Journal of Development Economics 153 (2021) 102732

Contents lists available at ScienceDirect

Journal of Development Economics

journal homepage: www.elsevier.com/locate/devec

Regular Article

Mass persuasion and the ideological origins of the Chinese

Cultural Revolution☆
Susan Ou a, Heyu Xiong b, *
a
Chicago, USA
b
Case Western Reserve University, USA

A R T I C L E I N F O A B S T R A C T

Keywords: We study the role of media in the transmission of ideology during the Cultural Revolution. We find that counties
State media with a stronger radio signal experienced higher revolutionary intensity as shown by the number of killings. We
Propaganda also find that the effects of radio were concentrated in areas where Mandarin, the language of state-sponsored
Chinese Cultural Revolution
broadcasts, was better understood. This suggests that the extent of state persuasion was constrained by the
Economics of conflict
linguistic heterogeneity of the population. Through investigation of later-in-life outcomes such as participation in
the Communist Party and attitudes on wealth and inequality, we provide evidence of persistence in beliefs among
the group most differentially exposed to media — native Mandarin speakers of an impressionable age at the start
of the Cultural Revolution.

1. Introduction
Autocratic regimes frequently use repression and violence as a means
to consolidate power and maintain political order. A striking develop­
ment in the 20th century, however, is the capacity of these states to build
civilian support for their actions not through overt coercion, but through
persuasion. This is particularly evident in the context of Mao-era China,
where the Chinese Communist Party repeatedly orchestrated mass
“revolutionary” campaigns as a method of governance. These campaigns
include the Land Reform (1947–1950), the Anti-Rightist Movement
(1957–1959), the Great Leap Forward (1958–1960), and perhaps most
notably, the Chinese Cultural Revolution (1966–1976). The success of
these movements relied, in part, on the mobilization of civilians who
were otherwise only tangentially connected to the political process. The
understanding of these events is arguably incomplete without an ac­
count of what caused the overwhelming civilian response and
participation.
In this paper, we focus on one of the largest and most significant
political campaigns launched by the Chinese Communist Party, the
Cultural Revolution, and study the role of mass persuasion in the orga­
nization of mass killing. We draw from a recent literature that discusses
how states use communication and media technology to facilitate
civilian involvement during periods of intrastate conflict (Adena et al.,
2015; Voigtländer and Voth, 2014; Yanagizawa-Drott, 2014). Our paper
contributes to this literature by examining the effect of media in a new
context, where, unlike other settings, the conflict was not ethnic in
origin. In doing so, we also shed light on the unique constraints to state
persuasion in a linguistically diverse country, as well as the long-run
effects of mass persuasion on individuals.
The Chinese Cultural Revolution was a large-scale political move­
ment characterized by collective violence and political persecution.
Violence against political opponents was state sanctioned, but typically
carried out by ordinary citizens. Estimates of the number of fatalities
ranges from 250,000 to 1.6 million, while the total number of victims,

☆
We are grateful to Georgy Egorov, Joel Mokyr, Matt Notowidigdo, and Nancy Qian for generous advice and feedback on this work. We also thank Lori Beaman,
Matthias Doepke, Joe Ferrie, Seema Jayachandran, Ruixue Jia, Cynthia Kinnan, Lee Lockwood, Mara Squicciarini, and participants of the Northwestern Applied
Microeconomics Seminar, History Seminar, NBER Conference on the Chinese Economy, and DEVPEC Conference at UC Berkeley for helpful advice and comments.
Heyu Xiong is grateful for financial support from the Balzan Foundation and Northwestern’s Center for Economic History. The authors are grateful to Hiroyuki
Kasahara and Bingjing Li for sharing Communist Party membership data, and to Andrew Walder for sharing Cultural Revolution violence data.
* Corresponding author.
E-mail addresses: susanou546@gmail.com (S. Ou), hxx226@case.edu (H. Xiong).

https://doi.org/10.1016/j.jdeveco.2021.102732
Received 27 January 2020; Received in revised form 13 July 2021; Accepted 8 August 2021
Available online 13 August 2021
0304-3878/© 2021 Elsevier B.V. All rights reserved.
S. Ou and H. Xiong
including those persecuted, ranges from 22 to 30 million (Walder,
2014).1 Despite a large literature documenting the consequences of this
traumatic episode on Chinese history and the cohorts which experienced
it, there is little consensus as to what caused ordinary citizens to
perpetrate violence.
First, we show that state-sponsored media led to more killings during
the Cultural Revolution. In the prelude to the Cultural Revolution, the
Communist Party developed a sophisticated wired radio infrastructure
from which politicized media was regularly broadcast. To identify the
contemporaneous effect of radio broadcast on revolutionary intensity,
we construct a proxy for county-level radio exposure based on the
strength of signal from each provincial relay stations to the county
receiving station.
The main empirical challenge to estimating the effect of radio pro­
paganda is the possibility that the variation in radio signal could reflect
unobservable characteristics which may also influence revolutionary
intensity. To address this, we follow a strategy pioneered by Olken
(2009) and used extensively in the media literature (see e.g. Yanagiza­
wa-Drott (2014), Enikolopov et al. (2011), and DellaVigna et al. (2014)).
Specifically, we regress outcomes on radio signal, controlling for the
hypothetical radio signal in the absence of geomorphological obstacles.
Doing so allows us to isolate the exogenous variation in radio exposure
attributable only to topographic features in the line-of-sight along the
signal transmission path. Identification comes from the assumption that,
conditional on the hypothetical free-space signal, the residual variation
in radio signal is exogenous to other determinants of conflict. While this
assumption is fundamentally un-testable, we assess its plausibility by
examining the relationship of radio signal with observable covariates.
Reassuringly, we show that radio coverage is uncorrelated with an
extensive set of both contemporaneous and historical county charac­
teristics, including population, ethnolinguistic fragmentation, urbani­
zation, gender ratio and the distance to provincial capitals.
One concern is the possibility that local radio development is related
to the penetration of Chinese Communist Party or the extent of their
influence in the region. To address this, in our main specification, we
control for the county-level Chinese Communist Party membership prior
to the onset of the Cultural Revolution in 1956. This is the most direct
measure of contemporaneous state capacity available, and the robust­
ness of our results to this control suggests that the effect of radio was
unlikely to be confounded by state capacity.
We find that counties with better radio reception experienced a
higher number of fatalities over the course of the Cultural Revolution. A
one standard deviation increase in radio signal corresponds to a 0.014
percentage point change in the population killed. We interpret this as
the causal effect of radio broadcast on the extent of conflict.
Next, having established the effect of radio broadcasts on violence,
we then demonstrate that the extent of state persuasion was affected by
the linguistic compatibility of the population. China is a linguistically
diverse country, composed of hundreds of mutually unintelligible di­
alects united by a common written script. In order to combat localism
and to facilitate national integration, the Chinese Communist Party
severely restricted the use of minority dialects over the airwaves in the
years prior to the Cultural Revolution. It was mandated that state news
and central broadcasts be conducted in the national language: Standard
Mandarin. We exploit this institutional detail to create a measure of
comprehension based on pre-existing heterogeneity in the local pop­
ulation’s ability to understand spoken Mandarin.
Using an interaction specification, we find that the effects of radio
1 Rwandan genocide increased militia violence. Qin et al. (2017) describe
The death toll exceeds some of the modern era’s worst incidents of
politically-induced mortality, such as the Soviet “Great Terror” of 1937–38, the
Rwandan genocide of 1994, and the Indonesian coup and massacres of sus­
pected communists in 1965–66. The death figure also likely understates the true
extent of the violence due to the presumably larger number of those who were
imprisoned or otherwise persecuted.
2
Journal of Development Economics 153 (2021) 102732
were concentrated primarily in locations where the local dialect aligned
with the language of radio broadcasts. Specifically, we employ two
measures of linguistic compatibility: 1) a binary indicator for whether
the local dialect belonged to the Mandarin language family, and 2) a
continuous measure of the degree of mutual intelligibility between the
local dialect and Standard Mandarin. Across both of these measures, we
find that the effect of radio is increasing in the Mandarin comprehension
of the local population. In other words, radio broadcast did induce more
conflict, but only in areas where Mandarin was relatively well under­
stood. In Mandarin speaking counties, we find that a standard deviation
shift in radio signal leads to a 0.022 percentage point change in the
population killed compared to non-Mandarin speaking counties. In non-
Mandarin counties, the effect size is insignificant and small in magni­
tude. Thereby, we show that the extent of exposure and success of
indoctrination hinged not only on the broadcast infrastructure but also
on the linguistic compatibility of the listeners.
Finally, we investigate the long-term consequences of state propa­
ganda on individual behavior using retrospective micro-data from the
2010 China Family Panel Studies (CFPS) survey. We build on our pre­
vious result relating media exposure to linguistic compatibility and
exploit differences in individuals’ native language as a source of iden­
tifying variation.2 We further utilize variation across birth cohorts
during the Cultural Revolution as another dimension of exposure.
We implement a difference-in-differences strategy using these two
sources of variation. We regard Mandarin speakers who lived through
the Cultural Revolution at an impressionable age, when they were more
likely to be persuaded by the political rhetoric, as the treatment group.
We compare the outcomes of this group with that of non-Mandarin
speakers from the same cohort as well as Mandarin speakers from
other cohorts. In other words, our empirical specification captures pat­
terns unique to Mandarin speakers among the birth cohorts most readily
exposed to radio propaganda. The combination of dialect and cohort
variation allows us to flexibly control for time-period and life-cycle ef­
fects as well as unobserved but cohort invariant heterogeneity arising
from being a native Mandarin speaker.
We show that native Mandarin speakers who experienced the Cul­
tural Revolution during the critical years of adolescence and early
adulthood exhibit behavior and held views consistent with Communist
ideology later in life. Namely, we find that these individuals were more
likely to join the Communist Party later in life. They were also less
tolerant of inequality and generally held more negative attitudes con­
cerning wealth. These results coincide with both the psychological
literature regarding the susceptibility of adolescents to media indoctri­
nation and the nature of state propaganda during the Cultural Revolu­
tion. The fact that we find these effects only among Mandarin speakers
within the critical age cohort is consistent with our county-level results:
that language was a barrier for the propagation of propaganda. The
evidence suggests that media formed a persistent and long-lasting
impression on individuals, which also facilitated the recruitment and
supply of future party members.
This paper relates to several distinct strands of literature. It com­
plements recent work exploring the political effects of media. Della­
Vigna and Kaplan (2007), Gerber et al. (2009), and Chiang and Knight
(2011) investigate media influence on voting behavior in developed
democracies. Enikolopov et al. (2011) and DellaVigna et al. (2014) show
effects of media on voting behavior in transitional democracies, namely
Russia and Croatia. Adena et al. (2015) attribute the rise of Nazi support
partially to the influence of radio propaganda. Less is known regarding
the impact of media in non-democracies. Notably, Yanagizawa-Drott
(2014) find that radio broadcasts encouraging violence during the
2
The CFPS asks respondents about the main language they speak at home,
which we assume is their native language. Even if individuals learned Mandarin
later in life, their spoken language at home would not have likely changed.
S. Ou and H. Xiong
governmental control of media in contemporary China, and Qin et al.
(2018) study whether market competition constrained the extent of
media bias in the same setting.
We contribute to this literature by exploring the effect of propaganda
in a novel context — the Cultural Revolution, which is substantially
distinct from the contexts found in existing studies. Prior papers have
shown the capacity of media to exploit pre-existing ethnic cleavages and
instigate violence exclusively along that dimension. Our paper illus­
trates the ability of the state to carry out mass violence through media
that is unrelated to ethnic dispositions.
Our unique institutional setting also allow us to explore heteroge­
neity based on the variation in spoken vernacular. We show that lin­
guistic compatibility was crucial for the success of indoctrination.3 The
results suggest that the effectiveness of persuasion can be highly het­
erogeneous and depend crucially on the identity of the audience. This
paper increases understanding of the characteristics that might make
individuals more able to be persuaded and more receptive to media
effects.
In addition to the contemporaneous effects, we explore long-run
consequences of exposure to state-sponsored media. The previous
literature has found that living through Communist regimes has a lasting
effect on individual beliefs (Alesina and Fuchs-Schündeln, 2007).
Similarly, Voigtländer and Voth (2015) show Nazi indoctrination had
persistent effect on fostering anti-Semitism in Germany. We attempt to
provide one explanation for why this occurs. Empirical evidence on the
persistence of state media is elusive. In general, studies have found that
political communication rarely has long term effects. The effect of po­
litical advertisement in US political campaigns has been found to be
short-lived (Gerber et al., 2011). Because of the political continuity in
the Chinese context, our setting is particularly well suited to answer this
question. The political environment is relatively stable over the time
period we study. China has not experienced a regime change since the
founding of the People’s Republic of China, and the party responsible for
the Cultural Revolution-era propaganda remains in control. The
post-Mao shift in party policy and leadership did not result in the
outright repudiation of Mao and his policies. With the exception of the
prosecution of the Gang of Four, public criticism of Mao-era policies
were understated. It stands to reason that 1950s propaganda would still
be consequential for the current regime. Our results suggest that pro­
paganda is effective at cultivating a more permanent support of the
government.
This paper also contributes to the literature on how linguistic di­
versity shapes economic and political outcomes. Linguistic fractional­
ization is a barrier to state capacity, hindering the government’s ability
to implement policy. Outcomes due to this limitation on state capacity
are found to have an unfavorable impact on the country. Numerous
studies have attributed political instability, poor institutions, and low
economic growth to ethnolinguistic fragmentation. For instance, East­
erly and Levine (1997) find that GDP growth is inversely related to
fractionalization across a large sample of countries. La Porta et al.
(1999) show fractionalization is important in determining the quality of
government. Alesina and La Ferrara (2000) document that participation
in social activities is lower in more ethnically or racially fragmented
localities in the United States. More recently, Michalopoulos (2012) and
Bazzi et al. (2017) explore the causes and consequences of ethno­
linguistic diversity in the setting of developing countries.
We provide a clear mechanism to the limits of centralization when
there is linguistic fragmentation using intra-country evidence. During
the Cultural Revolution in China, linguistic differences precluded the
state from carrying out its goals by constraining the audience of state
sponsored media, thereby limiting the scope of persuasion. This pro­
vides new evidence that promoting linguistic homogeneity through
3
Gupta et al. (2020) makes a similar point regarding language constraints
limiting access to information in the context of India.
3
Journal of Development Economics 153 (2021) 102732
standardization of language augments state capacity, broadly
conceived.
Finally, our paper adds to the growing empirical literature on the
Chinese Cultural Revolution. These studies have typically focused on the
outcomes for areas and the cohorts who experienced it (Bai, 2014; Gong,
Lu and Xie, 2014, 2015; Kinnan et al., 2015; Meng and Gregory, 2007;
Zhou, 2013).4 While a large empirical literature has explored the long
term implications of Cultural Revolution on the birth cohorts that
experienced it, we use variation in individual propaganda exposure to
disentangle the effect of propaganda and ideology from the “bundle” of
living through the Cultural Revolution. There is also relatively little
work studying the local determinants of the Cultural Revolution. To our
knowledge, we are the first to provide rigorous empirical evidence on
the cause of the violence, as well as the motivating factors compelling
individuals to be voluntarily rusticated. By showing that media in­
fluences ideology and individual behavior during the Cultural Revolu­
tion, our study also complements two recent studies about the
determinants of ideology (Cantoni et al., 2017) and foreign media up­
take (Chen and Yang, 2019) in China today.
The rest of the paper proceeds as follows: Section 2 provides a brief
history of radio broadcasting in China and the Cultural Revolution,
Section 3 describes our data, and Section 4 explains the contempora­
neous effects of media on violence during the Cultural Revolution. An
analysis of individual level results and the persistence of ideology is
presented in Section 5. Concluding remarks are offered in Section 6.
2. Background
It is beyond the scope of our paper to provide an extensive back­
ground of the events of the Cultural Revolution. Instead, in this section
we will focus on the particular details that are relevant for our analysis
of the relationship between radio propaganda and individual behavior.
We document that the Communist government undertook a campaign to
mobilize the public into action during this time period. The media, and
especially radio, were especially salient ways to communicate with the
masses and facilitate the transmission of state ideology.
The Cultural Revolution was a large-scale political campaign
launched by Mao Zedong in 1966 with a stated aim of preserving Chi­
nese Communism by “cleansing the class ranks” of capitalist elements
(Schoenhals and MacFarquhar, 2006; Su, 2011). The violence of the
Cultural Revolution was pervasive and widespread, especially during
the first two years, from 1966 to 1968. Even though much of the writing
concerning violence during the Cultural Revolution focuses on urban
violence, work has documented the extent of violence in the rural areas
(Walder and Su, 2003). Individuals deemed incompatible with the so­
cialist system were persecuted, including intellectuals, senior party of­
ficials, rich peasants, teachers and elites. These “class enemies” were
subject to public denunciations, forced self criticisms, and beatings if not
outright death. The violence and political purges during the Cultural
Revolution were typically perpetrated by ordinary individuals within a
4
Meng and Gregory (2007) find that those whose education was disrupted
during this time period faced a decrease in lifetime earnings. Within this
literature, special attention is paid to the impact of the Send Down Movement
specifically. Zhou (2013) explores the long term effects of the Send Down
Movement on individuals, finding that those who were sent down in fact have
better economic outcomes. Gong, Lu and Xie (2015) explores the persistent
effect of China’s the Send Down Movement on beliefs, finding that those in­
dividuals who were sent down to work in the countryside are less likely to
believe that external circumstances such as luck, control their lives. Gong, Lu
and Xie (2014) also find that these sent-down youth are more likely to expe­
rience mental health problems and chronic diseases. Bai (2014) investigates the
economic legacies of violence during the Cultural Revolution, finding that more
revolutionary regions were slower to industrialize and have a lower GDP.
Kinnan, Wang and Wang (2015) finds the effects of lasting inter-province links
created by migration due to the Send Down Movement.
S. Ou and H. Xiong
radicalized community rather than agents of the central government.
Oftentimes, the perpetrators and victims knew each other.
Red Guards, a revolutionary youth organization composed of ordi­
nary civilians, became the paramilitary force of the movement. In the
initial phase of the Cultural Revolution, traditional state apparatuses
such as the police were instructed not to intervene in Red Guard activ­
ities and individuals who attempted to do so would risk being labeled
counterrevolutionaries (Schoenhals and MacFarquhar, 2006). Red
Guards often engaged in power seizures. According to varying sources,
between 1.1 and 1.6 million were killed during the Cultural Revolution,
and many more were victims of persecution (Walder, 2014).
The nature of the violence during this time period was primitive.
Instead of guns and armed weaponry, victims were often beaten to death
with blunt objects, or forced to jump off cliffs. The perpetrators of
violence were not deemed as criminals, but rather, were accepted as
someone who was acting on behalf of the community. These commu­
nities were “willing” agents of the central government (Su, 2011).
The Cultural Revolution required mass involvement and compliance
at many levels of the public. Because the movement was initiated by
Mao partly to re-assert control over the party, it did not have consensus
support within the government itself, particularly provincially. To
implement the movement, Mao bypassed traditional party structures
and appealed directly to the masses (Fukuyama, 2014). Radio became
the means through which the overwhelming popular response was
realized.
Throughout the Cultural Revolution, radio allowed for the direct
communication between the central committee and the local commu­
nities. Radio broadcasts served the dual purpose of communicating
Mao’s directions and goals as well as inciting the masses to action and to
carry out these goals. State rhetoric heightened the revolutionary fervor
among ordinary civilians who were otherwise tangentially connected to
the political process. In the next few sections, we provide an overview of
the buildup of radio infrastructure in the years leading up to the Cultural
Revolution, the content of the broadcasts, and constraints on its
effectiveness.
2.1. Broadcasting infrastructure in communist China
The official use of radio by the Communist Party of China (CPC) dates
from September 1945, when it established a radio station within the CPC
controlled territory in Ya’an. Following political consolidation in 1949,
the regime nationalized existing private stations and developed an
extensive network of mass communication that was centrally operated.
Detailed instructions regarding its administration and establishment
were given in April 1950, when “Decisions Regarding the Establishment
of Radio-Receiving Networks” was announced (Pye, 2015).
Similar to the organization of the broadcasting network in the Soviet
Union at the time, the Chinese government organized its broadcasting
network at three distinct operational levels: central, regional, and local,
each of which corresponded to their respective geographic units and
political authority. The central tier referred to the Central People’s
Broadcasting Station, or Radio Beijing. It created programming, in
particular national and international news, and dictated policy. Radio
transmission from the central station was relayed wirelessly, first to
municipal or provincial radio stations, before being wirelessly re-
directed to radio receiving stations located at each county seat (Liu,
1964). The local radio stations broadcasted content created by the
Central People’s Broadcasting Station, and could only add local news
content (Liu, 1971).
From the county radio station, wires that carry the broadcasts were
extended to the rural villages in its jurisdiction and connected to
strategically-situated loudspeakers. Typical locations included floors of
manufacturing plants, poles in market places, roofs of government
buildings, communes, and dormitories. These public places were chosen
for their visibility to facilitate collective listening. The overall engi­
neering schematic is illustrated in Fig. 1. Strictly speaking, this was not a
4
Journal of Development Economics 153 (2021) 102732
Fig. 1. Radio Diffusion Exchange. Note: This is a reproduction of a figure in Liu
(1964). It shows the overall structure of the radio diffusion exchanges. The
dotted lines represent signals sent over the airwaves and solid lines represent
wired connections.
broadcasting system, but a system of point-to-point radio communica­
tion, with dissemination of selected programming at points of reception
through means of wired loudspeakers. This system and its predecessors
were known as “radio diffusion exchanges” (Houn, 1957).
The construction of these exchanges circumvented the lack of ordi­
nary radio equipment. By 1956, there were reportedly 1,500,000 radio
receiver sets capable of receiving programs from medium wave stations
nationally, including those that were controlled and operated by the
government (Jan, 1967). Because personal receiver sets were costly to
manufacture and private ownership was scarce, collective listening via
public loudspeakers constituted the bulk of radio reception through the
1950s and 1960s. Wired loudspeakers were more economical to build en
masse, and also allowed the Communist government to completely
regulate listening habits. Per governmental figures, the total cost for
building a new radio diffusion exchange, together with 150 wired
speakers, was about 7000 yuan and monthly expenses for operating such
exchanges did not exceed 90 yuan. In comparison, the cost of 150 regular
radios was more than 20,000 yuan and their monthly maintenance was
estimated at 1500 to 2000 yuan (Jan, 1967).
In the decade immediately prior to the Cultural Revolution, there
was an extensive buildup in the stock of this broadcasting infrastructure.
In December 1955, the Third National Radio Broadcasting Conference
announced a schedule to build more than 900 wired radio broadcasting
stations in 1956 with 45,000 to 50,000 loudspeakers attached, 80 % of
which would be installed in villages. This conference projected that by
the end of 1957 there would be more than 1,800 wired radio broad­
casting stations with more than 1,360,000 loudspeakers in villages. The
collectivization campaigns and introduction of communes during the
Great Leap Forward facilitated the extension of broadcasting into rural
areas. This initiative had dramatic local consequences. Loudspeakers
were installed in peasants’ homes and commune offices. According to
government sources at the time, by 1963, 95 % of all counties had access
to loudspeaker facilities, although this number may have been inflated
(Jan, 1967).
Sustained radio operation required a consistent source of electrical
power. The expansion in equipment was accompanied by an increase in
the electrical power dedicated to broadcasting. The combined strength
of the stations in 1952 rose to 475.2 kW; in 1954 the figure was more
than nine times that of 1952. In 1957, total kilowatts had increased by
460 % from 1954 (Houn, 1957) . However, by the mid 1960s, wide­
spread electrification still eluded significant portions of China. Large
power plants were concentrated in coastal locations and regions of
Manchuria which were formerly occupied by the Japanese. This was
S. Ou and H. Xiong
despite the call for the installation of large hydro and thermal plants in
Beijing’s first Five Year Plan, initiated in 1956. The actuality of the
electrification campaign consisted of crude generating plants of every
conceivable method: small hydro motors, hand generators, gas motors,
wind motors, etc. (Liu, 1964).
Consequently, radio development was confined to areas where the
power supply was relatively sufficient. This pattern of provincial and
municipal radio stations is supported in the data, as radio station pres­
ence is most prevalent along the main railroad routes at the timeas well
as near tributaries of major waterways. In some parts of the country,
radio infrastructure was bootstrapped from older telephone lines.
Nevertheless, by the eve of the Cultural Revolution in 1964, there
was a robust system of radio communication in China, consisting of 141
provincial and municipal stations (including the central station in Bei­
jing), more than 5,400 wired radio broadcasting stations, and approxi­
mately 6 million loudspeakers across the country, amounting to 1
loudspeaker per every 160 persons (Jan, 1967; Liu, 1964). Radio was an
invaluable tool for state transmissions and projected national authority
directly to its intended recipients.
2.2. Radio content and propaganda
From the outset, broadcasting infrastructure in China was designed
expressly for the purpose of mass persuasion. The political leaders were
keenly aware of mass media as an instrument for state indoctrination
and for the transmission of ideology. Virtually all senior members of the
party were actively involved in media and propaganda activities at some
point in their careers (Volland, 2003). The alacrity with which Chinese
government developed its network of state sponsored media following
its establishment in 1949 was proclaimed by contemporary Western
sources as the “most extensive propaganda effort” in history (Dittmer,
1998). Both scholars writing in the midst of the Cultural Revolution and
researchers writing retrospectively have noted the pervasiveness of
politicized rhetoric and have contemplated its role in facilitating the
events (Howse, 1960; Jan, 1967; Markham and Liu, 1969).
Initially, radio broadcasting was used in adjunct to the press. Due to
persistent illiteracy, the growth in the number and circulation of
newspapers from 1952 to 1959 was modest. Despite a sustained plan to
eradicate illiteracy, the literacy rate remained at only around 30 %
among those in the age group of 14–40, and substantially lower for older
cohorts. Peasants who had become literate often lapsed back into illit­
eracy after the conclusion of compulsory education. Additionally, due to
the varied and oftentimes physically inaccessible terrain, the medium of
radio proved a particularly effective channel of persuasion and state
communication (Liu, 1971).
The content of radio programming was highly integrated to national
policy and focused public attention to immediate goals of the state.
During the Cultural Revolution, political propaganda actively promoted
revolutionary fervor and ideology. The dominant style of broadcasting
focused on mass agitation and the coverage of mass campaigns with the
goal of arousing support and increasing mobilization. This was espe­
cially pronounced from the onset of the Cultural Revolution.
Propaganda constituted a significant portion of the content over the
airwaves. On average, radio broadcasts in 1964 would last 435 min per
day, consisting of program announcements (5 %), educational programs
(16 %), newscasts (29 %), weather (2 %), agricultural programs (7 %),
and entertainment (41 %) (Jan, 1967).
Newscasts consisted of broadcasts from the Central People’s Broad­
casting Station, the official party station, and included programs such as
“Quotations from Chairman Mao,” and leaders’ speeches (Xu, 2014).
One function of the news was to discredit Mao’s enemies and it featured
sensational exposes of those who were purged (Markham and Liu,
1969).
5
Journal of Development Economics 153 (2021) 102732
Entertainment included revolutionary songs and dances, aimed at
political agitation. All traditional or foreign cultural influences,
including music and opera, were purged, leaving only party propaganda.
The media propagandized the “literature of workers and peasants” over
high culture. The Communist Party demanded programs be “national­
istic and populistic” not “intellectual and foreign” (Liu, 1971).
Agricultural programming consisted of half technical advice and half
propaganda, which included speeches from model farmers. In total,
propaganda and indoctrination constituted 85 % of broadcast time (Jan,
1967). These radio broadcasts encouraged listeners to participate in the
revolutionary cause to build a greater China together.
To enforce compulsory listening, a system of collective listening was
adopted in villages and communes. This was implemented in two ways:
broadcasting assemblies and institutional listening. In the former, het­
erogeneous audiences of peasants were gathered together in audito­
riums and listened to designated programs in groups, commonly
monitored by party cadres (Li, 2020). In the latter, loudspeakers
broadcast for a set number of hours each day in public places such as
governmental offices, factories, and schools, where employees, workers,
and students were captive audience members. In the lead up to the
Cultural Revolution, the practices of institutional listening became
greatly promoted. A report from Shanghai dated August 9th, 1966, the
day after Radio Beijing had broadcast the Central Committee’s decision
on the Cultural Revolution, stated:
The broad revolutionary people enthusiastically listened to the
broadcast of the Central Committee’s decision last night. Early this
morning, parade columns appeared in major streets of Shanghai …
The commune members in the suburbs, who were busy in reaping
and planting, listened to the broadcast and were greatly excited (Liu,
1971).
Throughout the Cultural Revolution, radio would remain the direct
communication link between the central government and the people,
bypassing the interference of local political power and bureaucrats.
Local stations broadcasted programs from the Central People’s Broad­
casting Station, and only had power to create local news content. In fact,
whenever a purge within a given regional party occurred, the radio
stations in that region stopped broadcasting regional content entirely
and only broadcast news from Radio Beijing (Liu, 1971).
The radical politicization of media content in this time period
created a favorable climate of opinion for the Cultural Revolution: it
heightened the morale of Mao’s followers and identified the ideological
enemies, whether perceived or real. Exposure to radio increased the
impression of universality of the political struggle, induced otherwise
apathetic peasants to become state agents, and emboldened them to act.
2.3. Language standardization and policy
Despite the pervasive infrastructure of the radio network across
China, the comprehension of the messages was constrained. The uni­
formity of content in the broadcasts were also reflected in the uniformity
of the broadcast language.
In an attempt to combat localism, it was mandated that all official
news broadcasts through the wired rural broadcast network be con­
ducted in Standard Mandarin. As a result, in some counties where the
native dialect was not Mandarin, it was estimated that only five percent
of audience members could actually comprehend the centrally-relayed
broadcasts from Beijing (Liu, 1971). Zhou Enlai remarked in 1958:
Radio and the cinema are powerful publicity instruments. But as our
common speech has not yet been made universal, their effectiveness
in the districts where only local dialects are spoken is inevitably
limited (Liu, 1971).
S. Ou and H. Xiong
China is a linguistically diverse nation, within which the predomi­
nant language is Chinese, or Hanyu. Chinese itself refers to a collection of
related but often mutually-unintelligible dialects. The varieties of Chi­
nese resemble distinct spoken languages united by a single written script
and shared cognates.5 The varieties of Chinese differ mainly in their
phonology and to a lesser degree, syntax and vocabulary. Linguists have
categorized the varieties in several different ways, but most agree that
there are between seven and ten groups, only one of which is the
Mandarin family.6 There is a general consensus of a North and South
division with more pronounced variation in the rugged South.
The choice of a Mandarin-language broadcast in counties where it
was not understood should not be considered an oblivious oversight by a
non-optimizing bureaucrat. Rather, it was a calculated decision
reflecting the careful tradeoff between the static efficiency of persuasion
and the dynamic efficiency of linguistic standardization. A unified lan­
guage was thought to be instrumental in unified policy for a unified
country. Before the ascendance of the Communist Party in 1949, there
was no national standard language.7 In the incipient years of Communist
rule, the regime was quite tolerant of minority dialects as the Commu­
nist revolution had drawn its support from the largely dialect-speaking
rural peasant population.8 However, by 1955, the government had
become highly cognizant of linguistic barriers to national construction.
In a little publicized Conference on Standardization of Chinese Language
in 1956, a simplified version of Standard Mandarin was devised and
promoted as the common tongue (Liu, 1971; Ramsey, 1987).9
This language policy was implemented through a variety of mecha­
nisms including laws, regulations, education, exams, and restriction of
minority language use in public spaces (Barnes, 1982). In 1956, primary
school teachers were trained in the standard language and the use of
minority dialects in schools or over the airwaves was rebuked (Liu,
1971). Even the primacy of propaganda was subordinate to the directory
of language standardization.
3. Data
Our aggregate analysis makes use of a number of datasets, including:
(i) revolutionary intensity at the county level, proxied by number of
deaths, from Walder (2010–2013), (ii) the predicted strength of radio
signal received at each county seat, and (iii) information on county
characteristics. In our heterogeneity analysis, we also make use of data
5
To fix ideas, a helpful analogy can be drawn to variations within the
Western Romance languages. But unlike the Romance languages, the differ­
ences in Chinese dialects reflect only the spoken form. There is only one written
form of Chinese, which would be used for anyone writing or reading Chinese.
Henceforth we will refer to the varieties of Chinese as dialects for simplicity.
6
The common agreed upon groups include: Mandarin, Hakka, Cantonese,
Wu, Gan, Min, and Hui. This level of categorization subsumes a great deal of
underlying differences as there is local variation even within these broad
groups.
7 the entire sample.
A common governmental language existed in the form of Guanhua during
the dynastic periods but it was used only by the upper echelons of bureaucrats
and magistrates (Ramsey, 1987).
8
It is important to note that differences in spoken Chinese dialects had very
little to do with literacy. Prior to the 20th century, the written language Chinese
was based on classical Chinese used in ancient times and not vernacular Chi­
nese. It was perfectly feasible for an individual to be fully literate yet speak a
non-Mandarin dialect. In this period, there was no sense in which native
Mandarin speakers were more educated (although this may be the case today,
as Mandarin is the official language taught in schools). Since language stan­
dardization was still beginning to take effect at the start of the Cultural Revo­
lution, Mandarin comprehension was a matter of location, not of education.
9
The National Language Unification Commission established the Beijing di­
alect of Mandarin as the standard language of the country in 1932. For expe­
diency, the People’s Republic retained this standard when they took power in
1949. However, active enforcement and promotion only began from 1956 on­
ward (Ramsey,1987).
6
Journal of Development Economics 153 (2021) 102732
that shows the extent of Mandarin intelligibility locally. The rest of this
section describes each of these in detail.
3.1. Cultural revolution intensity
The main outcome of interest is the intensity of the Cultural Revo­
lution. On the county level, this is proxied by the number of killings due
to revolutionary violence. Our analysis uses a county-level dataset on
revolution-related fatalities and victims, digitized from regional gazet­
teers (Walder, 2010–2013). To the best of our knowledge, this is the
most comprehensive dataset of casualties available.
Although much research concerning the Cultural Revolution has
focused on the extent of violence in urban areas, recent research has
found that the violence in rural areas was also extensive. The violence in
rural areas were not simply spillovers from urban regions; instead, po­
litical activity extended far into the countryside. Walder (2014) de­
scribes how he documented the extent of this violence, using
county-level gazetteers. This was part of a two-decade long process to
gather data from gazetteers from every county. A near-complete
collection of information from 2,213 cities and counties was gathered
(Walder, 2010–2013).
Gazetteers are book-length encyclopedias detailing local histories,
demographics, and economics. Gazetteers covering the Cultural Revo­
lution were published in the late-1980s as a consequence of a central
policy directive issued in 1978. Each county was instructed to conduct
official investigations into the period in order to rehabilitate “wrongful”
victims and compensate remaining family members (Su, 2011). National
standards were established with much room for interpretation. Some of
the annals were very lengthy and detailed, including details on the
method of killing, while others were brief and conservative. Although
the resulting annals contained varying degrees of details, they all
included specifics on the number of abnormal deaths attributable to the
revolution.
For instance, these deaths include suicides of individuals under
persecution, deaths in clashes with military or factions, deaths in
struggle sessions or as a result of imprisonment, and executions during
political campaigns. Walder collected this information along with the
number of victims in each county, more loosely defined. Walder and Su
(2003) discuss the issue of data quality extensively. They state that it is
reasonable to assume that the degree of under-reporting in the data
should not be correlated with the severity of the Cultural Revolution
locally.
Walder and Su (2003) find that “no matter how remote from urban
centres, rural counties experienced many of the same political events
that we long assumed were common primarily in cities.” They also find
that out of the 1530 counties for which there was information as of the
publication of the article (out of a total of 1936 counties as of 1988),
almost 87 % experienced a power seizure by a mass organization, and
73 % experienced an armed battle between mass factions.
Fig. 2 illustrates the spatial variation in revolutionary intensity for
The average percent killed in a county is 0.048 percent, with a
standard deviation of 0.164. As noted in previous work, considerable
variation in revolutionary intensity exists (Su, 2011; Walder and Su,
2003).
3.2. Radio data
We construct a measure of radio reception using information on the
location of provincial and municipal radio stations. This data is obtained
from Liu (1971), who identified the location of 141 known provincial
and municipal stations in 1964, just prior to the Cultural Revolution.
The radio stations are located in 114 unique cities. Fig. 3a shows the
geographic distribution of these radio stations across China. As should
be expected, the spatial variation radio station was not random or uni­
form. While the Communist government pursued an ambitious plan to
S. Ou and H. Xiong
Fig. 2. Spatial Variation in Violence during the Cultural Revolution. Note: In
this figure, more darkly-shaded regions correspond to areas with a higher
number of fatalities directly attributed to the Cultural Revolution.
Fig. 3. Spatial Distribution of Radio Stations. Note: This map shows the loca­
tion of radio stations. In the bottom map, the radio stations are overlaid with
the maps of railroad and major waterways. The railroad are shown in dark red
and the rivers are shown in teal. (For interpretation of the references to colour
in this figure legend, the reader is referred to the Web version of this article.)
expand radio infrastructure nationwide, the reality of development was
uneven. This is in part due to the fact that one of the factors that
contributed to the growth of broadcasting power was electricity and the
growth of technology (Liu, 1964).
Sustained radio operation required a consistent source of electrical
power. By the mid 1960s, widespread electrification still eluded
7
Journal of Development Economics 153 (2021) 102732
significant portions of China. Large power plants were concentrated only
in coastal locations and former Japanese occupied regions of Manchuria.
This was despite the call for installation of large hydro and thermal
plants in Beijing’s first Five Year Plan, initiated in 1956. The actuality
electrification campaign consisted of crude generating plants of every
conceivable method; for example, small hydro motors, hand generators,
gas motors, and wind motors (Liu, 1964).
Fig. 3b shows suggestive evidence that as a result of the uneven
development of electrification, radio development was confined to areas
where the power supply was relatively sufficient. Radio station presence
is most prevalent along the main railroad routes at the time as well as
near tributaries of major waterways.
Based on the cross sectional data on radio station locations, we apply
the Irregular Terrain Model (Hufford, 2002) to calculate the predicted
radio signal strength in all localities.10 To implement the ITM algorithm,
we utilize information on radio station locations along with a high res­
olution geo-topographical map of China. For each county, we predict the
radio signal strength at the county seat, where historically the county
receiving stations would have been located.11 Ultimately, we compute
our explanatory variable, Signalc, by dividing the de-meaned signal in­
tensity by its standard deviation.
To account for potential endogeneity in the location of radio stations,
we follow Olken (2009) and simulate the hypothetical signal quality in
free-space (i.e., assuming that the terrain is flat and absent of any
geomorphological obstacles). Fig. 4 displays the geographic variation in
actual and hypothetical signal strength. Conditional on the “free-space”
signal, which captures the variation in signal strength driven by prox­
imity to transmitters, the coefficient on Signalc is identified only by the
residual variation in propagation patterns caused by idiosyncratic
topography features, which is plausibly exogenous.
3.3. Linguistic data
To explore the effects of differences in ability to comprehend radio
broadcasts on violence, we gather data on regional vernacular dialects
and heterogeneity in their compatibility with Standard Mandarin. We
assemble this data in two steps. First, we identify the spatial variation in
dialects across China from The Language Atlas of China. The Language
Atlas is a compilation of local linguistic studies documenting Chinese
dialects and their genealogical relationships. The digitized data is
organized at the county level. It records the primary dialect spoken in
each county, other minor dialects if present, and the dialect families they
belong to.12
Second, to measure each dialect’s linguistic distance to Mandarin, we
appeal to experimental data collected by linguists in the field. Tang and
Van Heuven (2009) study the strength of pairwise mutual intelligibility
between Chinese dialects. They relate functional intelligibility between
dialects to proximity in lexical structural and phonological regularity.
The authors conducted an extensive set of experiments in order to
find the mutual intelligibility between pairs of Chinese dialects. In
10
The Irregular Terrain Model (ITM) was originally developed by the US
government for frequency planning purposes and allows one to accurately
predict signal strength across narrow geographical cells (Phillips et al., 2011).
The model computes the signal loss between transmitting and receiving loca­
tions, accounting for the physical distance and topography that lies in between.
It has also been employed by Olken (2009), Enikolopov et al. (2011), and
DellaVigna et al. (2014).
11
Due to data limitations in our radio data, the values of broadcast frequency,
transmission power, and height of towers are imputed.
12
Although the majority of the languages are spoken by those of Han
ethnicity, in our robustness section we restrict our attention to linguistic di­
versity within the ethnically Han population and exclude observations, pri­
marily in autonomous regions, where non Chinese languages, such as Turkic,
Altaic, or Mongolian languages are observed. This mitigates the possible con­
founding bias of ethnicity.
S. Ou and H. Xiong
Fig. 4. Predicted Radio Signal. Note: The maps represent the geographic dis­
tribution of the simulated intensity of radio signal in 1964, under real condi­
tions (top) and in the absence of geomorphological obstacles (bottom). The
geographic boundaries of the map of China are dictated by the availability of
violence data from Fig. 2.
particular, 150 native speakers of each of the 15 different Chinese di­
alects were subjected to a listening exam where they were asked to
identify words and sentences read by speakers of another dialect,
including their own. The listening exam was administered via a
recording of 288 standard Chinese core words read by a native speaker
of each dialect. The participants resided in rural areas and were around
the age of 50 in 2009, and thus, were youths at the time of the Cultural
Revolution. Therefore, their intelligibility scores would closely reflect
intelligibility of youths during the time period we are interested in.
These participants were also selected because they had never traveled
outside of their home province. From this experiment a measure of
bilateral intelligibility between dialects was compiled. From the tables
in their paper, we compile the ability of listeners of each dialect to
correctly identify words from the Beijing Mandarin dialect, which
closely resembles Standard Mandarin.
We construct our analytical dataset by combining the spatial data on
the dialects spoken by county and the proximity of each of these dialects
to the Mandarin family and Standard Mandarin. Since only 15 dialects
were studied instead of the entirety of Chinese dialects, we use each of
8
Journal of Development Economics 153 (2021) 102732
these dialects as the representative of the family of dialects that they
originate from. We standardize these comprehension measures by sub­
tracting the percent understood of one’s native dialect. Fig. 5 maps the
geographical variation of the underlying data.
3.4. Control data
We complement the above dependent and explanatory variables
with additional sources of control data, which we briefly outline here.
The socio-economic and demographics information come from the 1964
Census, which was the last census enumeration prior to the Cultural
Revolution. This data is obtained from the University of Michigan’s
China Data Center. We also use the China Historical GIS (CHGIS) digital
map collection of Harvard University. Using the maps, we compute the
proximity to nearest navigable river and distance to the coast from the
centroid of each county as well as the number of historical Buddhist
temples contained in each county. Data on county-level railroad access
as of 1961 is created from rail network files provided by Baum-Snow
et al. (2017). Ruggedness and terrain feature data are constructed as per
instructions from Nunn and Puga (2012). Finally, data on Communist
Party membership in 1956 comes from Kasahara and Li (2020). Table 1
shows the summary statistics of select explanatory and dependent
variables.
4. Empirical strategy & results
In this section, we examine whether radio propaganda affected the
contemporaneous intensity of the Chinese Cultural Revolution. The
outcome we study is the number of killings directly attributable to the
Cultural Revolution. The main empirical challenge in our setting is
separating the causal effect of the radio signal from other potential
confounders that are correlated with signal and also affect violence. For
instance, one could be concerned that if radio receiving stations were
strategically placed in areas where individuals were expected to be more
receptive to the radio messages and more prone to violence.
In order to address this concern, we follow papers such as Yanagi­
zawa-Drott (2014), Olken (2009), and Enikolopov et al. (2011), and
construct a measure of the free-space signal, which is the radio signal in
the absence of geomorphological obstacles. By controlling for the
free-space signal, we are able to isolate the component of radio signal
that is due to topographical features, and plausibly exogenous. In
addition to controlling for the free-space signal, we also control for other
factors that are potentially correlated with violence, including popula­
tion measures, linguistic fragmentation, railroad access, distance to
Beijing and provincial capitals, and Communist Party membership.
Identification relies on the exogeneity of Signalc, conditional on ob­
servables. As a test for exogeneity, we regress other determinants of
violence on Signalc and the free-space signal, to assess whether radio
signal is correlated with other determinants of violence. The regression
specification for the test for exogeneity is as follows:
zc = φ Signalc + δFree Signalc + γXc + λp + ∈c
The left-hand side variable, zc, represents our other county-level
control variables. If Signalc is exogenous, we should expect ϕ to be
zero. Table 2 shows the results, and we see that ϕ is insignificant in all of
these regressions, with the sole exception of the log distance to pro­
vincial capital, which is significant at the 5 % level. In addition to the
chance of finding a significant result among multiple regressions, it is
also unlikely that distance to provincial capital would explain the
baseline results. Nevertheless, we still control for the distance to pro­
vincial capitals throughout our analyses. Since radio signal is not
correlated with nearly all observed determinants of violence, this lends
credibility to our identification strategy.
S. Ou and H. Xiong Journal of Development Economics 153 (2021) 102732

Fig. 5. Linguistic Distance to Mandarin. Note: The maps present the geographic distribution of linguistic distance to Mandarin as it is operationalized in the paper.
The geographic boundaries of the map of China are dictated by the availability of violence data from Fig. 2. On the left, the red region indicates areas whose main
dialect belongs to the Mandarin family while the white region indicates areas whose main dialect does not belong to the Mandarin family. On the right, redder shades
represent a higher fraction of Beijing Mandarin words identified by speakers (of the dialect of vernacular Chinese) in that region, while greener shades represent a
smaller fraction identified. The gray regions indicate missing data from the linguistic study, Tang and Van Heuven (2009). We note that there is variation in
comprehension of Beijing Mandarin even within the Mandarin family of dialects. . (For interpretation of the references to colour in this figure legend, the reader is
referred to the Web version of this article.)

Party membership in 1956).13 Importantly, to account for unobservable

Table 1
characteristics that could vary systematically with proximity to radio
Summary statistics.
stations, we include the free-space signal strength Free Signalc.
Variable Obs Mean Std. Dev.

Percent Killed per County 1985 .048 .164 4.1. Baseline results
Signal 1985 0 1
Mandarin 1985 .666 .472
We present our results in Table 3 and display the effect of signal
Distance from Beijing 1985 1201.898 649.694
Altitude 1985 .345 .428 strength on conflict across several different specifications. To start, in
Area 1985 3566.197 8725.911 column (1), the baseline controls are hypothetical radio signal in the
Population 1985 310695.8 287279.8 absence of geomorphological obstacles, ruggedness, log 1964 popula­
River dummy 1985 .455 .498 tion, elevation, county area, administrative status, distance to provincial
Coast dummy 1985 .12 .326
Ruggedness 1985 11.398 1.292
capitals, latitude and longitude. We successively add more controls in
Distance to closest provincial capital 1985 120.973 100.877 the subsequent columns. In column (2), we add province fixed effects. In
Railroad access 1985 .325 .469 column (3), we add controls for Communist Party Membership and
Non agricultural population 1985 .133 .248 distance to Beijing, which are intended to proxy for state capacity. In
Male female ratio 1985 1.071 .116
column (4), we include the controls for transport infrastructure: railroad
Households 1985 70587.17 64245.61
Township dummy 1985 .101 .301 access and waterway access.
Linguistic fractionalization 1985 .038 .139 Finally, in column (5), we add pre-existing socioeconomic condi­
Buddhist temples 1985 1.045 2.361 tions, population and demographics. This includes: 1964 county gender
CCP membership 1985 3707.15 3463.99 ratio, 1964 fraction of non-agricultural population, the number of
Note: All distance measurements are in kilometers (km). Area is in square km. households in 1964, the number of historical Buddhist temples and the
Mandarin is an indicator for if the main dialect of a particular county belongs to degree of ethnolinguistic fragmentation. We also include an indicator for
the Mandarin family of dialects. whether the main local dialect in a county belongs to the Mandarin
language group. The number of Buddhist temples is a proxy for historical
Eq. (1) describes the main specification: socio-economic status while ethnolignusitic fragmentation has been
show to affect intensity of conflict in other settings. This is our preferred
yc = β Signalc + δ Free Signalc + γXc + λp + ∈c , (1)
specification. Across all specifications, we cluster at the level of the
Signalc is the predicted signal strength in county c, Xc is a vector of nearest radio station.
county characteristics and λp is a province fixed effect. The outcome, yc, The estimated effect of radio reception is statistically significant and
is the casualty rate directly attributed to the events of the Cultural quantitatively important. A one standard deviation increase in signal
Revolution in county c. The coefficient of interest is β, which measures strength is associated with a 0.014 percentage point increase in the
the effect of radio signal. percent of population killed. The effect is meaningful as the average
The vector, Xc, is a set of county-level characteristics. This includes percent killed per county is 0.048 percent.
controls for preexisting demographic conditions (log population in Interestingly, we also detect a positive and significant coefficient on
1964, the share of non-agricultural population, gender ratio, number of the standardized CCP membership variable, indicating that counties
households, administrative status); controls for transportation infra­ with greater share of population belonging to the CCP experienced
structure and geography (the ruggedness of the terrain, county area, higher fatalities during the Cultural Revolution. Comparing the
elevation, river access, railroad access, latitude and longitude); controls
for pre-existing socioeconomic conditions (the number of Buddhist
13
temples within the county, an index of ethnolinguistic fractionalization, Previous work has shown that state capacity can facilitate the mobilization
whether the county is Mandarin speaking); and finally controls for state of civilians into violence. See Bonnier et al. (2020). Communist Party mem­
capacity (distance to Beijing, distance to provincial capitals, Communist bership data comes from Kasahara and Li (2020) and are normalized with
respect to the mean and standard deviation.

9
S. Ou and H. Xiong Journal of Development Economics 153 (2021) 102732

Table 2
Exogeneity test.
Log Log # Buddhist Gender Linguistic Non-Ag. Railroad Log Dist. Log Dist. CCP
population households temples ratio fragmentation population access Beijing Prov. Capital membership

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)

Signal − 0.001 0.005 − 0.039 0.007 − 0.063 − 0.003 0.024 − 0.003 − 0.112** − 75.631
strength (0.009) (0.009) (0.040) (0.007) (0.054) (0.008) (0.024) (0.011) (0.046) (128.449)
Observations 1985 1985 1985 1985 1985 1985 1985 1985 1985 1985
R2 0.989 0.990 0.242 0.329 0.792 0.200 0.271 0.935 0.472 0.463
Mean of dep. 12.305 10.819 0.473 1.071 9.716 0.038 0.325 13.805 5.116 3707.154
var.

Note: Each column is a separate regression showing the coefficient on signal strength in the regression of corresponding correlate (column header) on the signal
strength under real conditions, the radio signal under hypothetical terrain, and the full set of controls excluding the dependent variable. * p < .10, ** p < .05, *** p <
.01.

coefficient for radio signal to that of state personnel (which are both
standardized), we find the effect of radio signal to be notably larger.
4.2. Robustness
To assess the robustness of the results, we conduct the following
checks: using Conley standard errors; clustering by province and dialect
group; weighting by population; using different subsamples (omitting
areas with a high percentage of ethnic minorities and excluding counties
containing radio stations); and analyzing the effects at the intensive
margin. The results are displayed in Tables 4 and 5.
approaches to statistical inference. In column (1) we use Conley stan­
dard errors with a 100 km cutoff for spatial autocorrelation. In column
(2) we use a 150 km cutoff, and in column (3) we use a 200 km cutoff. In
columns (4) and (5), we cluster at the province and dialect level,
respectively. The results remain robust throughout.
4.2.2. Weighted regression
To show results that are representative for the average person rather
than the average county, we weight the regression by county population.
The results are shown in Table 5, column (1), and are very similar to the
results using the main specification.

4.2.1. Clustering 4.2.3. Restricted sample

In our main specification, we show our results clustered at the One potential concern is that our estimates are driven by targeted
nearest radio station level. In Table 4, we show results using alternative violence towards ethnic minorities in peripheral or autonomous prov­
inces. This would mean that our results could be confounded by the
effect of ethnic-based conflict. To address this, we restrict our attention
Table 3
to what is traditionally referred to as China “proper” and areas that are
Effect of radio broadcasts on cultural revolution intensity.
uniformly ethnically Chinese.
Dep. Var Percent of population killed Contemporary China consists of 23 provinces and 5 autonomous
(1) (2) (3) (4) (5) regions. By restricting samples to only the territories within the Great
Signal strength 0.021* 0.013* 0.014** 0.013* 0.014** Wall, or China “proper,” we demonstrate that our results are stable
(0.012) (0.007) (0.007) (0.007) (0.007) across sub-populations (see Table 5, column (2)). We exclude the border
CCP membership – – 0.007* 0.007* 0.007* provinces: the Northwestern provinces (Gansu, Qinghai, Ningxia,
(0.004) (0.004) (0.004)
Xizang (Tibet), and Xinjiang), the Southwestern provinces (Hainan,
Control variables
Baseline X X X X X Sichuan, and Yunnan), and the Northeastern provinces (Inner Mongolia,
Province F.E. – X X X X Heilongjiang, and Jilin). This implies that the effect of radio broadcasts
Distance to – – X X X on violence is not explained by a story of the core versus periphery parts
Beijing of China, or by the ethnic-based violence that may have occurred during
Transportation X X
the Cultural Revolution.
– – –
infrastructure
Demographics – – – – X
Buddhist temples – – – – X 4.2.4. Intensive margin
Linguistic – – – – X Another concern may be with regard to data quality. For instance,
fragmentation
counties that report zero killings may in fact be under-reporting due to
Observations 1985 1985 1985 1985 1985
# of Clusters 114 114 114 114 114
poor historical records. Therefore, there is a concern that we could be
R2 0.096 0.215 0.217 0.218 0.222
Mean of dep. var. 0.048 0.048 0.048 0.048 0.048

Note: This table shows the effect of radio exposure on violence. A unit of Table 4
observation is a county in the 1964 census. The dependent variable in each Effect of radio broadcasts on cultural revolution intensity: Alternative inference.
column is the percent of the population killed in each county during the Cultural Conley Conley Conley Cluster by Cluster by
Revolution. The explanatory variable of interest is the radio signal strength. The 100 km 150 km 200 km province dialect
baseline controls are hypothetical radio signal in the absence of geomorpho­
(1) (2) (3) (4) (5)
logical obstacles, ruggedness, log county population in 1964, county adminis­
trative status, elevation, county area, distance to provincial capital, latitude and Signal 0.014** 0.014** 0.014*** 0.014* 0.014**
longitude. Transportation infrastructure includes railroad and canal access. Strength (0.006) (0.006) (0.005) (0.007) (0.007)
Observations 1985 1985 1985 1985 1985
Demographics refer to the 1964 county gender ratio, 1964 fraction of non-
R2 0.222 0.222 0.222 0.222 0.222
agricultural population and the 1964 log number of households. Linguistic
fragmentation includes the number of dialects in the county and an indicator for Note: This table explores the robustness of the main findings from Table 3 to
whether the main language spoken in a county belongs to the Mandarin lan­ alternative clustering and statistical inference. Conley standard errors with 100
guage group. The Signal Strength and CCP Membership variables are both km, 150 km, and 200 km cutoffs are presented in columns (1)–(3). In columns
standardized. Standard errors are clustered at the level of the nearest radio (4) and (5), standard errors are clustered at the level of the province and the
station.* p < .10, ** p < .05, *** p < .01. local dialect group, respectively. * p < .10, ** p < .05, *** p < .01.

10
S. Ou and H. Xiong Journal of Development Economics 153 (2021) 102732

Table 5 Table 6
Effect of radio broadcasts on cultural revolution intensity: Robustness. Effect of radio broadcasts by local Mandarin comprehension.
Weighted Restricted Intensive Excl. Control Dep. Var: Percent of population killed
by pop. sample margin radio for
(1) (2) (3) (4) (5)
counties diffusion
Signal strength 0.014** 0.002 − 0.398 0.015** − 0.871
(1) (2) (3) (4) (5)
(0.007) (0.007) (0.872) (0.006) (0.926)
Signal 0.014** 0.027** 0.016* 0.012* 0.014* Signal strength × – 0.018* 0.022** – –
strength (0.006) (0.013) (0.009) (0.006) (0.008) Mandarin (0.010) (0.010)
Observations 1985 1330 1415 1889 1689 Signal strength × – – – 0.058* 0.080**
R2 0.226 0.250 0.254 0.225 0.233 Mandarin (0.035) (0.032)
intelligibility
Note: This table explores the robustness of the main findings from Table 3. Full controls X X X X X
Column (1) uses population-weighted least squares. Column (2) excludes border, Full controls × – – X – X
autonomous, and minority provinces — restricting the analysis to what is Signal strength
considered China “proper”. In column (3), we restrict the sample to counties Mandarin – – X – X
with non-zero values of killings. Column (4) excludes the counties containing (Intelligibility)
radio stations. In column (5), we exclude the counties containing radio stations × Free space
and control for the percent of population killed in counties containing radio sign
Observations 1985 1985 1985 1969 1969
stations. 2
R 0.222 0.224 0.232 0.224 0.231
*p < .10, ** p < .05, *** p < .01.
Note: This table illustrates how the effect of radio exposure varies with Mandarin
comprehension. A unit of observation is a county in the 1964 census. The
conflating missing data with zero casualty rates. To address this possi­
dependent variable in each column is the percent of the population killed in each
bility, we omit counties with no reported deaths. In Table 5, column (3),
county due to the Cultural Revolution. The Signal Strength variable is stan­
we show results on the intensive margin only (restricting to only
dardized. Mandarin is an indicator variable taking a value of one if the primary
counties with positive killings). The results are qualitatively similar, and dialect spoken in a county belongs to the Mandarin family. Mandarin Intelligi­
show a stronger effect, compared to the main result. bility is the percentage of Mandarin words correctly identified by a respondents
from each dialect group from the listening experiment conducted by Tang and
4.2.5. Excluding counties with radio station Van Heuven (2009). Full controls correspond to the full set of controls from
To rule out the concern that the main results may be driven by the column (5) of Table 3, including CCP Membership. * p < .10, ** p < .05, *** p <
counties with a radio station, we exclude counties having a radio station .01.
in Table 5, column (4). Again, the results are qualitatively similar to the
main result. to the Mandarin language group, and zero otherwise. To construct the
indicator variable, we use data from the Language Atlas of China, which
4.2.6. Controlling for diffusion we describe in more depth in the Data Section. The Language Atlas was
Finally, to rule out the concern that the main result may be driven by published in 1987; however, the spatial distribution of languages has
diffusion of conflicts, we repeat the previous exercise as in column (4), remained very similar over time. A linguistic map from 1943 shows that
but also control for the percent of population killed in cities with radio the language patterns are largely the same as the ones from the Lan­
stations. We show our result in column (5) and find a qualitatively guage Atlas (Chao, 1943).
similar result to the main result. Second, we use a continuous measure of linguistic distance, Man­
darin Intelligibilityc. This is the percentage of the 288 core Mandarin
words correctly identified in a listening exam, by a sample of 150
4.3. Heterogeneity speakers of the dialect in county c.14 The continuous intelligibility
measure, which utilizes more variation than a binary Mandarin indica­
In this section, we explore how the effect of radio signal on violence tor, addresses concerns that our results are driven by fundamental dif­
can vary with differences in comprehension of Mandarin. As discussed in ferences between Mandarin and non-Mandarin counties.
the background section, radio broadcasts were conducted primarily in We show the results in Table 6. In column (1), we show the main
Standard Mandarin, as opposed to the local dialect. This created a lan­ result from the uninteracted specification (see also column (5) of
guage barrier in the exposure to media that is unique to our setting. Table 3). In column (2), we show results using Mandarinc as a measure of
In China, differences in spoken dialects are stark and the degree of linguistic distance. The controls are the same as in column (1). In column
mutual intelligibility between them vary significantly. However, rela­ (3), we add a set of controls, interacting our controls with signal strength
tive to other modern nation states with the same level of linguistic di­ as well as Mandarinc with the free space signal. We add these controls to
versity, dialect variation in China stems less from ethnic differences. address concerns that Mandarin could be endogenous, or that the radio
Rather, these linguistic differences reflect historical migration patterns signal could be endogenous. We find a significant and positive coeffi­
and diffusion of groups that are often no longer salient in modern times. cient of 0.022 on the interaction term, meaning that a one standard
In fact, much of this meaningful dialect variation occurs in what is deviation shift in radio signal causes a 0.022 percentage point increase
considered China “proper” within a population of people that are in percent killed in a Mandarin-speaking county compared to a non-
broadly classified as Han Chinese, allowing us to explore linguistic Mandarin speaking county. This is qualitatively similar, and higher in
heterogeneity in a setting of relative ethnic homogeneity. magnitude, compared to our baseline result. We also find that the co­
To understand the effect of linguistic barriers on the effect of radio, efficient on Signalc is insignificant, showing that the combination of
the following specification is estimated: intelligibility and radio signal are necessary for radio broadcasts to
matter.
yc = β Linguistic Distc ⋅Signalc + α Signalc + σ Linguistic Distc
In columns (4) and (5), we show the analogous results to column (2)
+ δ Free Signalc + γXc + λp + ∈c , (2) and (3), but using Mandarin Intelligibilityc, a continuous measure of
The parameter of interest is β, which shows how the effect of Signalc
can vary with comprehension of Mandarin. We construct the Linguistic
Distc variable in two different ways. First, we define an indicator vari­ 14
The Mandarin Intelligibility measure is created from data collected in Tang
able, Mandarinc, equal to one if the main local dialect in county c belongs and Van Heuven (2009), discussed in detail in Section 3.3.

11
S. Ou and H. Xiong
intelligibility, as a measure of linguistic distance, instead of Mandarinc.
One might think that Mandarin speakers are somehow systematically
different and have different unobservables from non-Mandarin speakers.
This analysis mitigates those concerns by allowing us to utilize linguistic
variation within the Mandarin and non-Mandarin groups. We find a
positive and statistically significant coefficient on the interaction term,
and a statistically insignificant coefficient on the signal strength itself.
Altogether, our results suggest that the effectiveness of broadcasts
was constrained by local linguistic compatibility.
5. Individual-level long-term outcomes
Thus far, we have provided evidence that the violence of the Cultural
Revolution was higher in areas with stronger radio signal, and particu­
larly in those areas that were also Mandarin-speaking. In this section, we
examine the long-term effect of state propaganda at the individual level.
Living in a Communist regime has been found to have a persistent effect
on preferences and attitudes (Alesina and Fuchs-Schündeln, 2007). Our
paper provides context for this persistence by studying the effect of
media exposure at a critical age juncture. Using micro-data from the
retrospective China Family Panel Studies (CFPS) survey, we augment
the cross sectional analysis of the previous section by introducing cross
birth-cohort variation in exposure and within county variation in spoken
dialect. This builds on our previous result, relating media exposure to
linguistic compatibility.
We explore later-in-life outcomes, both in terms of behavior and
beliefs, which are suggestive of ideological indoctrination. Specifically,
we consider joining the Communist Party, tolerance of economic
inequality, and attitude towards wealth as measures of ideological ac­
tions and beliefs.
5.1. Empirical strategy and results
Individual-level micro-data comes from the China Family Panel
Studies (CFPS), a cross-sectional and retrospective survey conducted in
2010 by the Institute of Social Science Survey of Peking University in
China. It collected data relating to respondents’ educational outcomes,
family dynamics, migration and health. Demographic information
include date of birth, province of birth, province of residence at various
points throughout one’s childhood, gender, ethnicity, parents’ occupa­
tion, hukou status, and language spoken at home. The CFPS dataset
consists of 33,000 individual observations from selected counties across
China. After dropping observations with relevant missing data, we are
left with 23,930 observations.
We measure Communist Party membership through the following
questions in the CFPS questionnaire:
• “Are you a member of the Communist Party of China?” We refer to
joining the Communist Party within 45 years of birth as Communist.15
To measure long-term effects on ideological beliefs, we examine the
responses to the following questions:
• “How much do you agree with the following statement? The income
gap has to increase for economic prosperity?” This is referred to as
tolerance of inequality..16
15
We exclude joining the Communist Party after this age because we focus on
ideological motivation from one’s youth. The motivations to join in older age is
less plausibly a result of this. Only 5 % of individuals join after 45.
16
This is measured on a 5-point scale, which we convert to a binary variable
equal to 1 if an individual agrees with this statement.
12
Journal of Development Economics 153 (2021) 102732
• “How much do you agree with the following statement? Wealth
reflect an individual’s achievement?” We refer to this as attitude to­
wards wealth.17
A key feature of our data which allows us to determine the effect of
media on individuals is that we are able to observe the language re­
spondents spoke at home. Having highlighted the heterogeneity with
respect to linguistic compatibility in the previous section, we use indi­
vidual comprehension of Mandarin during the Cultural Revolution as a
proxy for media exposure.
Our identification strategy involves a difference-in-differences
framework in which the two sources of variation are: 1) living
through the Cultural Revolution during one’s youth (the ages during
which one is most susceptible to propaganda), and 2) the comprehen­
sion of Mandarin during one’s youth. In this framework, differences
between the Mandarin and non-Mandarin speakers for that cohort in
excess of the differences from other cohorts could be interpreted as the
effect of radio propaganda.18
To measure Mandarin comprehension, we use a dummy variable
indicating whether the primary language used in daily communication
with the respondent’s family is Mandarin Chinese, as opposed to a
Chinese local dialect or minority ethnic dialect. This variable is referred
to as Mandarin. We argue that this variable is not a choice variable. The
language spoken at home depends on one’s parents’ native language as
well as the location of upbringing. While individuals may have been
incentivized to learn Mandarin later in life, that second language
acquisition would unlikely have affected the dialect spoken at
home.19Hence, this measure of language spoken at home is plausibly
exogenous to the political environment, and reflects individuals’ ability
to understand Mandarin during their youth.
To account for time-invariant differences between Mandarin
speakers and non-Mandarin speakers, we introduce a second source of
variation: exposure and susceptibility to propaganda across age cohorts.
To identify the ages when individuals are most susceptible to propa­
ganda, we use evidence from the psychology literature, which suggests
that adolescents and youths are be more susceptible to media than those
of other age groups. The impressionable years hypothesis states that the
historical environment to which one is exposed to during the transition
between adolescence and adulthood has a profound impact on one’s
attitudes and world views. After this time of plasticity, beliefs become
set and permanent. Young adults are especially vulnerable to shifts in
attitudes in political beliefs (Alwin and Krosnick, 1991; Flanagan and
Sherrod, 1998). There is no consensus for which ages exactly constitute
the impressionable ages. Some see the age of 18 as the end point, but
others see the process of socialization lasting until the age of 25.
The history of the Cultural Revolution itself also suggests that ado­
lescents might be the most influenced by propaganda during this time
period. The Communist Party specifically targeted youth to join the
revolutionary cause, and most Red Guards were between 14 and 23
years old (Jing, 1991). The Cultural Revolution was not the first instance
in history during which youth were the targets of propaganda. During
the Nazi regime in Germany, youth were similarly targeted due to their
naivete (Hoffmann, 1996). Giuliano and Spilimbergo (2014) examines
outcomes on beliefs among those who lived through recessions during
their impressionable years. Medical literature has also found strong as­
sociations between adolescents and risky behavior (Escobar-Chaves and
17
This is measured on a 5-point scale, which we convert to a binary variable
equal to 1 if an individual agrees with this statement.
18
We are unable to perform the same specification in the previous section,
which uses signal strength as an additional source of variation, because the
nature of the data is coded. We cannot observe the actual identities of the
counties.
19
Even if individuals did choose to learn Mandarin, and choose to then speak
Mandarin at home, our results would be attenuated.
S. Ou and H. Xiong Journal of Development Economics 153 (2021) 102732

Table 7
Summary statistics: Individual-level long-term outcomes.
Non-Mandarin Mandarin All

Mean Std. Dev. Mean Std. Dev. Mean Std. Dev.

Dependent variables
Communist Party 0.0642 0.245 0.0857 0.280 0.0684 0.252
Tolerance of inequality 0.518 0.500 0.466 0.499 0.508 0.500
Attitude toward wealth 0.698 0.459 0.673 0.469 0.693 0.461
Independent variables
Urban (Hukou status) 0.418 0.493 0.607 0.489 0.455 0.498
Education (years) 6.658 4.845 8.200 4.698 6.956 4.855
Age 46.50 15.34 47.24 15.07 46.64 15.29
Income (RMB) 9738.2 19582.3 11649.7 24853.9 10107.5 20718.7
# of Siblings 2.734 1.894 2.578 2.068 2.703 1.930
Gender 0.498 0.500 0.478 0.500 0.494 0.500
Father’s level of education 1.929 1.091 2.224 1.219 1.986 1.123
Mother’s level of education 1.488 0.850 1.823 1.090 1.553 0.911
N 19,307 4623 23,930

Note: This table provides descriptive statistics for independent and dependent variables for all respondents in the analytic sample, as well as for Mandarin and non-
Mandarin speakers separately. Communist Party, Tolerance of inequality, and Attitude toward wealth are binary variables. Communist Party equals 1 if an individual attains
Communist Party membership. Tolerance of inequality equals 1 if an individual believes that the income gap has to increase for economic prosperity. Attitude toward
wealth equals 1 if an individual agrees that wealth reflects an individual’s achievement.

Anderson, 2008; Klein et al., 1993). with language use at home.

For these reasons, we consider individuals aged 11 to 20 at the start
of the Cultural Revolution, born between 1945 and 1957, as the cohorts
differentially exposed to the Cultural Revolution rhetoric. We refer to
this cohort as the Cultural Revolution Cohort. This definition incorporates
both the historical details as well as evidence from psychology literature
outlining the most impressionable years.
Thus, we explore the effects of understanding Mandarin among the
cohorts for whom propaganda was the most salient. Although we do not
measure radio directly as an explanatory variable, we believe that media
exposure is the most likely explanation for the measured effect. We
control for signal strength flexibly with a set of county fixed effects.
The difference-in-differences non-parametric event study model we
estimate is as follows:
∑ ∑
yijc = αMandarini + γj cohortj + δj cohortj ⋅Mandarini + ωc + Xi β + ∈i
j j
(3)
In the equation above, yijc denotes the outcome measure of individual
i of cohort j residing in county c prior to the Cultural Revolution. These
outcomes are: participation in the Communist Party, tolerance of
inequality, and attitudes on wealth. The independent variables, cohortj,
are a set of dummies for the birth cohorts that individuals belonged to.
The cohort categories are defined by the age at the start of the Cultural
Revolution in 1966: 30 and older, 21 to 30, 11 to 20, 0 to 10, and born
after the Cultural Revolution. The individual controls are interacted
with cohortj to allow them to vary flexibly by birth cohorts. Mandarini is a
dummy for speaking Mandarin at home. ωc is a county fixed effect which
partials out the difference in outcome arising from radio infrastructure
or local political conditions. The estimated vectors δj, the set of co­
efficients on the interaction term between cohortj and Mandarini, reveal
the effect of speaking Mandarin at home on the outcome for individuals
belonging to each birth cohort. If, for example, the Cultural Revolution
propaganda increased participation for the Cultural Revolution Cohort,
then we would expect δj to be positive and significant only for the birth
cohorts 11 to 20.
Xi is a vector of individual level controls, which include gender,
education, age, ethnicity, occupation, county of residence at age 12,
urban residence, number of siblings, income, and hukou status. Impor­
tantly, we also control for a robust set of parental characteristics that
may affect the intergenerational transmission of political status. This
includes father’s birth year, mother’s birth year, mother’s education,
and father’s education. Summary statistics are shown in Table 7. These
factors may jointly determine the outcome variable, and be correlated
For instance, we control for parental demographics and character­
istics because parental background may directly affect ideological
behavior through the intergenerational transmission of beliefs. Addi­
tional controls include interactions between birth cohort indicators and
education, gender, urban dummy, birth province, and ethnicity. This
allows the control variables to affect the outcome differently for the
Cultural Revolution cohort. While radio signal does not explicitly appear
in the estimation equations, by utilizing within county variation, we
control for differences in radio signal across counties.
The estimates of Eq. (3) are show in Table 8. In the first set of col­
umns for each outcome, we control for baseline individual-level controls
as described above. In the second set of columns for each outcome, we
interact each of these controls with birth cohorts, which is our preferred
specification. Several consistent patterns emerge. We find that in­
dividuals who are native Mandarin speakers and belong to the Cultural
Revolution Cohort are 4.5 percentage points more likely to have joined
the Communist Party, compared to non-Mandarin speakers in other
birth cohorts. The change in likelihood of joining the Communist Party
for native Mandarin speakers is also only statistically significant for the
Cultural Revolution Cohort.
This finding is significant given that joining the Communist Party is a
rigorous process which involves a lengthy application process and
scrutiny. Membership to the Communist Party is exclusive, consisting of
5 % of the population.20. The results here contributes to a strand of
literature that tries to understand the determinants of Communist Party
membership.21
Additionally, individuals exposed to state media were more con­
cerned about economic inequality and had less adherence to wealth. In
particular, those who belong to the Cultural Revolution Cohort, who are
native Mandarin speakers are 6.2 percentage points less likely to be
tolerant of inequality, and 5.0 percentage points less likely to have a
positive attitude toward wealth. In both of these specifications, the ef­
fects on native Mandarin speakers in other non-Cultural Revolution
Cohorts are small and not statistically significant. This precludes
competing explanations on the basis of persistent differences between
Mandarin and non-Mandarin speakers as driving these effects.
Overall, we find that native Mandarin speakers who lived through
the Communist regime during their impressionable years exhibit
20
Individuals must submit multiple applications and endure several rounds of
evaluations. See www.nytimes.com/ref/college/coll-china-politics-002.html
21
See Appleton et al. (2009) for a comprehensive literature review.

13
S. Ou and H. Xiong Journal of Development Economics 153 (2021) 102732

Table 8
Long-term individual-level outcomes.
Communist party member Tolerance of inequality Importance of wealth

(1) (2) (3) (4) (5) (6)

Age 31+ × Mandarin − 0.069 (0.048) − 0.074 (0.048) 0.013 (0.079) − 0.018 (0.077) − 0.027 (0.064) − 0.055 (0.067)
Age 21-30 × Mandarin 0.013 (0.029) 0.016 (0.028) − 0.075* (0.045) − 0.056 (0.043) − 0.037 (0.043) − 0.020 (0.044)
Age 11-20 × Mandarin 0.043*** (0.016) 0.045*** (0.016) − 0.062** (0.026) − 0.062** (0.026) − 0.051* (0.028) − 0.050* (0.027)
Age 0-10 × Mandarin 0.009 (0.014) 0.011 (0.014) − 0.019 (0.024) − 0.016 (0.025) − 0.005 (0.024) − 0.007 (0.025)
Born After CR × Mandarin − 0.003 (0.010) − 0.001 (0.010) 0.027 (0.023) 0.022 (0.022) 0.018 (0.019) 0.014 (0.019)
Baseline controls: X X X X X X
× Birth Cohorts: – X – X – X
Observations 23930 23930 23930 23930 23930 23930
R2 0.190 0.202 0.193 0.196 0.132 0.134

Note: This table shows the effect of speaking Mandarin on individual outcomes by birth cohorts. The outcomes we consider are: joining the Communist Party, tolerance
of economic inequality, and attitude towards wealth. Communist Party, Tolerance of inequality, and Attitude toward wealth are binary variables. Communist Party equals 1
if an individual attains Communist Party membership. Tolerance of inequality equals 1 if an individual believes that the income gap has to increase for economic
prosperity. Attitude toward wealth equals 1 if an individual agrees that wealth reflects an individual’s achievement. Mandarin is an indicator for speaking Mandarin at
home. Birth cohorts are defined based on age at the start of the Cultural Revolution, in 1966. Baseline controls include gender, education, age, ethnicity, county of
residence at age 12, urban residence, number of siblings, hukou status, income, occupation, father’s education, mother’s education, father’s birth year, mother’s birth
year, and province fixed effects. Birth cohort dummies and an indicator for whether the primary language used in daily communication with the respondent’s family is
Mandarin Chinese are also included. Standard errors are clustered on the county of residence before the Cultural Revolution, and county of birth if born after the
Cultural Revolution. * p < .10, ** p < .05, *** p < .01.

behavior and form long-term ideological beliefs which are aligned with
Communist Party ideology. Our results show that the differential out­
comes of Mandarin speakers were unique to those belonging to the
Cultural Revolution cohort. This is consistent with the notion that in­
dividuals of this particular cohort were exposed to Mandarin-language
propaganda at a critical age juncture. This is also consistent with our
contemporaneous findings: that exposure to propaganda during the
Cultural Revolution affected individual behavior and beliefs, and that
linguistic compatibility affected exposure to propaganda.
6. Conclusion
The difficulty associated with projecting state influence across space
is a recurring theme in the setting of developing nations. To this end,
mass communication technologies are frequently invoked as in­
struments which potentially augment state capacity. In this paper, we
study how mass media enabled civilian participation and compliance in
the context of political campaigns directed by the Communist Party in
China.
We utilize a previously unexplored institutional detail of a widely
studied historical movement, the Chinese Cultural Revolution, to study
the causal impact of state-sponsored propaganda on individual and
collective behavior during and after the time period. Our identification
is based on the interaction of the geographic variation in radio reception
and variation arising from the degree of intelligibility between the local
dialects and the language used in broadcasts.
Contemporaneously, we find that state radio broadcasts had positive
and significant effects on the incidences of conflict during the Cultural
Revolution. Localities where both radios were readily available and
Mandarin was reasonably well understood experienced a greater in­
tensity of conflict as proxied by the number of individuals killed.
The results suggest that linguistic diversity constrained the state’s
ability to conduct persuasion, highlighting an important tension faced
by the state between standardization of policy and effective adminis­
tration. This provides a context for why linguistic standardization is
often an implicit or explicit policy of the nation building strategy pur­
sued by the state.
In addition to investigating the contemporaneous effects of a lin­
guistic barrier to state propaganda, we also explore these long-term
consequences, by exploiting within county variation in birth cohorts
and language spoken at home. We find that in contrast to insights from
the political communication literature, the effect of radio propaganda
was persistent. Individuals who both were able to understand and were
more exposed to the radio messages were significantly more likely to
join the Communist Party later in life and hold beliefs consistent with
party ideology. This suggest that exposure to propaganda at critical
junctures in life has long-lasting implications for life-trajectories and
choices.
Our paper sheds light on the role of propaganda in building state
capacity in both the short and long-run. In the short-term, authoritarian
regimes utilize media for the successful implementation of mass cam­
paigns that would be otherwise difficult to coordinate. In the long-run,
media cultivates permanent support for the regime through the supply
and recruitment of future party members. The ability of propaganda to
influence individual behavior contemporaneously as well as through
time is central to the administrative capacity of authoritarian
governments.
Author statement
Susan Ou and Heyu Xiong: Conceptualization, Methodology, Soft­
ware Data curation, Writing – original draft preparation Visualization,
Investigation Supervision Software, Validation Writing- Reviewing and
Editing.
Data availability
The authors do not have permission to share data.
References
Adena, Maja, Enikolopov, Ruben, Petrova, Maria, Santarosa, Veronica,
Zhuravskaya, Ekaterina, 2015. Radio and the rise of the Nazis in Prewar Germany.
Q. J. Econ. 130 (4), 1885–1939.
Alesina, Alberto, La Ferrara, Eliana, 2000. Participation in heterogeneous communities.
Q. J. Econ. 115 (3), 847–904.
Alesina, Alberto, Fuchs-Schündeln, Nicola, 2007. “Good-bye Lenin (or not?): the effect of
communism on people’s preferences. Am. Econ. Rev. 97 (4), 1507–1528.
Alwin, Duane F., Krosnick, Jon A., 1991. Aging, cohorts, and the stability of
sociopolitical orientations over the life span. Am. J. Sociol. 169–195.
Appleton, Simon, Knight, John, Song, Lina, Xia, Qingjie, 2009. “The economics of
Communist Party membership: the curious case of rising numbers and wage
premium during China’s transition. J. Dev. Stud. 45 (2), 256–275.
Bai, Liang, 2014. Economic Legacies of the Cultural Revolution.
Barnes, Dayle, 1982. Nationalism and the Mandarin movement: the first half-century.
Language spread: Studies in diffusion and social change 260–290.
Baum-Snow, Nathaniel, Brandt, Loren, Henderson, J Vernon, Turner, Matthew A.,
Zhang, Qinghua, 2017. Roads, railroads, and decentralization of Chinese cities. Rev.
Econ. Stat. 99 (3), 435–448.
Bazzi, Samuel, Arya, Gaduh, Rothenberg, Alexander D., Wong, Maisy, 2017. Unity in
Diversity? Ethnicity, Migration, and Nation Building in Indonesia.

14
S. Ou and H. Xiong
Bonnier, Evelina, Poulsen, Jonas, Rogall, Thorsten, Stryjan, Miri, 2020. Preparing for
genocide: quasi-experimental evidence from Rwanda. J. Dev. Econ. 147, 102533.
Cantoni, Davide, Chen, Yuyu, Yang, David Y., Yuchtman, Noam, Zhang, Y Jane, 2017.
Curriculum and ideology. J. Polit. Econ. 125 (2), 338–392.
Chao, Yuen Ren, 1943. Languages and dialects in China. Geogr. J. 102 (2), 63–66.
Chen, Yuyu, Yang, David Y., 2019. The Impact of Media Censorship: 1984 or Brave New
World? Am. Econ. Rev.
Chiang, Chun-Fang, Knight, Brian, 2011. Media bias and influence: evidence from
newspaper endorsements. Rev. Econ. Stud., rdq037
DellaVigna, Stefano, Kaplan, Ethan, 2007. The Fox News effect: media bias and voting.
Q. J. Econ. 122 (3), 1187–1234.
DellaVigna, Stefano, Enikolopov, Ruben, Vera, Mironova, Petrova, Maria,
Zhuravskaya, Ekaterina, 2014. Cross-border media and nationalism: evidence from
Serbian radio in Croatia. Am. Econ. J. Appl. Econ. 6 (3), 103–132.
Dittmer, Lowell, 1998. Liu Shaoqi and the Chinese Cultural Revolution. Taylor & Francis.
Easterly, William, Levine, Ross, 1997. “Africa’s growth tragedy: policies and ethnic
divisions. Q. J. Econ. 1203–1250.
Enikolopov, Ruben, Petrova, Maria, Zhuravskaya, Ekaterina, 2011. Media and political
persuasion: evidence from Russia. Am. Econ. Rev. 101 (7), 3253–3285.
Escobar-Chaves, Soledad Liliana, Anderson, Craig A., 2008. Media and risky behaviors.
Future Child. 18 (1), 147–180.
Flanagan, Constance A., Sherrod, Lonnie R., 1998. Youth political development:An
introduction. J. Soc. Issues 54 (3), 447–456.
Fukuyama, Francis, 2014. Political order and political decay: From the industrial
revolution to the globalization of democracy. Macmillan.
Gerber, Alan S., Dean, Karlan, Bergan, Daniel, 2009. Does the media matter? A field
experiment measuring the effect of newspapers on voting behavior and political
opinions. Am. Econ. J. Appl. Econ. 1 (2), 35–52.
Gerber, Alan, Gimpel, James, Green, Donald, Shaw, Daron, 2011. How large and long-
lasting are the persuasive effects of televised campaign ads? Results from a
randomized field experiment. Am. Polit.. Sci. Rev. 105 (1), 135–150.
Giuliano, Paola, Spilimbergo, Antonio, 2014. Growing up in a recession. Rev. Econ. Stud.
81 (2), 787–817.
Gong, Jie, Lu, Yi, Xie, Huihua, 2014. Adolescent Adversity and Long-Run Health.
Gong, Jie, Lu, Yi, Xie, Huihua, 2015. Adolescent Environment and Noncognitive Skills.
Gupta, Apoorv, Ponticelli, Jacopo, Tesei, Andrea, 2020. Information, Technology
Adoption and Productivity: the Role of Mobile Phones in Agriculture. National
Bureau of Economic Research.
Hoffmann, Hilmar, 1996. The Triumph of Propaganda: Film and National Socialism,
19331945, vol. 1. Berghahn Books.
Houn, Franklin W., 1957. Radio broadcasting and propaganda in communist China.
Journal. Mass Commun. Q. 34 (3), 366–377.
Howse, Hugh, 1960. The use of radio in China. China Q. 2, 59–68.
Hufford, George A., 2002. The its Irregular Terrain Model, Version 1.2. 2 the Algorithm.
Institute for Telecommunication Sciences, National Telecommunications and
Information Administration, US Department of Commerce. http://flattop.its.bldrdoc.
gov/itm.html.
Jan, George P., 1967. Radio propaganda in Chinese villages. Asian Surv. 7 (5), 305–315.
Jing, Lin, 1991. The Red Guards’ Path to Violence: Political, Educational, and
Psychological Factors. Praeger Publishers.
Kasahara, Hiroyuki, Li, Bingjing, 2020. “Grain exports and the causes of China’s Great
Famine, 1959–1961: county-level evidence. J. Dev. Econ., 102513
Kinnan, Cynthia, Wang, Shing-Yi, Wang, Yongxiang, 2015. Relaxing Migration
Constraints for Rural Households.
15
Journal of Development Economics 153 (2021) 102732
Klein, Jonathan D., Brown, Jane D., Dykers, Carol, Walsh Childers, Kim, Oliveri, Janice,
Porter, Carol, 1993. “Adolescents’ risky behavior and mass media use. Pediatrics 92
(1), 24–31.
La Porta, Rafael, Lopez-de Silanes, Florencio, Shleifer, Andrei, Vishny, Robert, 1999. The
quality of government. J. Law Econ. Organ. 15 (1), 222–279.
Li, Jie, 2020. Revolutionary Echoes: radios and Loudspeakers In The Mao Era. Twentieth-
Century China 45.
Liu, Alan PL., 1964. Radio Broadcasting in Communist China. DTIC Document).
Liu, Alan PL., 1971. Communications and National Integration in Communist China.
Univ of California Press.
Markham, James W., Liu, Alan PL., 1969. “Mass communication and media in China’s
cultural revolution. Journal. Mass Commun. Q. 46 (2), 314–319.
Meng, Xin, Gregory, Robert, 2007. Exploring the Impact of Interrupted Education on
Earnings: the Educational Cost of the Chinese Cultural Revolution.
Michalopoulos, Stelios, 2012. The origins of ethnolinguistic diversity. Am. Econ. Rev.
102 (4), 1508–1539.
Nunn, Nathan, Puga, Diego, 2012. Ruggedness: the blessing of bad geography in Africa.
Rev. Econ. Stat. 94 (1), 20–36.
Olken, Benjamin A., 2009. Do television and radio destroy social capital? Evidence from
Indonesian villages. Am. Econ. J. Appl. Econ. 1 (4), 1–33.
Phillips, Caleb, Douglas, Sicker, Grunwald, Dirk, 2011. The Stability of the Longley-Rice
Irregular Terrain Model for Typical Problems. arXiv preprint arXiv:1109.1843.
Pye, Lucian W., 2015. Communications and Political Development.(SPD-1). Princeton
University Press.
Qin, Bei, Strömberg, David, Wu, Yanhui, 2017. Why does China allow freer social media?
Protests versus surveillance and propaganda. J. Econ. Perspect. 31 (1), 117–140.
Qin, Bei, Strömberg, David, Wu, Yanhui, 2018. Media bias in China. Am. Econ. Rev. 108
(9), 2442–2476.
Ramsey, S Robert, 1987. The Languages of China. Princeton University Press.
Schoenhals, Michael, MacFarquhar, Roderick, 2006. Mao’s Last Revolution. Belknap
Press of Harvard University Press.
Su, Yang, 2011. Collective Killings in Rural China during the Cultural Revolution.
Cambridge University Press.
Tang, Chaoju, Van Heuven, Vincent J., 2009. Mutual intelligibility of Chinese dialects
experimentally tested. Lingua 119 (5), 709–732.
Voigtländer, Nico, Voth, Hans-Joachim, 2014. Highway to Hitler. National Bureau of
Economic Research.
Voigtländer, Nico, Voth, Hans-Joachim, 2015. Nazi indoctrination and anti-Semitic
beliefs in Germany. Proc. Natl. Acad. Sci. Unit. States Am. 112 (26), 7931–7936.
Volland, Nicolai, 2003. The Control of the Media in the People’s Republic of China (PhD
diss).
Walder, Andrew G., 2014. Rebellion and repression in China, 1966–1971. Soc. Sci. Hist.
38 (3–4), 513–539.
Walder, Andrew G., Su, Yang, 2003. The cultural revolution in the countryside: scope,
timing and human impact. China Q. 173, 74–99.
Principal Investigator Walder, Andrew G., 2010-2013. Political Movements in an
Authoritarian Hierarchy. National Sciencohortinvariantce Foundation Grant SBS-
1021134.
Xu, Lanjun, 2014. Translation and internationalism. In: Cook, Alexander (Ed.), Mao’s
Little Red Book: A Global History. Cambridge University Press.
Yanagizawa-Drott, David, 2014. Propaganda and conflict: evidence from the Rwandan
genocide. Q. J. Econ. 129 (4), 1947–1994.
Zhou, Weina, 2013. How Does a Traumatic Experience during Youth Affect Life Later?
the Long-Term Impact of the Send-Down Program during the Chinese Cultural
Revolution. Working Paper).