During these periods officials gather andattention is focused on the activities of the regime assuch successful protests can garner greater attentionPrior to these meetings measures such as a preemptiveredistribution and preemptive repression are put intoplace to decrease the likelihood of social mobilization(Pan 2015 Truex 2016)

5 The Purpose of 50c Posts Although our leakedarchive includes specific directions to 50c workers itdoes not reveal whether these directions originate fromZhanggong or from higher levels of the government orparty This and the nearly infinite phenomena that wemight identify as potential precipitating events pre-vents us from determining the immediate cause of ev-ery burst of 50c activity However our inference aboutdistraction being the goal of the regime is consistentwith directions to 50c party members in emails fromthe Zhanggong propaganda department They ask 50cmembers to ldquopromote unity and stability through pos-itive publicityrdquo ( )and ldquoactively guide public opinion during emergencyeventsrdquo () In thiscontext ldquoemergency eventsrdquo are events with collectiveaction potential5

5 For example a website developed by the Ministry of Public Securityand Ministry of Education to help young people better understand

489

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44


We now turn to a more systematic analysis of theseposts their accounts and others like them beyondZhanggong

CONTENT OF 50c POSTS

We now reveal the content of 50c party posts acrossChina by estimating the distribution of these posts overthe five main content categories introduced previously(with details in Appendix A) We do this in five sepa-rate analyses and datasets that successively expand theinitial set of posts from Zhanggong to larger and largerareas across the country

Ex ante we do not know how 50c party activity inZhanggong might differ from that in other countiesOriginally part of the Jiangxi Soviet established in 1931by Mao Zedong Zhu De and other leaders Zhang-gong has a rich revolutionary history These and otherfactors may make it unusual However directives fromthe central government or common interests of dif-ferent counties in keeping their populations in checkmay keep the purpose and content 50c party activity indifferent counties aligned As it turns out for each ofthe five separate analyses and in the survey validationin the next section we find very similar results with 50cparty posts largely comprised of cheerleading and dis-traction rather than engaged argument In other wordsthe patterns found in the leaked data from ZhanggongDistrict do extrapolate

We conclude this section with a sixth part report-ing on an event that occurred during our observationperiod that provides strong evidence of coordinationacross counties and very clear top down control

1 Leaked 50c Posts We first analyze the 43757 50csocial media posts that we harvested from the leakedarchive from Zhanggong These posts were made bynumerous authors on many different social media sitesincluding national-level platforms run by private sectorfirms such as Sina Weibo and Baidu Tieba as well asgovernment forums at the national provincial prefec-tural and county levels To study these data we beganby hand coding a random sample of 200 posts into ourcategories (again with high intercoder reliability)

One result is immediately apparent the number ofposts from this sample that fall in the categories ldquotaunt-ing of foreign countriesrdquo or ldquoargumentative praise orcriticismrdquo is exactly zero This is an important surprise

safety issues (httpjmpEmergEvents) explains ldquoEvery emergencyevent involves the self-interest of a particular group of people lead-ing to psychological pressure and change among this group andunderstandably leading to concern and worry Especially for emer-gency events of a societal nature [as distinct from natural disasters]most are organized by a small group of people who through theirpublicity seeking and encouragement get others involved Recentlyemergency events due to issues like territorial disputes land requi-sitions and housing demolition in certain areas are often organizedby one person and involve many making collective eventsrdquo ( 13)

as it is essentially the opposite of the nearly unanimousviews espoused by scholars journalists activists andsocial media participants This result would be highlyunlikely to have resulted from (binomial) sampling er-ror if the true share of the full set were even as largeas a few percentage points (at 5 which would stillbe a major surprise the probability of seeing the sam-ple that we obtained is essentially zero) To push evenfurther we did extensive searches and reading amongthe remaining posts and finally found a few that fit thiscategory (see the examples in Categories 2 and 3 inAppendix A) but the overall result is that 50c partyposts are extremely rare in these categories

We thus infer that the leaked posts contain very littletaunting of foreign countries or argumentative praiseor criticism we verify this by formally estimating allcategory proportions in the entire set of posts Usinga text-analytic method known colloquially as ReadMe(named for the open source software that implementsit) we estimate the category proportions directly with-out having to classify each post into a category (Hop-kins and King 2010) This is fortunate as individualclassifiers that manage to achieve high (but imperfect)levels of the percentage correctly classified may stillgenerate biased estimates of the category proportionsFor example an estimate indicating that zero countrydyad-years since WWII were at war achieves a pre-dictive accuracy of about 999 but aggregating theseclassifications yields an obviously biased (and useless)estimate of the prevalence of war In contrast ReadMedoes not give individual classifications but it has beenproven to give approximately unbiased and consistentestimates of the category proportions which here isthe relevant quantity of interest The other advantageof ReadMe in this context is that its statistical assump-tions are met by our sampling procedures

The estimated proportions of 50c posts by categoryfor all datasets appear in Figure 3 the results for ourfirst dataset (of all posts found in the leaked emails inZhanggong) are represented by a histogram formedby the set of solid disks (bull) for the point estimate andsolid line for the confidence interval for each of thecategories Other results to be described in the follow-ing in order from left to right within each category alsoappear in the same graph

The categories in Figure 3 are arranged so that thetwo emphasized in the literature appear on the leftand our main empirical results on the right For thisanalysis the results indicate that approximately 80fall within the cheerleading category 14 in nonargu-mentative praise or suggestions and only tiny amountsin the other categories including nearly zero in argu-mentative praise or criticism and taunting of foreigncountries Clearly these results clearly indicate that 50cposts are about cheerleading not argumentation

2 Posts from Leaked 50c Weibo Accounts One possi-bility that we now consider is whether 50c party mem-bers differentially reported cheerleading posts back tothe propaganda department even though they postedabout topics at the behest of the regime from other cat-egories as well To study this question we constructed

490

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44


FIGURE 3 Content of Leaked and Inferred 50c Posts by substantive category (with details inAppendix A) and analysis (given in the legend)

a second dataset by first identifying all Weibo socialmedia accounts revealed in the leaked email archiveWe chose Weibo because it is the most widely usedsocial media site that enables mass distribution andwe were able to obtain access in the manner we neededit We then found these accounts on the web and keptall 498 Weibo accounts that made at least one postFinally we downloaded all social media posts fromthese accounts yielding a set of 167977 knownmdashbutnot previously leakedmdashposts from 50c accounts

We drew a random sample (stratified by account)of 500 of these 167977 social media posts and codedthem into our categories as a training set In this ran-domly selected training set like the last we find noevidence of taunting of foreign countries although wedid find a handful of posts in the category of argumen-tative praise or criticism constituting only 3 of theposts As earlier we then used (a stratified sample and)ReadMe to estimate the five category proportions forthe set of all posts The results reported in the secondbar of the histogram in Figure 3 are very similar tothat from the first dataset The point estimates (por-trayed as solid triangles with confidence intervals asdashed lines) indicate that again the bulk of 50c postsfrom leaked accounts are cheerleading (51) 20 infactual reporting 23 in nonargumentative praise orsuggestions and only 6 in argumentative praise orcriticism

3 Partitioning Leaked Accounts for ExtrapolationWe designed our third analysis to further explore theleaked data and to prepare the ground for extrapola-tion The key idea here was to partition the Sina Weiboaccounts (from Analysis 2) into those easy to identify

outside the leaked archive (which we do for Zhang-gong in Analysis 4 and in other counties in Analysis 5)and those more difficult to identify We developed analgorithm to distinguish these two account types andthen showed that we only need to extrapolate the firsttype because they post the same types of content

To find a useful partition we began by studying thestructure of the 498 known 50c Weibo accounts andtheir 167977 social media posts In each type we oftenfound many commercial posts which fall in our ldquootherrdquocategory (see the Appendix) since we remove andcondition on this category for all analyses we do notdefine account types on this basis either The first typeof account which we call ordinary is used by appar-ently ordinary people in China to post about their chil-dren funny videos commercial advertisements sportsteams pop stars personal opinions and many othersubjects Embedded within the stream of these postsare those which these authors indicate in their com-munication with the propaganda department to be 50cparty posts The second type which we call exclusiveaccounts is (aside from commercial posts) almost ex-clusively devoted to 50c posts Near as we can tell viaextensive cross checking with external data sourcesordinary accounts are genuine registered in the nameof a person (usually a government employee) postingon it whereas exclusive accounts are pseudonymousdesigned solely to fool those who see it In both casesthe 50c posts on these accounts are those directed bythe government rather than necessarily reflecting theopinions of ordinary people

Distinguishing between ordinary and exclusive ac-counts in our leaked archive is easy (the number ofreal 50c posts reported to the propaganda department

491

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44


as a proportion of all posts on the account is a directmeasure) but our goal is to extrapolate to other coun-ties where we have no known 50c posts Thus we needa formal partitioning algorithm to sort accounts intothese two categories without needing the inside infor-mation that we have from our extraordinary leakeddata Moreover since our goal is to determine the con-tent of 50c posts we must be able to discern whetheran account was written by a 50c party member withoutusing the text of the posts

To develop this partitioning algorithm we followedthe logic of ldquoBayesian falling rule listrdquo methodologywhich is accurate and also highly interpretable (Lethamet al 2015) The interpretability also enabled us tocombine qualitative knowledge with modern machinelearning as well as to make choices that were mucheasier to apply outside of Zhanggong With this ap-proach as a guide we found that two simple rules aresufficient to partition our 498 50c accounts into ex-clusive and ordinary First we obtained candidate 50caccounts by collecting all accounts that comment onor forward any post on the Zhanggong governmentrsquosWeibo account (httpweibocomu3880516376) Sec-ond we narrowed this to accounts with 10 or fewerfollowers The result is our definition of exclusive ac-counts These two simple interpretable rules are highlyplausible and consistent with what is known about so-cial media After all accounts that engage with gov-ernment websites and have no more than a handful offollowers are likely used for a very specific purpose(Because of how Weibo differs from platforms likeTwitter users of Weibo accounts with few followerscan still be highly influential by commenting on othermore popular accounts)

We now show that the 50c posts appearing on exclu-sive and ordinary accounts have essentially the sametypes of content where we can verify both To do thiswe applied our partitioning algorithm to the set of 498known 50c accounts from our archive and then com-pared the content of ordinary and exclusive accountsWe found that 202 (41) are exclusive accounts andthe remaining 296 (59) are ordinary accounts Thispartition of the data is neither right nor wrong (andthus statistics like ldquopercent correctly classifiedrdquo do notapply) but it is useful only to the extent that using onlythe exclusive posts causes no bias Thus we estimateand compare the distribution of posts within the ordi-nary and exclusive account types across our five contentcategories To do this we applied ReadMe within eachpartition and compared the results

Fortunately the results are very close to each otherand (as a result) to the overall results we presentedpreviously This implies that bias is unlikely to beinduced by narrowing our search outside our leakedarchive to exclusive accounts Point estimates for thecategory proportions appear in Figure 3 (marked asand in red) For both the bulk of 50c posts appear inthe cheerleading category (46 for exclusive accountsand 58 for ordinary accounts) In contrast the sumof taunting of foreign countries and of argumentativepraise or criticism is very small (5 for exclusive and11 for ordinary)

4 Unleaked 50c Posts in Zhanggong We now use theresults about ordinary and exclusive accounts (fromAnalysis 3) and expand our extrapolation beyond the50c posts in the leaked archive (from Analysis 1) andnew unleaked Sina Weibo posts that we found fromthe accounts identified in the leaked archive (fromAnalysis 2) The key for this extrapolation is that allthree of these analyses yielded very similar estimatesof the distribution of 50c posts across our five categoriesof interest We thus now narrow our extrapolation toWeibo posts from exclusive accounts which are easierto find even though we strongly expect 50c posts to bemade in many different platforms including those runby private firms and different levels of government

In this section we focus on previously unidentified50c posts in Zhanggong To do this we chose exclusiveaccounts (by applying the two rules from the previoussection) With this procedure we found 1031 accountsof which 829 accounts are not mentioned in our leakedarchive We then found and scraped all 22702 socialmedia posts available from the front page of each ofthese accounts Each front page has up to 45 separateposts We then analyzed these posts with ReadMe asearlier

Results from this analysis appear in Figure 3 (withpoint estimates represented by times) The result againis very similar to previous analyses 57 of the postsmade on these accounts engaged in cheerleading 16engaged in factual reporting 22 engaged in nonargu-mentative praise and suggestions about 4 in tauntingof foreign countries and essentially zero in argumen-tative praise or criticism

5 Unleaked 50c Posts in Counties with County Govern-ment Weibo Accounts We now extrapolate to coun-ties across China To do this we started with all 2862counties (and county-level divisions) We then took asour target of inference 50c behavior in 1338 of thesecounties that were structured same way as Zhanggongwith a propaganda department that has a public web-site We then drew a simple random sample of 100 ofthese counties and identified all exclusive accounts anda sample of their social media posts6

To be more specific for each county governmentWeibo account we collected all 151110 postsrandomly sampled up to 200 posts of these identifiedall outside Weibo accounts that commented on orforwarded any one downloaded all metadata fromthose accounts and subsetted to those with 10 orfewer followers We then downloaded the first pagecomprising up to 45 social media posts from eachaccount as our candidate 50c posts

Figure 3 provides our results (with point estimatesrepresented as a diamond ) Again we find verysimilar results highly focused on cheerleading and dis-traction rather than argumentation and criticism 64

6 Many of the remaining 1524 counties have Weibo accounts run bygovernment bureaus and agencies (eg the public security depart-ment and civil affairs department) but not by the county governmentOur informal study of these counties revealed no systematic differ-ences from those we studied but following up with systematic studyin these counties would be a good topic for future research

492

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44


of the posts made on these accounts are categorizedas cheerleading 18 in factual reporting 9 nonar-gumentative praise and suggestions 4 in taunting offoreign countries and only 4 in argumentative praiseor criticism

6 Coordination and Top Down Direction The analy-ses thus far suggest a high level of coordination in thetiming (see Figure 2) and content (see Figure 3) of 50cparty activity Here we offer evidence that these effortsmay be directed from the highest levels of the regime

In late February 2014 Chinese president Xi Jinpingled the first meeting of the Central Leading Groupfor Internet Security and Informatization The meetingwas also attended by two other top leaders Li KeqiangChinarsquos premier and Liu Yunshan head of the ChineseCommunist Party propaganda department During thismeeting President Xi stressed the need for governmentofficials to ldquohave a good grasp of the timing degreeand efficacy of online public opinion guidance so thatonline spaces are clear and uncloudedrdquo () (Xi 2014)Xirsquos phrase public opinion guidance is the official termfor Chinese Communist Party policies and practicesdesigned to control or influence public opinion whichincludes ldquotraditionalrdquo guidance such as Chinese Com-munist Party control of the press as well as newer typesof opinion guidance for social media such as 50c partyactivity censorship and the Great Firewall PresidentXi repeatedly stressed in the meeting the need for theregime to build infrastructure and a solid foundationfor ensuring ldquoInternet securityrdquo (which refers to cy-bersecurity more broadly in addition to public opinionguidance)

As this event occurred near the middle of our datawe can look for evidence that it had an effect Thus wecalculate that over the 2 years we observed in Zhang-gong 50c party members created an average of 77 so-cial media accounts per month Yet 156 accounts werecreated the month of the meeting and 39 the monthafter Similarly in our predicted data an average of 19accounts were created per month Yet they created41 accounts in the month following and 174 in themonth after We interpret these strong patterns as ev-idence that governments all across China respondeddirectly to Xirsquos call

VERIFICATION BY DIRECT SURVEY

We now attempt to go an extra step to verify the accu-racy of our extrapolation presented earlier to predicted50c party members across China To do this we take theunusual step in this context of conducting a samplesurvey of predicted 50c party members along with goldstandard elements designed to validate this method ofvalidation 7

7 We had full IRB approval for our study (although we cannot makepublic our exact question wording in Chinese sample size certainstatistics or the original data) We also added our own additionalethics rules not required by the IRB designed to further protect theidentities of our respondents and to keep our large research team

1 Design We began by creating a large number ofpseudonymous social media accounts This requiredmany research assistants and volunteers having a pres-ence on the ground in China at many locations acrossthe country among many other logistically challeng-ing complications We conducted the survey via ldquodi-rect messagingrdquo on Sina Weibo which enables privatecommunication from one account to another WithIRB permission we did not identify ourselves as re-searchers and instead posed like our respondents asordinary citizens Since information in our archive ap-pears to indicate that government monitoring of 50cparty member activities occur only through voluntaryself-reporting up the chain of command our surveyquestions and the responses are effectively anonymouswhich are conditions that have been shown to makerespondents more sincere in responding to sensitivequestions (Tourangeau et al 2013)

We drew a random sample of social media accountsthat we predicted earlier to be 50c and asked eachwhether the owner of that account was indeed a 50cparty member (in a special manner described in thefollowing) Of course the difficulties of interpretingthese answers is complicated by the fact that our surveyrespondents are conducting surreptitious operationson behalf of the Chinese government designed to foolparticipants in social media into thinking that they areordinary citizens and we are asking them about thisvery activity In most cases the government is also theiremployer and so they have ample incentives to notcomply with our requests or to not comply sincerely

We addressed these uncertainties with two entire ad-ditional surveys designed to provide internal checks onour results as well as a carefully worded survey ques-tion in our anonymous survey context In most surveysresearchers are left trusting the answer perhaps aftera stage of pretesting or cognitive debriefing In oursurvey we are in the unusual position of being able togo further by offering a gold standard validation wherefor some respondents we know the outcome to thequestion that we are posing In other words we ask thesame question of a random sample of known 50c partymembers from our Zhanggong leaked archive If theresults of our survey of predicted 50c party membersgive similar results as this survey then we should havemore confidence in the results

We also fielded a third entire survey that approx-imates the opposite gold standard by asking thoseknown not to be 50c party members To do this wedrew a random sample from Weibo accounts acrossChina among those who do not engage with govern-ment Weibo accounts and have more than 10 followersOur results would be confirmed if the percentage whosay they are 50c in this sample are significantly lowerthan those who acknowledge being 50c in our predicted50c sample A tiny fraction of these accounts may ac-tually be 50c but that would merely bias the results

safe Our rules followed the principle articulated in footnote 20 ofKing et al (2014) of trying to avoid influencing the system that wewere studying which has the added advantage of reducing the chancefor bias

493

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44


against the test of our hypothesis of the difference inmeans from our set of predicted 50c members

The final way we reduce uncertainty is in the de-sign of our survey question We followed best practicesin designing survey questions about sensitive topicsincluding adjusting the perceived social environment(Naher and Krumpal 2012) and using familiar languageand positive ldquoloadingrdquo of sensitive questions (Groveset al 2011) We also studied a large volume of socialmedia interactions both via automated means (Kinget al 2016) and by direct reading and found a waywithin the cultural context to ask the question so thatit would be more likely to elicit a sincere answer Wealso pretested our survey on an independent sampleAlthough preserving the confidentiality of our respon-dents and research team makes us unable to share theexact text of our question here we report a similarversion in English which will also enable us to explainits features

I saw your comment itrsquos really inspiring I want to askdo you have any public opinion guidance management oronline commenting experience

To avoid interfering or influencing the system that weare studying and to avoid putting our respondents in anuncomfortable position the question discusses onlinepropaganda in positive terms We used the terms ldquoopin-ion guidance managementrdquo and ldquoonline commentingrdquowhich is the terminology the government uses to dis-cuss these tasks We avoided terms like ldquo50crdquo whichhas negative connotations for some Instead of ask-ing someone to ldquooutrdquo themselves as a 50c party mem-ber we asked for advice on where the person learnedto write in such a motivating inspiring manner thusavoiding generating defensiveness on part of the re-spondent

2 Results High-quality web surveys have responserates of about 35 (Pew Research Center 2014) Theresponse rate for our survey was almost twice that at65 which although small on an absolute level is en-couraging given our more challenging environment Inaddition unlike most web surveys we were able toperform some checks for selection bias because wecollected available information on our entire targetsample before administering our survey question Thisinformation to test for selection bias included vari-ables such as the number of followers gender yearof creation average number of posts for each monthand enabling geolocation we also observed each ofthese variables within the five separate data sourcesmentioned previously Most tests that we conductedindicated statistically insignificant differences betweenrespondents and nonrespondents The few differencesthat appeared were negligible compared to the largeeffect sizes that we present in the following As mightbe expected the data contained some evidence that 50cparty members are less likely to respond to our ques-tion than non-50c party members which has the effectof making it more difficult to confirm our hypothesisAlthough we could weight the following results by the

TABLE 1 Survey About 50c Status

50c Status Origin Yes ()

Predicted 50c Across China 59Known 50c Leaked Zhanggong archive 57Known ldquoNotrdquo 50c Random sample 19

Note The first line is from our survey the second two are goldstandard evaluation surveys The difference between the firstand second lines is not statistically significant the differencebetween the first and the third is statistically significant (both atα = 005)

differences we found they are small enough that wechose to present the raw unprocessed data instead

The results for our three surveys appear in Table 1Overall we found that 59 of our predicted 50c partymembers admitted to being 50c party members If weare correct that they are all 50c party members thenthe remaining 41 gave an insincere answer whichwould not be surprising given that doing so is essen-tially their job To test this we used our gold standardsample of known 50c party members revealed in ourleaked Zhanggong email archive In this sample 57admitted to their 50c party status The two percentagepoint difference between these two figures is not statis-tically significant (at = 005) suggesting that indeedall respondents in our predicted sample are 50c

Also as a test we use our gold standard samplethat approximates those known to not be 50c partymembers In this sample only 19 said that they were50c the substantial 40 percentage point difference be-tween this figure and that from our predicted 50c partymember sample (59) is very large and statisticallysignificant revealing a strong signal of actual 50c partymembership among our predicted 50c sample (Near aswe can tell if we had asked much more directly whetherour respondents were 50c party members those whowere not would have responded with angry denialsThis would have had the advantage of dropping the19 figure nearer to 0 but it would likely also havethreatened our entire project The survey would alsohave failed because then few or no actual 50c partymembers would have answered our survey question)Overall the results from this survey strongly supportthe validity of the predictions of 50c party membershipconducted previously

SIZE OF THE 50c PARTY

In this section we study how widespread 50c activity isacross the country Overall we find a massive govern-ment effort where every year the 50c party writes ap-proximately 448 million social media posts nationwideAbout 527 of these posts appear on governmentsites The remaining 212 million posts are inserted intothe stream of approximately 80 billion total posts oncommercial social media sites all in real time If theseestimates are correct a large proportion of governmentwebsite comments and about 1 of every 178 social me-dia posts on commercial sites are fabricated by the

494

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44


government The posts are not randomly distributedbut as we show in Figure 2 are highly focused anddirected all with specific intent and content The restof this section explains how we estimate these numbersThroughout in lieu of the possibility of formal standarderror calculations we offer transparent assumptionsthat others can easily adjust to check sensitivity or im-prove as more information is unearthed

1 Number of Social Media Posts To understand thecontext into which 50c posts are inserted we began byestimating the total number of Chinese social mediaposts nationwide As of December 2012 netizens wereposting approximately 100 million messages a day or365 billion a year on Sina Weibo alone (Zhao et al2014) which is one of at least 1382 known social me-dia sites (King et al 2013) In our data the ratio ofSina Weibo posts to all posts is 185 meaning that anestimate of the total number of posts on all platformsis (185 times 365 billion =) 675 billion However thisrequires the strong assumption that 50c party mem-bers use specific commercial social media platforms inthe same proportions as the entire user populationWe therefore used the detailed survey from iiMediaResearch Group (2014) and calculated the ratio of to-tal posts to Sina Weibo posts to be 210 and the totalnumber of posts per year to be about 804 billion Thisis an underestimate because it is based on microblogsand ignores blogs but blogs probably number in themillions which is rounding error on this scale

2 Number of 50c Posts in Zhanggong Among the43757 confirmed 50c posts 30215 were made duringa 365-day period between February 11 2013 (the firstday on which we observed a 50c post) and February10 2014 We have evidence of at least 1031 exclusive(Sina Weibo) accounts in Zhanggong including 202 ac-counts in the leaked archive and 829 that we identifiedoutside the archive (by following the rules presentedpreviously

In our archive a 50c party member needing to makea post chooses an exclusive account on Weibo (68943757 =) 157 of the time compared to all other choices(an ordinary account on Weibo or another social mediasite) We assume that this ratio is approximately thesame for nonleaked 50c posts in Zhanggong which inturn implies that the ratio of total 50c posts to 50c postsin the archive is the same as the ratio of total exclusiveaccounts to exclusive accounts in the archive As suchan estimate of the total number of posts in Zhanggongin 2013 is (30 215 times 1 031202 =) 154216

3 Number of 50c Posts in Jiangxi Province Zhang-gong is an urban district of Ganzhou City within JiangxiProvince According to the 2014 China Internet Net-work Information Centerrsquos Statistical Report on Inter-net Development in China the 2013 Internet penetra-tion of urban residents was 620 and of rural residentswas 275 (CNNIC 2014) According to the NationalBureau of Statistics of China 4887 of the 4522 mil-lion people in Jiangxi Province lived in urban areas or2210 million with 2312 million living in rural areas(National Bureau of Statistics of China 2014)

We first compute the number of 50c posts per Inter-net user in Zhanggong which is (154 216468 461 times062 =) 0531 We then assume that this rate is roughlythe same in Jiangxi and then scale up Thus we esti-mate the total number of 50c posts in Jiangxi during2013 as (0531 times [062 times 221M + 0275 times 231M] =)1065 million

4 Number of 50c Posts in China Finally to scalethis result to all of China we assume that the ratioof 50c posts to Internet users in other parts of China isroughly the same as in Jiangxi This ratio of posts perInternet user is (1065M1468M =) 07255 Applyingthis assumption to the country as a whole reveals thepresence of (07255 times 61758M =) 4480 million 50cposts in China during 2013 (see CNNIC 2014)

WHAT MIGHT BE WRONG

Inferences in this article depend on the veracity of theleaked archive that we analyze The size and extraor-dinary complexity of this archive makes it highly likelyto be real There are no signs of it having been gener-ated by automated means and fabricating it by handto mislead would have been a monumental task Wealso verified numerous external references from thedatamdashto specific individuals email addresses phonenumbers government departments programs web-sites social media accounts specific posts etcmdashandevery one checks out Nevertheless we have no infor-mation about how the leak actually occurred

Chinese government astroturfing efforts may existthat do not follow the model that we unearthed inZhanggong For example based on anecdotal evidencethat we came across it is possible that the public secu-rity bureaucracy and Communist Youth League mayalso be involved in fabricating social media content Itis possible that other organizations may hypotheticallyfollow different rules and practices perhaps varyingin different places and may generate 50c posts withdifferent types of content Determining whether it ismust wait for new evidence to be unearthed Perhapsthe window that this article opens on this large and pre-viously opaque government program may help othersdiscover different aspects of it in China and eventuallyin other related authoritarian regimes

We have observed that the content of 50c party postsacross China is largely about cheerleading and to alesser extent nonargumentative praise or suggestionsand factual reporting Since humans have highly lim-ited attention spans and the volume of informationcompeting for their attention is growing quickly in thedigital age huge bursts of irrelevant posts about cheer-leading will certainly be distracting to at least somedegree We are not able to quantify how distractingthese posts are in practice or as a result the overalleffectiveness of 50c strategy Our results do suggestsome interesting experiments that could be run by fu-ture researchers

We have also gone another step and inferred thatthe purpose of 50c activity is to (1) to stop argu-ments (for which distraction is a more effective than

495

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44


counterarguments) and (2) to divert public attentionfrom actual or potential collective action on the groundAs inferences these are by definition more uncertainthan observations and so we now briefly consider fivealternative possible interpretations of our evidence

First perhaps 50c activity is a simple extension ofthe traditional functions of the propaganda system andnot always focused on collective action This point isdefinitely possible that propaganda workers engage incheerleading because they are not motivated to exceland because they are guided by what Han (2015b) de-scribes as a ldquopersistent state propaganda logicrdquo thatcontravene covert activity However the cheerleadingthat we identify departs from the traditional focus ofthe Chinese Communist Party propaganda departmenton guiding the content of media and shaping publicopinion (Brady 2009 Lynch 1999) In addition wehave offered clear evidence that most 50c posts fromour data appear in highly coordinated bursts aroundevents with collective action potentialmdasheither afterunexpected events or before periods of time such asthe Qingming festival and political meetings when col-lective action is perceived by the regime to be morelikely Of course it may also be that these bursts of 50cposts have different purposes depending on the needas perceived by the regime

Second it may be that cheerleading about (essen-tially) irrelevant topics merely creates a general senseof positiveness that transfers over to positiveness aboutother things including the regime This may well betrue but such an effect is not likely to be large Thishypothesis would however be testable by experimentperhaps even in a lab setting

Third might the purpose of 50c posts be to dilutenegative opinion through generally positive cheerlead-ing In fact this is unlikely as 50c posts are aboutirrelevant issues and thus do not change the balanceof positive versus negative comments It is true that50c posts do change the percentage of negative com-ments as a proportion of all posts but more research isneeded to determine how 50c posts interact with char-acteristically bursty and highly variable social mediaposts about every possible issue unrelated to politicsand whether the influx of 50c comments to change thepercentage of negative comments as a proportion ofall posts has any tangible effect on public beliefs andperceptions

Fourth perhaps the point of 50c activity is to signal tothe people that they are under surveillance Althoughwhen sent through censorship a signal like this maybe effective in getting people to self-censor their postsand other activities which posts are 50c is not knownto the Chinese people and so this strategy if it exists isunlikely to be successful

Finally we might ask whether some of the few postsappearing in the empirically small categories of nonar-gumentative praise or factual reporting might actuallybe sarcastic backhand ways of making arguments Thisis possible but our methods are human led and com-puter assisted and thus such sophisticated and subtlearguments would have to confuse our human codersand yet still not mislead Chinese social media partic-

ipants In fact even in the unlikely situation where100 of these posts were misclassified from argu-mentative praise or suggestions most would still becheerleading and our conclusions would remain largelyunchanged

THEORETICAL IMPLICATIONS

The empirical results offered earlier seem clear butwhat do they suggest about the overall strategy of theChinese government or for authoritarian regimes ingeneral We first explain these results by generalizingprior findings on (human) censorship and (automated)filtering all led by the same propaganda department inthe same government as the 50c party (King et al 20132014) We then extend these ideas to the authoritarianliterature in general

1 China One way to parsimoniously summarize ex-isting empirical results about information control inChina is with a theory of the strategy of the regimeThis theory which as with all theories is a simplificationof the complex realities on the ground involves twocomplementary principles that the Chinese regime ap-pears to follow one passive and one active The passiveprinciple is do not engage on controversial issues do notinsert 50c posts supporting and do not censor posts crit-icizing the regime its leaders or their policies The sec-ond active principle is stop discussions with collectiveaction potential by active distraction and active censor-ship Cheerleading in directed 50c bursts is one way thegovernment distracts the public although this activitycan be also be used to distract from general negativitygovernment-related meetings and events with protestpotential and so forth (Citizens criticize the regimewithout collective action on the ground in many waysincluding even via unsubstantiated threats of protestand viral bursts of online-only activitymdashwhich by thisdefinition do not have collective action potential andthus are ignored by the government)

These twin strategies appear to derive from the factthat the main threat perceived by the Chinese regimein the modern era is not military attacks from for-eign enemies but rather uprisings from their own peo-ple Staying in power involves managing their govern-ment and party agents in Chinarsquos 32 provincial-levelregions 334 prefecture-level divisions 2862 county-level divisions 41034 township-level administrationsand 704382 village-level subdivisions and somehowkeeping in check collective action organized by thoseoutside of government The balance of supportive andcritical commentary on social media about specific is-sues in specific jurisdictions is useful to the govern-ment in judging the performance of (as well as keepingor replacing) local leaders and ameliorating other in-formation problems faced by central authorities (Dim-itrov 2014andashc Wintrobe 1998) As such avoiding anyartificial change in that balancemdashsuch as from 50c postsor censorshipmdashcan be valuable

Distraction is a clever and useful strategy in informa-tion control in that an argument in almost any humandiscussion is rarely an effective way to put an end to

496

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44


an opposing argument Letting an argument die orchanging the subject usually works much better thanpicking an argument and getting someonersquos back up(as new parents recognize fast) It may even be thecase that the function of reasoning in human beingsis fundamentally about winning arguments rather thanresolving them by seeking truth (Mercier and Sperber2011) Distraction even has the advantage of reducinganger compared to ruminating on the same issue (Den-son et al 2012) Finally since censorship alone seemsto anger people (Roberts 2014) the 50c astroturfingprogram has the additional advantage of enabling thegovernment to actively control opinion without havingto censor as much as they might otherwise

2 Authoritarian Politics For the literature on author-itarian politics in general our results may help re-fine current theories of the role of information andparticularly what is known as common knowledge intheories of revolutionary mobilization Many theoriesin comparative politics assume that autocrats slow thespread of information critical of the regime to minimizethe development of common knowledge of grievanceswhich in turn may reduce the probability of mobiliza-tion against the regime The idea is that coordination isessential to revolution and coordination requires somecommon knowledge of shared grievances (Chwe 2013Egorov et al 2009 Hollyer et al 2014 Persson andTabellini 2006 Tilly 1978)

In contrast our results suggest that the Chineseregime differentiates between two types of commonknowledgemdashabout specific grievances which they al-low and about collective action potential which theydo a great deal to avoid Avoiding the spread of com-mon knowledge about collective action events (and notgrievances) is consistent with research by Kuran (19891991) Lohmann (1994) and Lorentzen (2013) whofocus specifically on the spread of information aboutreal-world protest and ongoing collective action ratherthan the generic spread of common knowledge morebroadly

The idea is that numerous grievances of a populationruled autocratically by nonelected leaders are obviousand omnipresent Learning of one more grievance inand of itself should have little impact on the power ofa potential revolutionary to ignite protest The issuethen appears not to be whether such grievances arelearned by large enough numbers to foment a revolu-tion Instead we can think of creative political actorsincluding those aspiring to lead a revolution or coup astreating issues ideologies events arguments ideas andgrievances as ldquohooks on which politicians hang theirobjectives and by which they further their interestsrdquoincluding interests that entail initiating or fostering apolitical uprising (Shepsle 1985) If one hook is notavailable they can use another

By this logic then common knowledge of grievancesis already commonplace and thus allowing more infor-mation about them to become public is of little risk tothe regime or value to its opponents Since disruptingdiscussion of grievances only limits information that isotherwise useful to the regime the leaders have little

reason to censor it argue with it or flood the net withopposing viewpoints What is risky for the regime andtherefore vigorously opposed through large-scale cen-sorship and huge numbers of fabricated social mediaposts is posts with collective action potential

CONCLUDING REMARKS

Academics and policymakers have long been focusedon contested physical spaces over which military warshave been or might be fought For example in theSouth China Sea the Chinese regime is presently build-ing artificial islands and the United States is conduct-ing military exercises both highly expensive shows ofpower As important as this focus may be we believethat scholars and policymakers should focus consid-erably more effort on the Chinese Internet and itsinformation environment which is a contested virtualspace one that may well be more important than manycontested physical spaces The relationship betweenthe government and the people is defined in this spaceand thus the world has a great interest in what goeson there We believe that considerably more resourcesand research should be devoted to this area Whateverthe appropriate relationship between governments andtheir people a reasonable position is that it be open andknown This is an area where academic researcherscan help By devoting great effort they can open upthis knowledge to the world It is our hope that othersfollow up on the research reported here

More specifically most journalists activists partic-ipants in social media and some scholars have un-til now argued that the massive 50c party is devotedto engaging in argument that defends the regime itsleaders and their policies Our evidence indicates theoppositemdashthat the 50c party engages in almost no ar-gument of any kind and is instead devoted primarilyto cheerleading for the state symbols of the regimeor the revolutionary history of the Communist PartyWe interpret these activities as the regimersquos effort atstrategic distraction from collective action grievancesor general negativity and so forth

It also appears that the 50c party is mostly com-posed of government employees contributing part timeoutside their regular jobs not as has been claimedordinary citizens paid piecemeal for their work Thisnevertheless is still an enormous workforce that weestimate produces 448 million 50c posts per year Theireffectiveness appears maximized by the effort we foundof them concentrating the posts into spikes at appro-priate times and by directing about half of the posts tocomments on government websites

Appendix A CATEGORIZATION SCHEME

Our categorization scheme for social media posts includesthe six categories below along with examples of each Non-Chinese speakers should be aware when reading these exam-ples that the Chinese language even on social media tendsto be quite flowery and formal with frequent creative andoften (to English speakers) stagy-sounding wordings

497

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44


(1) Taunting of Foreign Countries Favorable compar-isons of China to other countries insults to other coun-tries taunting of pro-democracy pro-West pro-individualliberties or pro-capitalist opinions within China Exam-ples from leaked Zhanggong 50c posts

bull 23ldquo13131313rdquo[Last year at the Shangri-la Dialogue where Obamainvited 23 countries to participate in the containmentof China he said ldquoChina has 13 billion people thefaster China rises the more difficult it will be for usto live because the earthrsquos resources are limited Forus to remain at our current living standard we mustcontain Chinarsquos developmentrdquo]

bull 131313mdashmdash [Chinarsquos rise is now inevitable On onehand the US publicly asserts that if China does notperish the West will wither on the other hand it tellsthe Chinese people that your government is problem-atic you have to overthrow it so you can live a betterlife Is there a more ridiculous and contradictory logicthan this]

(2) Argumentative praise or criticism Comments oncontroversial ProCon (non-valience) issues as well asclaims of wrongdoing or unfairness praise (usually of thegovernment) or criticism (usually of opponents of the gov-ernment) taking a position or explaining why a particularviewpoint is correct or (more often) wrong These postsare often part of a debate in opposition to a previous postExamples from leaked Zhanggong 50c posts

bull 1313 [My dear friends you ifyou go through your Weibo yoursquoll discover that thesystem automatically had you follow Xue Manzi LiKaifu Zuo Yeben Han Han Li Chengpeng and otherpopulist Weibo users This is a typical tactic of indoc-trination and brainwashing I suggest you unfollowthem]

bull 136013 13 [Li Kaifu says that youcan buy a villa for $600000 USD in New York muchcheaper than in Beijing But what he doesnrsquot tell you isthat this so-called villa is actually a warehouse whichis more than a four hour drive from New York City]

(3) Non-argumentative Praise or Suggestions Noncon-troversial valience issues which are hard to argue againstsuch as improving housing and public welfare praiseof current government officials programs or policies Itdoesnrsquot respond to alternative opposing viewpoints andit includes positive sentiment It is distinguished from cat-

egory (2) in that it praises something specific such as thegovernment its officials government programs or initia-tives but does not take issue with another post Includesa small number of constructive suggestions for what gov-ernment policies might include (ie added benefits ratherthan critical complaints) It does not argue against a spe-cific viewpoint but just says ldquoit would be nice if the gov-ernment did Xrdquo which usually the government is alreadyin the process of implementing Some examples of knownZhanggong 50c posts

bull [The government has done a lot of practicalthings among which is solving a significant part of thehousing problem]

bull 13 13 [The policy of renovating mud-brick houses hasallowed villagers to move out of mud-brick dwellingsinto small Western-style buildings The village hasbeen transformed we are so grateful]

bull 13 [We hope the central govern-ment provides us with even more support]

bull 13 [We hopethere will be more good policies like rdquoVarious Opin-ionsrdquo (the abbreviated name of an economic devel-opment policy)]

bull 13 [We lookforward to the leadership of our party secretaryWehope that he can carry out more policies that willbenefit the people in different aspects especially ineducation and health care]

(4) Factual Reporting Descriptions of current govern-ment programs projects events or initiatives or plannedor in progress initiatives Does not include any praise ofthese programs or events (which would be category (3))just that they are occurring Reporting on what govern-ment government officials are doing Some examples ofknown Zhanggong 50c posts

bull 137 [During the Qing-ming festival three-day holiday [the freeway] will re-main free to 7-seater buses]

bull 613271313$1313$1313[On June 27 the Jiangxi provincial committee pro-mulgated an opinion to learn from comrade ZhenGongquan calling on all provincial party membersand cadres to study Zhen Gongquanrsquos firm convic-tion staunch support of the Partyrsquos spirit service tothe masses straightforward dedication to the peopledevotion to duty abiding dedication indifference tofame and fortune selfless dedication to moral charac-ter and hardwork]

bull 1131613amp [On January 16 Jiangxi Party Committee

498

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44


Member and Ganzhou City Party Secretary Shi Wen-qing will communicate with netizens on the ChinaGanzhou Web to hear comments suggestions anddemands from netizens]

(5) Cheerleading for China Patriotism encouragementand motivation inspirational quotes and slogans inspi-rational quotes from government officials thankfulnessgratefulness inspiration or thankfulness for historical andaspirational figures or events and cultural references andcelebrations (eg describes traditions actions suggestionsfor the community) Excludes positive sentiment towardparticular government leaders or specific policies (whichwould be category (3)) but includes positive sentiment orgeneral praise toward life historical figures model citizens(eg Lei Feng Gong Quanzhen a model teacher GuoChuhui a patriotic villager) or China in general Someexamples of known Zhanggong 50c posts

bull 13 1313 13 [Many revolutionary martyrsfought bravely to create the blessed life we have to-day Respect these heroes]

bull 1313 [Respect to all the peoplewho have greatly contributed to the prosperity andsuccess of the Chinese civilization The heroes of thepeople are immortal]

bull [[I will] carry the red flag stained with the blood ofour forefathers and unswervingly follow the path ofthe CCP]

bull 1313 [We allhave to work harder to rely on ourselves and to takethe initiative to move forward]

bull [I love China]bull [[If] everyone

can live good lives then the China Dream will berealized]

bull 1313 [Way to go Ganzhou]

(6) Other Irrelevant posts that are entirely personalcommercial (such as ads) jokes or empty posts that for-ward information not included This category is removedand conditioned on in all analyses in this article

Appendix B AN UNINTENDED ldquoSURVEYrdquoOF THE CHINESE GOVERNMENT

We describe here a rare tacit confirmation of the existence ofthe 50c party as well as an apparent admission to the accu-racy of our leaked archive and the veracity of our empiricalresults all unexpectedly offered by the Chinese governmentin response to our work

Due to a set of unusual and unintended circumstances anearly draft of this article received considerable internationalattention so much so that the Global Times wrote an editorialabout it (this is a newspaper published by the Peoplersquos Dailythe CCPrsquos primary mouthpiece see Wade 2016)8 Although

8 We shared an early draft of this article privately with colleaguesand others to solicit comments We were in the process of revising

this editorial is not an official statement of the Chinese gov-ernment it is reasonable to interpret it as a close approx-imation or at worst of a faction of the government (Weoffer a translation of the editorial along with a contextualexplanation of its content in the Supplementary Appendixto this paper)

The main purpose of the editorial is to strongly defend thegovernmentrsquos unique system of public opinion guidance (seeitem 6 in Section 4) The editorial claims that ldquoChinese societyis generally in agreement regarding the necessity of lsquopublicopinion guidancerdquorsquo (ldquordquo) To understand the governmentrsquos position andperspective it is helpful to use the viral discussion of our pa-per in social media following its unexpected news coverageto test the editorialrsquos claim To do this we downloaded postsfrom two sources comments on the Global Times site and abroader sample from Weibo responding to the editorial Weused ReadMe as above to analyze each corpus separately

We would expect more support for public opinion guid-ance from comments on a nationalist newspaper websiteand much less support (than the regime acknowledges) froma more general population (consistent with Roberts 2014)Indeed this is just what we found Our estimates indicatethat 82 of the comments on the newspaperrsquos website whichexpressed an opinion supported Chinarsquos system of publicopinion guidance (with 15 critical) Yet among the likelybroader audience found on Weibo only 30 were supportive(with 63 critical) clearly contradicting the editorialrsquos rosyview of the governmentrsquos popularity

The fact that the regimersquos central strategy for controllingthe dynamic and highly contested social media space lacksuniversal support likely made the regime feel it all the moreurgent to defend public opinion guidance in this forum Au-thoritarian regimes like China with strong international andmilitary power are usually focused on threats to their rulefrom their own people rather than in this case the interna-tional press (or scientific community) Confirming the fol-lowing four points central to our article (as opposed to deny-ing their previously surreptitious behavior) was of incidentalrelevance to government leaders but served the purpose ofenabling them to engage the discussion and explicitly defendtheir information control practices

First although the Global Times has English and Chineseeditions with many articles published in both languages theeditorial about our paper was published only in Chinese Thatis even though it objected to how the story was covered in theinternational press the CCP was primarily addressing its ownpeople This seems to be a regular strategy of the regime andis consistent with our interpretation of their main perceivedthreats being their own people rather than Western powers

when a reporter from a major international news outlet somehowobtained a copy and contacted us about publishing a story about itWe asked him to hold off because we had not finished revising nor ofcourse was it even under review yet He refused explaining that ourresults were too important and he did not want to be scooped by hisjournalistic competitors Freedom of the press obviously gave us norecourse and so we gave in answered his questions and posted thepaper on our web sites Within a few days over 5000 news outletsacross the globe published stories on our paper (according to Googlenews) The reporter who broke the story turned out to be right ashis next fastest competitor published only a few hours after he didfollowed soon by many others

499

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44


Second the editorial appears to admit to the existenceof the 50c party and at least tacitly confirms the veracity ofour leaked archive They made these admissions apparentlyin order to turn the conversation into an explanation fortheir people about why public opinion guidance is essentialThey also use the editorial to explain that traditional publicopinion guidance is no longer sufficient to prevent the in-crease in viral messaging under control of those outside thegovernment which can spark or fuel collective action Due tothe rise of social media the editorial says the government hasldquono choicerdquo but to implement stronger information controlpractices designed for this new form of communication suchas 50c party activity In other words the 50c party exists butthe Chinese people should not be focused on it

Third in a forum that regularly expresses opinions includ-ing disapproval and disagreement the editorial began witha summary of our empirical results and took no issue withany of our conclusions9 Thus for all practical purposes theeditorial constitutes the answer to a simple sample surveyquestion That is instead of asking 50c party members abouttheir status as we do in Section 5 we (inadvertently) asked theChinese government whether they agreed with our resultsand they effectively concurred Although social scientists of-ten conduct interviews of individual public officials we aregrateful for the unusual if not unprecedented chance to posequestions to an organ of the Chinese government and haveit respond for all practical purposes as a government or atleast in a way that represents it

Finally in the editorial the government also acknowledgesthat the purpose of public opinion guidance is to constrainor stop the spread of ldquohot button issuesrdquo that go viral on-line or ldquograssroots social issuesrdquo that have collective actionpotential This also confirms a central point of our work

SUPPLEMENTARY MATERIALS

To view supplementary material for this article pleasevisit httpsdoiorg101017S0003055417000144

Replication files can be found at httpsdoiorg107910DVNQSZMPD

REFERENCES

Bambauer Derek E 2013 ldquoCensorship v31rdquo IEEE Internet Com-puting 17(3) 26ndash33

Bandurski David 2008 ldquoChinarsquos Guerrilla War for the Webrdquo FarEastern Economic Review 171(6) 41ndash44

Barr Michael 2012 ldquoNation Branding as Nation Building ChinarsquosImage Campaignrdquo East Asia 29(1) 81ndash94

Brady Anne -Marie 2009 Marketing Dictatorship Propaganda andThought Work in Contemporary China Lanham MD Rowmanamp Littlefield

Bremmer Ian 2010 ldquoDemocracy in Cyberspacerdquo Foreign Affairs89(6) 86ndash92

9 The editorial did take one issue with our work (in addition todisagreeing with how the international press covered it) which wasto imply that we had confused 50c party posts with those from othergroups unauthorized by the Chinese government but involved insimilar activities In fact the only posts we analyzed and called 50cwere those officially generated by the government In case this wasa misunderstanding we clarify this point by adding what is nowFootnote 2 to our paper

Cairns Christopher and Allen Carlson 2016 ldquoReal-World Islandsin a Social Media Sea Nationalism and Censorship on WeiboDuring the 2012 DiaoyuSenkaku CrisisrdquoChina Quarterly 22523ndash49

Chen Jidong Jennifer Pan and Yiqing Xu 2016 ldquoSources of Author-itarian Responsiveness A Field Experiment in Chinardquo AmericanJournal of Political Science 60(2) 383ndash400

China Digital Space 2016 ldquoFifty Centsrdquo China Digital Timeshttpjmpcdt50cent Chwe Michael Suk-Young 2013 RationalRitual Culture Coordination and Common Knowledge Prince-ton NJ Princeton University Press

China Internet Network Information Center (CNNIC) 2014 Sta-tistical Report on Internet Development in China Beijing CNCNNIC httpjmpStatDevC

Chwe Michael Suk-Young 2013 Rational ritual Culture coordina-tion and common knowledge Princeton NJ Princeton UniversityPress

Deibert Ronald and Rafal Rohozinski 2010 ldquoLiberation vs Con-trol The Future of Cyberspacerdquo Journal of Democracy 21(4)43ndash57

Denson Thomas F Michelle L Moulds and Jessica R Grisham2012 ldquoThe Effects of Analytical Rumination Reappraisal andDistraction on Anger Experiencerdquo Behavior Therapy 43(2)355ndash64

Dimitrov Martin K 2014a ldquoInternal Government Assessments ofthe Quality of Governance in Chinardquo Studies in Comparative In-ternational Development 50(1) 50ndash72

Dimitrov Martin K 2014b ldquoTracking Public Opinion Under Au-thoritarianismrdquo Russian History 41(3) 329ndash53

Dimitrov Martin K 2014c ldquoWhat the Party Wanted to Know Citi-zen Complaints as a lsquoBarometer of Public Opinionrsquo in CommunistBulgariardquo East European Politics and Societies 28(2) 271ndash95

Economist 2013 ldquoCat and Mouse How China Makes Sure itsInternet Abides by the Rulesrdquo The Economist httpwwweconomistcomnewsspecial-report21574629-how-china-makes-sure-its-internet-abides-rules-cat-and-mouse

Egorov Georgy Sergei Guriev and Konstantin Sonin 2009 ldquoWhyResource-Poor Dictators Allow Freer Media A Theory and Ev-idence from Panel Datardquo American Political Science Review103(4) 645ndash68

Freedom House 2009 ldquoChina and the Internet An Uphill Fight forFreedomrdquo Harvard International Review 31(2) 68ndash73

Greitens Sheena Chestnut 2013 ldquoAuthoritarianism Online WhatCan We Learn from Internet Data in NondemocraciesrdquoPS Po-litical Science and Politics 46(02) 262ndash70

Groves Robert M Floyd J Fowler Jr Mick P CouperJames M Lepkowski Eleanor Singer and Roger Tourangeau2011 Survey Methodology Vol 561 John Wiley amp Sons Pleaseprovide publisher location for Groves reference

Haley Usha 2010 ldquoChinarsquos Fifty Cent Party for Internet Pro-pagandardquo Huffington Post httphuffingtonpostcomusha-haleychinas-fifty-cent-party-f 1 b 749989html

Han Rongbin 2015a ldquoDefending the Authoritarian Regime On-line Chinarsquos lsquoVoluntary Fifty-Cent ArmyrsquordquoChina Quarterly 2241006ndash25

Han Rongbin 2015b ldquoManufacturing Consent in CyberspaceChinarsquos lsquoFifty-Cent ArmyrsquordquoJournal of Current Chinese Affairs44(2) 105ndash34

Hassid Jonathan 2012 ldquoSafety Valve or Pressure Cooker Blogs inChinese Political Liferdquo Journal of Communication 62(2) 212ndash30

Henochowicz Anne 2014 ldquoThousands of Local Internet Pro-paganda Emails Leakedrdquo China Digital Times httpjmpleakedEmails

Hollyer James R B Peter Rosendorff and James R Vree-land 2014 ldquoMeasuring Transparencyrdquo Political Analysis 22(4)413ndash34

Hopkins Daniel and Gary King 2010 ldquoA Method of AutomatedNonparametric Content Analysis for Social Sciencerdquo AmericanJournal of Political Science 54 (1) 229ndash47 httpjmpjNFDgI

iiMedia Research Group 2014 ldquo2013 (rdquo China Mobile Social Share Annual Report () httpiimediacn37109html

Johnson Ian 2016 ldquoThe Presence of the Past A Codardquo In TheOxford Illustrated History of Modern China ed Jeffrey N Wasser-strom Oxford UK Oxford University Press 301ndash324

500

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44


Keating Joshua 2011 ldquoChinarsquos lsquo50-Centrsquo Party Takes on the JasmineRevolutionsrdquo Foreign Policy httpforeignpolicycom20110301chinas-50-cent-party-takes-on-the-jasmine-revolutions

King Gary Jennifer Pan and Margaret E Roberts 2013 ldquoHowCensorship in China Allows Government Criticism but SilencesCollective Expressionrdquo American Political Science Review 1071ndash18 httpjmpLdVXqN

King Gary Jennifer Pan and Margaret E Roberts 2014 ldquoReverse-Engineering Censorship in China Randomized Experimentationand Participant Observationrdquo Science 345 (6199) 1ndash10 httpjmp1KbwkJJ

King Gary Patrick Lam and Margaret Roberts In PressldquoComputer-Assisted Keyword and Document Set Discovery fromUnstructured Textrdquo American Journal of Political Science Copyat httpjmp2nxUa8N

King Gary Jennifer Pan and Margaret E Roberts 2017 ldquoReplica-tion Data for How the Chinese Government Fabricates Social Me-dia Posts for Strategic Distraction Not Engaged Argumentrdquo Har-vard Dataverse V1 [UNF63IIELdmrcyZm+v5mx0OJg==]DOI httpsdoi107910DVNQSZMPDURLdxdoiorg107910DVNQSZMPD

Knockel Jeffrey Masashi Crete-Nishihata Jason Q NgAdam Senft and Jedidiah R Crandall 2015 ldquoEvery RoseHas Its Thorn Censorship and Surveillance on Social VideoPlatforms in Chinardquo In Proeedings of the 5th USENIX Workshopon Free and Open Communications on the Internet (FOCI rsquo15)

Kuran Timur 1989 ldquoSparks and Prairie Fires A Theory of Unan-ticipated Political Revolutionrdquo Public Choice 61(1) 41ndash74

Kuran Timur 1991 ldquoNow Out of Never The Element of Surprisein the East European Revolution of 1989rdquo World Politics 44(1)7ndash48

Lam Oiwan 2012 ldquoChina Challenging the 50 Cent Partyrdquo GlobalVoices httpsglobalvoicesorg20120304china-challenging-the-50-cent-party

Lam Oiwan 2013 ldquoChina Beefs Up lsquo50 Centrsquo Army ofPaid Internet Propagandistsrdquo Global Voices httpsadvoxglobalvoicesorg20131017china-beefs-up-50-cent-army-of-paid-internet-propagandists

Letham Benjamin et al 2015 ldquoInterpretable Classifiers Using Rulesand Bayesian Analysis Building a Better Stroke PredictionModelrdquo Annals of Applied Statistics 9(3) 1350ndash71

Lohmann Susanne 1994 ldquoThe Dynamics of Informational Cas-cades The Monday Demonstrations in Leipzig East Germany1989ndash1991rdquo World Politics 47(1) 42ndash101

Lorentzen Peter L 2013 ldquoRegularizing Rioting Permitting PublicProtest in an Authoritarian Regimerdquo Quarterly Journal of PoliticalScience 8(2) 127ndash58

Lynch Daniel C 1999 After the Propaganda State Media Politicsand ldquoThought Workrdquo in Reformed China Stanford CA StanfordUniversity Press

MacKinnon Rebecca 2012 Consent of the Networked The World-wide Struggle For Internet Freedom New York NY Basic Books

Mercier Hugo and Dan Sperber 2011 ldquoWhy Do Humans ReasonArguments for an Argumentative Theoryrdquo Behavioral and BrainSciences 34(02) 57ndash74

Miller Blake Andrew Phillip 2016 ldquoAutomatic Detection of Com-ment Propaganda in Chinese Mediardquo Available at httpspapersssrncomsol3Paperscfmabstract id=2738325

Naher Anatol-Fiete and Ivar Krumpal 2012 ldquoAsking SensitiveQuestions The Impact of Forgiving Wording and Question Con-text on Social Desirability Biasrdquo Quality and Quantity 46(5) 1601ndash16

National Bureau of Statistics of China 2014 China Statistical YearBook Beijing CN China Statistics Press httpjmpChinaSY

Ng Jason 2011 ldquoWu Maordquo Blocked on Weibo blog httpsblockedonweibotumblrcom

Ng Jason Q 2015 ldquoPolitics Rumors and Ambiguity Track-ing Censorship on WeChatrsquos Public Accounts PlatformrdquoMunk School of Global Affairs httpscitizenlaborg201507tracking-censorship-on-wechat-public-accounts-platform

Pan Jennifer 2015 ldquoBuying Inertia Preempting Social Disorderwith Selective Welfare Provision in Urban Chinardquo PhD disserta-tion Harvard University Cambridge MA

Persson Torsten and Guido Tabellini 2006 ldquoDemocracy and De-velopment The Devil in the DetailsrdquoAmerican Economic Review96 319ndash24

Pew Research Center 2014 ldquoThe Political Typology Beyond Redvs Bluerdquo Pew Research httpwwwpeople-pressorg20140626the-political-typology-beyond-red-vs-blue

Philipp Joshua 2015 ldquoLeaked Emails Show Chinese Regime Em-ploys 500000 Internet Trollsrdquo Epoch Times httpjmptwomill

Roberts Margaret Earling 2014 Fear Friction and Flooding Meth-ods of Online Information Control Dissertation Harvard Univer-sity

Roberts Margaret E 2015 ldquoExperiencing Censorship Embold-ens Internet Users and Decreases Government Support inChinardquo Unpublished Working Paper httpmargaretrobertsnetwp-contentuploads201507fearpdf

Shepsle Kenneth A 1985 ldquoComment of Why the Regulators Choseto Deregulaterdquo In Regulatory Policy and the Social Sciences edRoger Noll Berkeley CA University of California Press 231ndash39

Shirk Susan L 2011 Changing Media Changing China New YorkNY Oxford University Press

Sonnad Nikhil 2014 ldquoHacked Emails Reveal Chinarsquos Elaborateand Absurd Internet Propaganda Machinerdquo Quartz httpjmpSonnad

Stockmann Daniela 2013 Media Commercialization and Authori-tarian Rule in China Cambridge UK Cambridge University Press

Stockmann Daniela and Mary E Gallagher 2011 ldquoRemote Con-trol How the Media Sustain Authoritarian Rule in Chinardquo Com-parative Political Studies 44(4) 436ndash67

Strafella Giorgio and Daria Berg 2015 ldquolsquoTwitter BodhisattvarsquoAi Weiweirsquos Media Politicsrdquo Asian Studies Review 39(1) 138ndash57

Tang Min Laia Jorba and Michael J Jensen 2012 ldquoDigital Mediaand Political Attitudes in Chinardquo In Digital Media and PoliticalEngagement Worldwide A Comparative Study ed Eva AnduizaCambridge UK Cambridge University Press 221ndash39

Tilly Charles 1978 From Mobilization to Revolution New YorkNY McGraw-Hill

Tong Yanqi and Shaohua Lei 2013 ldquoWar of Position and Microblog-ging in Chinardquo Journal of Contemporary China 22(80) 292ndash311

Tourangeau Roger Frederick Conrad and Mick Couper 2013The Science of Web Surveys Oxford England Oxford UniversityPress

Truex Rory 2016 ldquoFocal Points Dissident Calendars and Preemp-tive Repressionrdquo SSRN httpspapersssrncomsol3paperscfmabstract id=2802859

Wade Samuel 2016 ldquoWho Does Global Times Speak Forrdquo ChinaDigital Times httpjmpGT-CCP

Wintrobe Ronald 1998 The Political Economy of Dictatorship NewYork NY Cambridge University Press

Xi Jinping 2014 ldquoOverall Plan Involving All Parties for Innovationand Development to Strive to Build Our Country into an InternetPowerrdquo httpjmpXiJingingSph

Yang Guobin 2009 The Power of the Internet in China CitizenActivism Online New York NY Columbia University Press

Yang Xiaofeng Qian Yang and Christo Wilson 2015 ldquoPenny forYour Thoughts Searching for the 50 Cent Party on Sina WeibordquoIn Proceedings of the 9th International AAAI Conference on Weband Social Media (AAAI rsquo15)

Zhang Qingpeng Dominic DiFranzo and James A Hendler 2014ldquoSocial Networking on the World Wide Webrdquo In Encyclopedia ofSocial Network Analysis and Mining Berlin Germany Springer1879ndash92

Zhao Juanjuan Weili Wu Xiaolong Zhang Yan Qiang Tao Liu andLidong Wu 2014 ldquoA Short-Term Trend Prediction Model of Topicover Sina Weibo Datasetrdquo Journal of Combinatorial Optimization28(3) 613ndash25

501

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44

INTRODUCTION


LEAKED INTERNET PROPAGANDA OFFICE COMMUNICATIONS




WHAT MIGHT BE WRONG


CONCLUDING REMARKS


REFERENCES


The nearly unanimous view of journalists (and hu-man rights activists) is that 50c party posts stronglyargue with and debate against those who criticize thegovernment its leaders and their policies This is alsothe view of many scholarly publications discussing thisactivity (and was our view as well prior to the re-search reported here) We systematically summarizethese views including a quantitative analysis of socialmedia posts openly accused of being written by 50cparty members Unfortunately until now no methodhas existed for detecting 50c party members whichposts they write their content the size of the opera-tion or why they write them and so the best anyonecould do was base claims on intuition logic occasionalanecdotes rumors and leaked government directives

In this article we offer the first systematic empiricalevidence for the content of 50c party posts their au-thors and the governmentrsquos strategic objectives3 Webegin by analyzing an archive of emails leaked from theInternet Propaganda Office of Zhanggong a district ofGanzhou City in Jiangxi Province These emails giveexplicit details of the work of numerous 50c accountsin this district Although in the public domain and re-ported in the press (eg Henochowicz 2014 Sonnad2014) the structure of the archive is complicated toolarge to understand by traditional qualitative meth-ods and in formats (and attachments) far too diverseto make standard methods of automation feasible Assuch it has never before been systematically analyzedand little of it has been explored We have developedan approach to analyze this dataset and have extractedmore than 43000 known 50c party posts and their au-thors from it

We first characterize the patterns in these data viatheir network and time series structures Then we sys-tematically analyze the content of the 50c posts in ourleaked archive and extrapolate to the rest of Chinain stages We then use this methodology to study thecontent of the posts and finally infer the goals behindthis massive government program and how it may re-veal broader government strategies We validate our50c party member predictions with a novel sample sur-vey of predicted 50c party members as well as severalunusual gold standard evaluations that we develop tovalidate our validation We estimate and reveal the sizeof what turns out to be a massive government operationthat writes approximately 448 million 50c posts a yearWe also discuss our assumptions interpretations andwhat might go wrong with our evidence and inferences

At every stage our results indicate that prevailingviews about the 50c party are largely incorrect We

the government the ldquolittle red flowersrdquo () unpaid red guardswho also attack opponents of the regime online the ldquoAmerican CentPartyrdquo () who express Western democratic values and criti-cize the Chinese communist regime online and the ldquoInternet waterarmyrdquo () which refers to for-hire astroturfers working forand advancing the interests of companies and other actors willing topay their fees None are known to be organized as an entire collectiveOf course political parties do not exist in China and so despite thename the 50c party is not a political party For an excellent overviewof the purpose and tactics of the volunteer 50c party see Han (2015a)3 Because China is a single-party regime we use the terms govern-ment and regime interchangeably to refer to those in power

show that almost none of the Chinese governmentrsquos50c party posts engage in debate or argument of anykind They do not step up to defend the governmentits leaders and their policies from criticism no matterhow vitriolic indeed they seem to avoid controversialissues entirely Instead most 50c posts are about cheer-leading and positive discussions of valence issues Wealso detect a high level of coordination in the timing andcontent in these posts A theory consistent with thesepatterns is that the strategic objective of the regime is todistract and redirect public attention from discussionsor events with collective action potential

The theoretical implications of our findings are pre-sented later in the article We give a unified parsimo-nious summary of Chinese government internal infor-mation control efforts and show how these findingsmay cause scholars to rethink the notion of ldquocommonknowledgerdquo in theories of authoritarian politics moregenerally Finally we conclude and then give a sum-mary of what we might have missed and how scholarscan follow up on this work4


We summarize here views about the 50c party of (1)journalists (2) academics and (3) social media par-ticipants accusing others of being 50c party membersThe dominant view of most is that 50c party membersengage in ldquohand-to-handrdquo verbal combat making spe-cific directed arguments that support the governmentits leaders and their policies and opposing argumentsto the contrary they do this by engaging in debate withand criticism of Chinarsquos enemies including those whooppose it inside the country and from abroad For (1)and (2) we offer brief literature reviews for (3) we findand analyze posts accused by others of being 50c Inthat section we also introduce and validate a schemeto categorize 50c posts on the web we use it in thissection to understand the posts accused of being writ-ten by 50c members and then for many other purposesthroughout this article

Although the difficulties of collecting data on aninherently secret operation means that most priorliterature includes ldquono successful attempts to quan-tify regime-sponsored commentary in Chinardquo (Miller2016) the work cited in this section involves consider-able effort and creativity and even a few clever effortsto guess what might be 50c posts before turning totry to explain or predict the guesses (Miller 2016 Han2015b) For example Han (2015b) uses informationfrom leaked censorship directives and local media re-ports of the training of online commentators in an on-line ldquoguerrilla ethnographyrdquo Still the lack of ground

4 Appendix B briefly describes the (unintended) events and implica-tions of the Chinese governmentrsquos unexpected decision to respondto an early version of this article through an editorial in a state-owned newspaper In so doing they appear to give a rare admissionof the existence of the 50c party and confirmation of our empiricalresults In doing this they also provide important information aboutgovernment strategy that we discuss (Our Online SupplementaryAppendix provides additional details and a translation of the originaleditorial)

485

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44











486

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44













487

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44



010

0020

0030

0040

00


Cou

nt o

f Pos

ts



2 ChinaDream(May)



Congress (Nov)

5 Two meetings(Feb)



(JulminusAug)

8 MartyrsDay

(Oct)







488

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44
















489

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44















490

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44









491

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44













492

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44














493

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44
















494

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44









WHAT MIGHT BE WRONG





495

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44














496

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44








CONCLUDING REMARKS






497

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44



















498

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44






















499

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44








REFERENCES































500

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44















































501

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44

INTRODUCTION






WHAT MIGHT BE WRONG


CONCLUDING REMARKS


REFERENCES











486

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44













487

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44



010

0020

0030

0040

00


Cou

nt o

f Pos

ts



2 ChinaDream(May)



Congress (Nov)

5 Two meetings(Feb)



(JulminusAug)

8 MartyrsDay

(Oct)







488

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44
















489

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44















490

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44









491

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44













492

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44














493

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44
















494

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44









WHAT MIGHT BE WRONG





495

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44














496

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44








CONCLUDING REMARKS






497

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44



















498

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44






















499

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44








REFERENCES































500

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44















































501

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44

INTRODUCTION






WHAT MIGHT BE WRONG


CONCLUDING REMARKS


REFERENCES













487

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44



010

0020

0030

0040

00


Cou

nt o

f Pos

ts



2 ChinaDream(May)



Congress (Nov)

5 Two meetings(Feb)



(JulminusAug)

8 MartyrsDay

(Oct)







488

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44
















489

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44















490

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44









491

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44













492

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44














493

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44
















494

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44









WHAT MIGHT BE WRONG





495

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44














496

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44








CONCLUDING REMARKS






497

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44



















498

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44






















499

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44








REFERENCES































500

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44















































501

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44

INTRODUCTION






WHAT MIGHT BE WRONG


CONCLUDING REMARKS


REFERENCES



010

0020

0030

0040

00


Cou

nt o

f Pos

ts



2 ChinaDream(May)



Congress (Nov)

5 Two meetings(Feb)



(JulminusAug)

8 MartyrsDay

(Oct)







488

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44
















489

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44















490

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44









491

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44













492

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44














493

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44
















494

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44









WHAT MIGHT BE WRONG





495

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44














496

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44








CONCLUDING REMARKS






497

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44



















498

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44






















499

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44








REFERENCES































500

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44















































501

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44

INTRODUCTION






WHAT MIGHT BE WRONG


CONCLUDING REMARKS


REFERENCES
















489

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44















490

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44









491

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44













492

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44














493

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44
















494

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44









WHAT MIGHT BE WRONG





495

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44














496

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44








CONCLUDING REMARKS






497

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44



















498

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44






















499

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44








REFERENCES































500

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44















































501

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44

INTRODUCTION






WHAT MIGHT BE WRONG


CONCLUDING REMARKS


REFERENCES















490

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44









491

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44













492

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44














493

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44
















494

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44









WHAT MIGHT BE WRONG





495

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44














496

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44








CONCLUDING REMARKS






497

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44



















498

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44






















499

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44








REFERENCES































500

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44















































501

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44

INTRODUCTION






WHAT MIGHT BE WRONG


CONCLUDING REMARKS


REFERENCES









491

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44













492

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44














493

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44
















494

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44









WHAT MIGHT BE WRONG





495

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44














496

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44








CONCLUDING REMARKS






497

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44



















498

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44






















499

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44








REFERENCES































500

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44















































501

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44

INTRODUCTION






WHAT MIGHT BE WRONG


CONCLUDING REMARKS


REFERENCES













492

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44














493

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44
















494

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44









WHAT MIGHT BE WRONG





495

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44














496

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44








CONCLUDING REMARKS






497

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44



















498

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44






















499

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44








REFERENCES































500

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44















































501

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44

INTRODUCTION






WHAT MIGHT BE WRONG


CONCLUDING REMARKS


REFERENCES














493

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44
















494

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44









WHAT MIGHT BE WRONG





495

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44














496

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44








CONCLUDING REMARKS






497

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44



















498

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44






















499

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44








REFERENCES































500

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44















































501

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44

INTRODUCTION






WHAT MIGHT BE WRONG


CONCLUDING REMARKS


REFERENCES
















494

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44









WHAT MIGHT BE WRONG





495

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44














496

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44








CONCLUDING REMARKS






497

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44



















498

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44






















499

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44








REFERENCES































500

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44















































501

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44

INTRODUCTION






WHAT MIGHT BE WRONG


CONCLUDING REMARKS


REFERENCES









WHAT MIGHT BE WRONG





495

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44














496

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44








CONCLUDING REMARKS






497

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44



















498

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44






















499

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44








REFERENCES































500

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44















































501

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44

INTRODUCTION






WHAT MIGHT BE WRONG


CONCLUDING REMARKS


REFERENCES














496

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44








CONCLUDING REMARKS






497

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44



















498

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44






















499

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44








REFERENCES































500

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44















































501

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44

INTRODUCTION






WHAT MIGHT BE WRONG


CONCLUDING REMARKS


REFERENCES








CONCLUDING REMARKS






497

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44



















498

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44






















499

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44








REFERENCES































500

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44















































501

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44

INTRODUCTION






WHAT MIGHT BE WRONG


CONCLUDING REMARKS


REFERENCES



















498

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44






















499

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44








REFERENCES































500

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44















































501

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44

INTRODUCTION






WHAT MIGHT BE WRONG


CONCLUDING REMARKS


REFERENCES






















499

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44








REFERENCES































500

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44















































501

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44

INTRODUCTION






WHAT MIGHT BE WRONG


CONCLUDING REMARKS


REFERENCES








REFERENCES































500

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44















































501

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44

INTRODUCTION






WHAT MIGHT BE WRONG


CONCLUDING REMARKS


REFERENCES















































501

Dow

nloa

ded

from

htt

ps

ww

wc

ambr

idge

org

cor

e H

arva

rd U

nive

rsity

on

28 A

ug 2

017

at 1

416

56

sub

ject

to th

e Ca

mbr

idge

Cor

e te

rms

of u

se a

vaila

ble

at h

ttps

w

ww

cam

brid

geo

rgc

ore

term

s h

ttps

do

iorg

10

1017

S00

0305

5417

0001

44

INTRODUCTION






WHAT MIGHT BE WRONG


CONCLUDING REMARKS


REFERENCES

how the chinese government fabricates social media posts for … · 2017-08-28 · how the chinese...

Documents