Hate is in the air: The effect of Czech and German radio on elections in pre-war Czechoslovakia (Paper)

Master Project

Hate is in the air: The effect of the Czechand German radio on the elections in

prewar Czechoslovakia

SUMMER FORUMB A R C E L O N A G S E

PROGRAM | JUNE 10-28, 2013

SEVERO OCHOA RESEARCH EXCELLENCE

In 2011, Barcelona GSE wasdistinguished as one of eight topresearch institutions of excellencewith an international impact inthe framework of the SeveroOchoa Program, sponsored by theSpanish Ministry of Economy andCompetitiveness.

Bruno Baranek, Krystof Krotil & Samuel Skoda

MSc. in Economics

2013/2014

Hate is in the air: The effect of the Czechand German radio on the elections in

prewar Czechoslovakia

Bruno Baranek, Krystof Krotil & Samuel Skoda

Abstract

In this paper we assess the role of radio broadcasting in parliamentaryelections of 1935 in Czechoslovakia. In our main specification, we regress thevote shares of multiple parties on the signal strengths of Czech and Germanradio while controlling for demographic and socio-economic characteristics. Inparticular, we focus on SdP – the ethnic German party with separatist ten-dencies, which was supported by Hitler’s NSDAP. We find that propagandacontained in the German broadcasts had a polarizing effect on the Czechoslo-vak political spectrum as it increased the number of votes for SdP and alsofor Czech communists and nationalists. This increase was compensated bythe fall in votes for centrist democratic parties. On the other hand, the Czechradio, which was politically neutral, tended to neutralize the effect of theGerman radio.

Keywords election, voting, media, radio broadcasting, political economy

JEL Classifications D72, L82, P16

Acknowledgements We remain greatly indebted to Ruben Enikolopov and MariaPetrova for their help with the computation of radio signal strengths and also fortheir valuable comments. We also want to express our gratitude to Pavel Hajek fromthe Czech Statistical Office for recommending relevant data sources, to Jan Krızfor helping with data collection and to Antonın Petrzelka for providing accurateinformation on radio transmitters.

1

1 Introduction

A large part of Czechoslovakia was annexed by Germany on the eve of the Second

World War in 1938. It is no doubt that this process was enhanced by increasing na-

tionalistic attitudes of a significant German minority. In 1935, the German national-

istic party SdP surprisingly won the Czechoslovak parliamentary elections, receiving

15.18 percent of votes. Later on, this party collaborated closely with Hitler’s NSDAP

and was instrumental to the fall of democratic Czechoslovakia. Historians provide

several explanations for the increased extremism among German citizens; perhaps

the most common is the worsening of economic conditions and high unemployment

in areas settled by the Germans as a bleak aftermath of the Great Depression. In

our paper we focus on an alternative explanation of the electoral triumph for SdP

– influence of Nazi propaganda from Germany. Extensive evidence from political

economy has already showed that media can have an enormous influence on political

beliefs and subsequently on electoral outcomes. In our paper we analyze the effects

of two different radios – Czech and German – on the support for the major politi-

cal parties in Czechoslovakia. Bilingualism was fairly widespread both between the

ethnic Germans and Czechs living in Czechoslovakia and therefore the overall pop-

ulation was at least to some extent able to listen to both stations. Czech radio was

often criticized for apolitical, dry and boring broadcasting, most importantly there

was no counterpropaganda or anti-German content. On the other hand, the German

radio was part of the sophisticated propaganda endorsed by Joseph Goebbels. In

some areas one could listen to both radios, in others only to one of them. This vari-

ation allows us to study whether these media had an influence on electoral outcomes

and to separate their effect.

We find a significant effect for both radios. The German radio seems to polarize

the voters by increasing the votes for the German nationalistic SdP but also for

the Czech nationalists NarSj and the communists. The increase in votes for these

extremist parties is compensated by the decrease in votes for mainstream centrist

parties. On the other hand, the Czech radio seems to lower the social tension

by decreasing the votes for SdP. These results show that a heavily biased radio

persuades a number of listeners but also provokes anger among those who oppose

its content. Importantly, we show that an apolitical radio, even with minimum

content of news, can significantly reduce extremism. Our results support the view

that media plays an important role in democratic institutions and non-biased media

might be critical for hindering extremism.

2

The rest of the paper is organized as follows. Section 2 presents the related literature,

while Section 3 provides short historical background. Section 4 describes data.

Section 5 discusses the empirical strategy as well as some identification challanges.

Section 6 presents the empirical results. Section 7 concludes.

2 Related literature

We study the effects of two prewar radios: the German radio that was heavily pro-

Nazi biased and the Czech radio that was notably apolitical. Below we present the

literature related to our research. Firstly, our study extends the literature studying

the connections between media and electoral outcomes. For theoretical models about

the interaction of media and governments in both democratic and non-democratic

countries see Besley and Prat (2006), Egorov et al. (2009) or Gehlbach and Sonin

(2013). Furthermore we will focus on empirical studies of media’s effect. Numerous

papers in the past have found that media might significantly influence political

beliefs and more importantly electoral results. The following studies use variation

in access to media in order to study outcomes relevant for public policy: DellaVigna

and Kaplan (2007), Stromberg (2004), Reinikka and Svensson (2004).

For our analysis we exploit variation in radio signal by utilizing the Irregular Terrain

Model (ITM). In particular we use historical data about height, position and power

of Czechoslovak and German transmitters to calculate signal strengths of the Czech

and German radio in Czechoslovakia. This approach was first introduced by Huf-

ford (2002). As Phillips et al. (2011) write in their overview of this method, “The

model predicts the median attenuation of the radio signal as a function of distance

and additional losses due to refractions at intermediate (terrain) obstacles.” More

broadly speaking, the Irregular Terrain Model is becoming increasingly popular in

studies in political economy. Olken (2009) use variation in television signal to study

the effect of watching TV on social behavior. The study by Adena et al. (2013)

in turn focuses on a question very similar to ours: they estimate the effect of Nazi

propaganda on votes for NSDAP in several elections in Germany before WWII. The

authors find a significant positive effect of listenership to radio in times when the

radio became politically biased with the aim of supporting the Nazi party. The

study shows that propaganda also helped to advance other nationalistic behavior

as well as animosity against Jews. Other papers have examined the effects of me-

dia using ITM in more recent times. DellaVigna and Kaplan (2007) show that the

introduction of Fox TV positively affected the vote share of the Republican Party.

3

Enikolopov et al. (2011) analyze how the only independent TV in Russia reduced

the vote share for the government party having control over all other media. An-

other study concentrating on biased media is Chiang and Knight (2011). Finally,

also media without bias and political content at all can affect political preferences

(Durante and Knight, 2012).

Our paper is also closely related to DellaVigna et al. (2014, forthcoming), who study

the cross border effects of Serbian radio on Croats who listen to it in Croatia mainly

for other than political reasons. They find that listening to Serbian radio provokes

extremist behavior among the Croats. The bilingual composition of Czechoslovakia,

where people could listen to both Czech and German radio, allows us to study

whether there is a similar effect. Relatedly, Gentzkow and Shapiro (2004) shows a

direct effect of media on extremism and nationalism among Muslims.

3 Background

3.1 The political system in prewar Czechoslovakia

As a result of the Austro-Hungarian empire’s collapse caused by the First World

War, Czechoslovakia declared independence on October 28th, 1918. Composed of

several regions of Austria-Hungary, it also blended together a number of distinct

ethnic groups. Of the total population of 13.4 million, the Czechs and Slovaks1

formed 66%, Germans 23%, Hungarians 6% and Rusyns 4%. In the Czech regions

of the country the share of German population was even higher and reached almost

a third of the local population. The formation of Czechoslovakia in this extent hap-

pened in spite of the opposition from German population that found itself within the

boundaries of the newly established Czechoslovakia and that soon after Czechoslovak

declaration of independence demanded to incorporate German-dominated provinces

into Austria. The German and Hungarian defiance was however quickly suppressed

after the breakaway regions were seized by the Czechoslovak army. Furthermore,

these de facto borders were confirmed by the Treaty of Versailles in 1919 and its

complementary peace agreements with the neighboring countries Austria and Hun-

gary.

The new constitution was adopted on February 29th, 1920 and it defined Czechoslo-

vakia as a republic and democracy, guaranteed universal adult suffrage (for men and

1The 1921 census used the category “Czechoslovaks” and therefore did not distinguish betweenthe two ethnicities.

4

women alike) and also warranted relatively wide minority rights and freedom of reli-

gion. The legislative power was assigned to a bi-cameral parliament divided into the

Chamber of Deputies (consisted of 300 members elected for 6 years) and the Senate

(150 members elected for 8 years). The executive power was shared by cabinet and a

president. Unlike the Austrian parliamentary system that was in place before 1918,

Czechoslovakia implemented a system of proportional representation that allowed

even relatively small parties to gain seats in the parliament.

Moreover, the parliamentary system of Czechoslovakia allowed for the coexistence

of a wide variety of parties, traditionally divided not only on the ideological basis

but also on ethnic basis. During the existence of the First Republic (1918–1938),

the ruling government coalitions were characteristic by their breadth, i.e. they were

often formed from five or more parties. Among the most important ones were the

the following:

• The Agrarian Party (Republikanska strana zemedelskeho a malorolnickeho lidu,

RSZML) was primarily focused on people from the more rural parts of the

country. The Agrarians had one of the broadest network of local support.

• Similarly powerful were the Social Democrats (Ceskoslovenska socialne demo-

kraticka strana delnicka, CSDSD) that were oriented towards the working class

and the middle class city dwellers. The party’s influence decreased for some

time after the radical fraction of the party seceded and formed the Communist

party of Czechoslovakia in 1921 (Komunisticka strana Ceskoslovenska, KSC).

• The Czechoslovak People’s Party (Ceskoslovenska strana lidova, CSL) was

composed of several Christian movements present in Bohemia and especially

in Moravia that was traditionally Catholic.

• The National Democrats (Ceskoslovenska narodnı demokracie, CSND) were

nationalist oriented and economically liberal. The party strongly opposed

any form of socialism and was secular. National Democrats promoted Czech

nationalism as a uniting element. In 1934 the party merged with two minor

nationalist parties and ran in the elections as the National Unity (Narodnı

Sjednocenı, NarSj).

German parties started participating in the democratic system despite the hostility

to the Czechoslovak state that followed the declaration of independence in Octo-

ber 1918. This change of attitude is often attributed to quick implementation of

laws that guaranteed the use of minority languages in education and administration

for counties with at least 20% of the minority population, which effectively cov-

5

ered more than 90% of the overall German population. In Figure 1 we present a

map of all counties according the shares of German population. Furthermore, the

forceful expropriation of assets was forbidden and sanctioned by the new constitu-

tion. Another stabilizing element was the fact that Czechoslovakia skillfully avoided

hyperinflation and political chaos after the end of the First World War unlike the

neighboring countries Austria, Germany and Hungary. All these facts contributed

to the increased engagement of ethnic Germans in the democratic system of prewar

Czechoslovakia.

Of course not all German parties were created equal. Historians traditionally divide

German parties into two main groups: activist and negativist. The activist parties

accepted the creation of Czechoslovakia with the inclusion of German minorities,

while the two negativist parties were anti-system and demanded the right of self-

determination for Germans in Czechoslovakia. The activist parties were essentially

counterparts to the Czechoslovak parties. The most important German parties

include the following:

• Deutsche sozialdemokratische Arbeiterpartei (DSAP) was the German equiv-

alent to the CSDSD. Especially since 1926 both parties cooperated closely.

In the first parliamentary election in 1920 German Social Democrats gained

the highest share of votes from all German parties (11.1%), however its sup-

port was lower after a communist fraction seceded from the party. The party

opposed German nationalism and was strongly pro-democratic.

• Bund der Landwirte (BdL) was the sister party of the Agrarian party. Its sup-

port stemmed from the ethnic German countryside that was largely comprised

of small farmers.

• The two negativist parties were the German National Party (Deutsche Nation-

alpartei, DNP), which was a conservative nationalist party, and the German

National Socialist Workers’ Party (Deutsche Nationalsozialistische Arbeiter-

partei, DNSAP), a proto-fascist party modeled on Hitler’s NSDAP that oper-

ated in Germany. In the elections of 1920, these two parties formed a coalition

that received 5.3% of all votes. Both parties were banned in 1933 and most of

their members later participated in SdP (described in the next subsection).

6

3.2 The rise of extremism

Until 1926 none of the German parties was a part of the government coalition,

though the division into activist and negativist parties was already established. In

1926, based on the election results from the previous year, the main center-right

parties (i.e. Agrarian Party, Czechoslovak People’s Party, Czechoslovak National

Democratic Party) together with their three German counterparts formed a new

government. This coalition lasted until the new elections in 1929, when the support

for the negativist parties declined to 4%. The new coalition included both center-left

and center-right parties, including German ones.

The Great Depression was a major blow to the Czechoslovak economy. The recession

lasted until the spring of 1934 and the economy did not return to pre-depression

production levels until 1938. The most affected industries were textile, glassware

and porcelain production that relied heavily on exports and were located mainly in

ethnic German areas. The number of officially unemployed increased almost by a

factor of five over a single year. In 1934, some of the German counties (e.g. Kraslice,

Stemberk or Rumburk) had an unemployment rate around 30%.

With deteriorating economic conditions and the rise of Nazis over western borders,

the Czechoslovak political landscape was also transforming. In 1933, DNSAP openly

expressed sympathies to Hitler’s rise to power in Germany and DNP was following

the line of DNSAP. As a result, both parties were outlawed in October 1933. Most

of the members of these two banned parties joined the newly established Sude-

tendeutsche Heimatsfront, a German nationalist movement, that tried to unite all

Czechoslovak Germans under one banner. Originally without a clear political agenda

to avoid the fate of DNSAP and DNP, the movement was before the elections in

1935 transformed into Sudeten German Party (Sudetendeutsche Partei, SdP) and

became critical of the mainstream Czechoslovak political parties. Despite the fact

that the party received funding and advice from Nazi Germany, its leader Konrad

Henlein, formerly a leader of a German gymnastics organization, was reluctant to

support the Nazis overtly.

In the elections of 1935 the SdP obtained a vote share of 15.18%, more than any

other party. The remaining German parties suffered a significant loss in the elec-

tions. Although SdP did not become a part of the grand government coalition, it

did however begin to gradually increase its separatist demands. It also became ag-

gressively attacking its rivals from other German parties. In September 1938, just

before the Munich crisis, SdP in cooperation with SS units unsuccessfully attempted

7

a violent coup. Two weeks later, on September 29th, the Munich agreement forc-

ing Czechoslovakia to hand the Sudetenland over to Germany was signed. After

that SdP was formally incorporated into NSDAP. The rest of the country was in-

vaded in March 1939 and the remaining Czech regions became occupied by Nazi

Germany.

3.3 Radio content and availability in Czechoslovakia

3.3.1 The Czech radio

The first regular radio broadcasting in Czechoslovakia started in 1923, around the

same time as in Germany. Broadcasting was provided by a private company called

Radiojournal spol. s r.o. originally owned by a local producer of radio receivers.

Radiojournal obtained a license from the Ministry of Posts and relied mostly on

revenues from selling listening subscriptions. Originally most of the program was

devoted to classical music, culture or educational programs, which were distinctly

apolitical. The number of subscribed listeners reached 570 000 in 1933 and the

millionth listener was recorded in 1936. The first transmitter in the Czech part

of the republic was set up in Prague, Brno followed in 1925 and Ostrava in 1929.

In the same year the transmitter in Brno was upgraded to a more powerful one.

The transmitter in Prague was since its establishment improved twice, until a new

120kW transmitter was built in 1931 on the outskirts of Prague. At the time it was

one of the most powerful transmitters in Europe.

News were originally devoted only to weather, sports or stock markets. Especially

the broadcasting of stock market news became popular among entrepreneurs and

was eventually extended to five transmissions a day. In 1925, Radiojournal became

primarily owned by the state with a 51% share. The following year it made an

agreement with the state-owned Czechoslovak Press Agency (Ceskoslovenska tiskova

kancelar, CTK) over the broadcasting of news that became solely under the control of

the CTK. Radiojournal had no entitlement over content produced by CTK and could

affect only culture programs and press summaries. Also a new set of programs for

more specific audiences emerged: agricultural as well as workers’ and industrialists’

programs were set up to educate listeners and advertise new job offers. Despite the

fact that Radiojournal tried to maintain its content strictly apolitical, these program

blocks sometimes included lectures by speakers that could be often associated with

a specific political party. However, control over content played a significant role in

the case of lectures as all speeches had to be approved by police at least 48 hours

8

before broadcasting and in case of any changes, the radio was obliged to stop the

live broadcast.

The news aired in Radiojournal were remarkable for their similarity to press news,

which was a result of the CTK’s involvement. Of all the broadcasted content, only

2% of the whole airing time was attributed to news, excluding sport and weather

service. The news were sporadically criticized for their dryness, incomprehensibility

and lack of support for pro-democratic and pro-government stances. Combined with

censorship that did not allow the expression of political opinions in other programs,

broadcasting during the studied period, i.e. before the 1935 elections, can be clearly

described as neutral.

Broadcasting of the Czechoslovak radio in German language started in 1925. Every

day, 30 minutes of the broadcasting airtime was dedicated to German programs. In

1933 this was extended to 60 minutes a day. Transmissions in German were mainly

focused on agricultural, workers’ and educational topics. A significant amount of

time was devoted to regional educational programs that presented information about

the history and folklore of different Bohemian regions. Outside of the 60 minutes

window, there were sometimes also live shows broadcasted from the German the-

atre in Prague. Despite the fact that the creation of an exclusively German radio

was discussed already in the late 20s, this plan was not realized until 1937, when

Radiojournal started broadcasting in German language from Melnık.

3.3.2 The German radio

With the rapid advance of radio in Germany and transmitters becoming more

and more powerful, German broadcasting became available also in some regions

of Czechoslovakia. In particular, this overlap was generated by transmitters near

Lepzig, Breslau and Munchen. The distribution of transmitters across Germany was

decided in a manner that aimed to cover the largest share of population in Germany.

Radio availability and content of the German broadcasting is described in detail by

Adena et al. (2013). For our purposes it is sufficient to characterize the prewar era

in three periods. Before 1929 the radio was mainly apolitical and partisan content

was explicitly forbidden. However, there was a change in 1929, when the German

radio started to transform, eventually offering more political content. Especially

with the growing popularity of the Nazi Party, more airing time was devoted to po-

litical and economic news and pro-democratic speeches. In fact, during the German

election campaigns in 1930 and 1932 all major parties except the Nazi Party and

9

the Communist Party were offered airing time. As was the radio becoming more

centralized, the government also used a notable share of broadcasting time to sup-

port government policies in regular radio speeches by ministers. According to Ross

(2006), the radio content was negatively skewed against the Nazi Party.

In January 1933, Hitler was appointed the chancellor of Germany and despite the

fact that Nazis had before no control over radio broadcasting, the situation was com-

pletely reversed in a matter of weeks. Nazis restricted the access to radio for other

parties and soon started to broadcast from rallies and demonstrations to support

the image of a mass Nazi movement present in Germany. Nazis themselves strongly

believed in radio as a very effective tool of propaganda and therefore broadcast-

ing remained sharply pro-Nazi throughout the entire time of the dictatorship in

Germany.

4 Data

Below we describe variables used in our analysis that we collected from a number

of different sources. For the purpose of easier presentation we adopt jargon of

randomized experiments. Summary statistics are reported in Table 1.

4.1 Treatment variables

Our treatment variable, radio signal strength, is computed using the Irregular

Terrain Model (see Hufford (2002), also used by Olken (2009), DellaVigna et al.

(2014, forthcoming) and others). Data on Czech transmitters are based on infor-

mation from two publications on history of the Czechoslovak radio: Prvnıch de-

set let Ceskoslovenskeho rozhlasu by A. J. Pacakova and Profesionalizace vysılanı

1930–1938 by Jirı Hrase. Corresponding data on German transmitters, i.e. their

location, height, power and frequency come from Adena et al. (2013). For all com-

putations we utilize the division of Czechoslovakia into judicial counties (soudnı

okresy) and obtain the predicted signal strength for the geographical location of the

main town of each county. Our analysis is restricted to the 328 “Czech” counties

of Czechoslovakia, i.e. we do not look at Slovakia and Subcarpathian Rus’, which

were distinct administrative units with different ethnic and political conditions. In

Figure 2 we present a map of the German signal strength divided into 6 quantiles.

Figure 3 is an analogous map for the Czech signal strength.

10

4.2 Outcome variables

Vote shares for a number of parties that participated in the 1935 parliamentary

elections were used as an outcome variable of our interest. We also exploit electoral

data from three previous parliamentary elections in 1920, 1925 and 1929. All his-

torical Czechoslovak electoral results on the level of judicial counties are obtained

from Volby do zakonodarnych organu Ceskoslovenska a Ceske republiky 1920 - 2006

published by the Czech Statistical Office in 2008. We do not compute turnout rates

as voting in the elections was compulsory by law. For illustration we mention that

the overall turnout rate in 1935 was approximately 92%.

4.3 Control variables

Demographic controls come from the national census that took place on Decem-

ber 1st, 1930. We process data from the official census publication of the Czechoslo-

vak Statistical Office. In particular we collect data on ethnicity, religion, age and

gender that were also reported on the level of judicial counties. We also create a

dummy variable for counties with German majority (i.e. Sudetenland) and another

dummy for counties that have a town with population over 10 000 inhabitants.

Furthermore, we employ two additional sources of data for economic controls.

First, we compute the unemployment rate from the number of the registered un-

employed in 1933 obtained from the publication Nezamestnanost a podpurna pece

v Ceskoslovensku by Jaromır Necas. Second, we process data on income per capita

and the number of eligible taxpayers per 1000 citizens in 1933 from Statistika dane

duchodove placene prımo, dane z vyssıho sluzneho, dane rentove placene prımo,

vseobecne a zvlastnı dane vydelkove podle predpisu za rok 1933 published by the

Czechoslovak Statistical Office. All three above mentioned variables are reported on

the level of political counties (politicke okresy). Political counties were at a higher

level of aggregation than judicial counties; usually a political county contained be-

tween 1 and 4 judicial counties. In total, we utilize 150 political counties.

5 Empirical strategy

In this section we present our empirical model and explain the main identification

assumption. We will estimate the effect of exposure to radio on electoral outcomes

11

in parliamentary votes in 1935 using the following specification:

vote shareij = β0j + β1jCzech signali + β2jGerman signali + β3jXi + εij

where vote shareij is vote share of party j in judicial county i, Czech signali and

German signali are measures of radio signal strength in county i, Xi is a set of

county-specific controls and εij is an idiosyncratic error. Since we have the economic

control variables only on the level of political counties we cluster the errors on the

level of political districts.

We analyze the vote shares of the four largest parties according to the 1935 elections

(i.e. SdP, RSZML, CSDSD, KSC) and we also include NarSj as a representative party

of the Czech nationalists and DSAP as the second most influential German party

in Czechoslovakia. Our first crucial assumption is that signal strength is a relevant

measure for real exposure to radio broadcasting, i.e. radio signal is correlated with

radio listenership so that variation in radio signal results in variation in exposure

to this media. This assumption seems intuitive and its plausibility has indeed been

confirmed by Adena et al. (2013) and the others. Moreover, significant results in

the latter parts of our paper speak in favor of this assumption. Unfortunately, we

do not have available data on listenership at other than the national level, so we

cannot test this assumption nor instrument listenership by radio signal. Therefore,

our results need to be interpreted as intention to treat effects.

Our main identifying assumption is that signal strength is not correlated with the

error term conditional on controlling for numerous observable characteristics of ju-

dicial counties. First, the signal of the German radio was generated mainly by three

big transmitters located in notable distance from the borders of Czechoslovakia.

Neither their position nor the historical evidence suggest that the German minority

in Czechoslovakia was taken into account at the time they were built. Moreover,

the transmitters were built before the radio had adopted any pro-Nazi content. The

Czech transmitters were located in the three biggest cities in terms of population.

Taking into account these facts we can rule out reverse causality, i.e. that radio

transmitters were built with respect to political beliefs of specific regions. In or-

der to study the exogeneity of the radio signals we present the regressions of radio

signals in 1935 on our control variables reflecting urbanization, demographic and

socio-economic conditions as well as older electoral outcomes standing as a proxy

for political beliefs within a county. For these regressions we chose electoral data

from 1925 because radio was very rare at this time as opposed to the next elections

in 1929 when radio was already becoming more popular. We see that radio signal is

12

correlated with some of our control variables. Importantly, while the signal of the

German radio is correlated only with vote share of CSNS (a small party), the signal

of the Czech radio is correlated with vote shares of several parties.

This represents a threat to our identification strategy since it means that there

might be correlation between signal strength and unobservable characteristics of

counties that determine political outcomes. If the correlation exists, this would

bias our results. To alleviate these concerns we perform the Altonji-Elder-Taber

test as conducted in Adena et al. (2013). In the first stage we separately predict

signal strength both for the Czech and the German radio in each county, taking the

fitted values from the regressions of signal strength on the full set of controls from

Table 2 with the exception of ethnicity variables. Then, we regress vote shares of

political parties in 1935 on these fitted values while controlling for ethnicities and

report the results in Table 3. Most importantly, we find that the fitted values for

both the Czech and German radio signal strength are unrelated with the vote share

of SdP. These results hence suggest that under the assumption that unobservables

are correlated with the index of observables, our estimates of the effects of radio

on the support for SdP are not likely to be biased in any direction. If we further

assume that the correlation is positive, the results are likely to be biased against

the findings we report in our main specification for CSDSD and DSAP in the case

of German radio. On the other hand the test shows that some of the effect of the

German radio on KSC, NarSj and RSZML might be found only by the inconsistency

of the estimators. When considering the Czech radio, one can see that we might be

overestimating the effect on vote share for KSC but also that the coefficients could

be biased toward zero for DSAP and NarSj.

6 Results

The results in Table 4 support the main hypothesis: we find a significant positive

effect of German radio availability on the vote share of SdP, a party most closely

associated with NSDAP. In our full specification the results indicate that an increase

in signal strength by one standard deviation yields additional 1.1 percentage point

for SdP. For comparison, the magnitude of the effect of exposure to Nazi radio is

even larger than the effect found in Adena et al. (2013), which found an increase by

0.5 percentage points for NSDAP.

Furthermore, we also find a significant negative effect of Czech radio on the vote

share of SdP. An increase by one standard deviation leads to a decrease in the vote

13

share of SdP by 1.5 percentage points. This result suggests that being exposed to

politically neutral media decreases the support for extremist parties and by exten-

sion gives prominence to the importance of democratic media. We would like to

emphasize here that the impact is sizable even in a case of a radio that provided

only limited time for news and potential politically sensitive content.

We also study the effect of radio for a number of other parties than SdP. Listening

to German radio works as a double-edged sword as it increases support for a party

with a clearly anti-German agenda (NarSj) and for the communists that formed

an anti-system party. These findings are consistent with the results of DellaVigna

and Kaplan (2007), where the authors also find an increase in nationalistic behavior

caused by listening to a radio of an antagonized nation. The magnitudes of these

effects are much smaller than for SdP; an increase by one standard deviation in signal

increases vote share by 0.2 and 0.3 percentage points for NarSj and KSC respectively.

On the other hand, more mainstream parties (Social Democrats and Agrarians) were

negatively affected. We conclude that the Nazi radio had a polarizing effect: a large

positive effect for SdP, a substantially smaller but also positive effect for anti-German

and far-left parties, nonetheless an effect of the similar size but of the opposite sign

for more centrist parties.

Next, we turn our attention on the effect of the politically neutral Czech radio

on other parties. Most notably we observe that listening to the Czech radio has

a positive significant effect on the vote share of pro-democracy, pro-government

DSAP, who were the main rivals of SdP. The effect is highly significant and in

standardized terms translates into a 0.6 percentage points increase in the DSAP

vote share. Another interesting observation is that the vote share of NarSj, an anti-

German right-wing party, is negatively affected by the Czech radio (signal improved

by one standard deviation decreases support by 0.4 percentage points). However

the effect on KSC, a communist party, is significant and positive (an increase by 0.3

percentage points).

In Table 5 we report how coefficients on radio signal strengths react when controls are

gradually added to the regression in the case of SdP. While the coefficient estimates

are sensitive to the inclusion of the ethnicity controls, the addition of the other

controls generally increases the magnitude of the coefficients and increases precision.

For the rest of the parties, the coefficients are similarly robust to the inclusion of

other covariates, except voting controls. Finally, we conduct placebo tests where

we run the specification for 1935 on electoral outcomes of 1920 (when radio did

not exist) and 1929 (when both radios were apolitical). The results are reported in

14

Table 6 and Table 7. While we do not find any association between the German

radio availability in 1935 and these outcomes, we find many significant effects in the

case of the Czech radio.

Throughout the paper we have discussed the endogeneity of radio signal. The

Altonji-Elder-Taber test suggested that some of our results might be biased for

our hypothesis and some against it. When further examining this problem with

placebo tests we find that the signal of the German radio in 1935 does not help in

explaining electoral outcomes in the previous years. However, the Czech radio signal

is correlated with past vote shares. Also the approach in which we gradually added

control variables to the specification to show robustness suggests that the German

radio can be considered exogenous in our analysis. On the other hand there may be

serious concerns about the exogeneity of the Czech radio. This is probably caused

by the location of transmitters in the three largest metropolitan areas. We have to

take this into account when interpreting results.

7 Conclusion

Our paper demonstrates the importance of media, radio in particular, in the elec-

tion during the eve of World War II in Czechoslovakia. In the election in 1935,

the propaganda contained in the German broadcast had a polarizing effect on the

Czechoslovak political spectrum, as it increased the number of votes for the German

Nazis, Czech nationalists and Czech communists. This increase was compensated

by the fall in votes for centrist democratic parties. On the other hand, the Czech

radio, which was politically neutral, tended to offset the effect of the German ra-

dio. However, in the case of the Czech radio we cannot entirely reject the potential

endogeneity that could be driving our results.

This study confirms that media has an important and significant effect on electoral

outcomes. Moreover, the work provides a unique comparison of effects of two radios

with a completely different contents – one heavily politically biased and one without

bias and with minimal political content. These findings underline the crucial role of

media in the process of building a democracy and also during political coups. We also

provide empirical evidence for the hypothesis that the German radio propaganda

played an important role in the fall of democracy in prewar Czechoslovakia.

15

References

Adena, M., R. Enikolopov, M. Petrova, V. Santarosa, and E. Zhu-

ravskaya (2013): “Radio and the rise of Nazis in pre-war Germany,” Working

paper.

Besley, T. and A. Prat (2006): “Handcuffs for the grabbing hand? Media

capture and government accountability,” The American Economic Review, 720–

736.

Chiang, C.-F. and B. Knight (2011): “Media bias and influence: Evidence from

newspaper endorsements,” The Review of Economic Studies, 78, 795–820.

DellaVigna, S., R. Enikolopov, V. Mironova, M. Petrova, and E. Zhu-

ravskaya (2014, forthcoming): “Cross-border media and nationalism: Evidence

from Serbian radio in Croatia,” American Economic Journals: Applied Eco-

nomics.

DellaVigna, S. and E. Kaplan (2007): “The Fox News effect: Media bias and

voting,” The Quarterly Journal of Economics, 122, 1187–1234.

Durante, R. and B. Knight (2012): “Partisan control, media bias, and viewer

responses: Evidence from Berlusconi’s Italy,” Journal of the European Economic

Association, 10, 451–481.

Egorov, G., S. Guriev, and K. Sonin (2009): “Why resource-poor dictators

allow freer media: A theory and evidence from panel data,” American Political

Science Review, 103, 645–668.

Enikolopov, R., M. Petrova, and E. Zhuravskaya (2011): “Media and

political persuasion: Evidence from Russia,” The American Economic Review,

101, 3253–3285.

Gehlbach, S. and K. Sonin (2013): “Government control of the media,” Avail-

able at SSRN 1315882.

Gentzkow, M. A. and J. M. Shapiro (2004): “Media, education and anti-

Americanism in the Muslim world,” The Journal of Economic Perspectives, 18,

117–133.

Hufford, G. A. (2002): “The ITS Irregular Terrain Model, Version 1.2. 2 the

Algorithm,” Institute for Telecommunication Sciences, National Telecommunica-

tions and Information Administration, US Department of Commerce, available at

http://flattop.its.bldrdoc.gov/itm.html.

16

Olken, B. A. (2009): “Do television and radio destroy social capital? Evidence

from Indonesian villages,” American Economic Journal: Applied Economics, 1,

1–33.

Phillips, C., D. Sicker, and D. Grunwald (2011): “The Stability of The

Longley-Rice Irregular Terrain Model for Typical Problems,” CoRR.

Reinikka, R. and J. Svensson (2004): “Local capture: Evidence from a central

government transfer program in Uganda,” The Quarterly Journal of Economics,

119, 679–705.

Ross, C. (2006): “Mass politics and the techniques of leadership: The promise and

perils of propaganda in Weimar Germany,” German History, 24, 184–211.

Stromberg, D. (2004): “Radio’s impact on public spending,” The Quarterly Jour-

nal of Economics, 119, 189–221.

17

Appendix

Figure 1: Share of German population

100% 70% 50% 30% 20% 10% 0%

Figure 2: German signal (divided to 6 quantiles according to strength)

best signal worst signal

Figure 3: Czech signal (divided to 6 quantiles according to strength)

best signal worst signal

18

Table 1: Summary statisticsVariable Mean Std. Dev. Min. Max. NCzech signal strength 17.25 10.645 -8.048 53.783 328German signal strength 10.112 6.006 -2.245 34.8 328Czechs & Slovaks share 0.632 0.407 0.01 0.996 328Germans share 0.348 0.402 0 0.984 328Jews share 0.001 0.003 0 0.021 328Other ethnicity share 0.007 0.046 0 0.646 328Catholicism share 0.835 0.145 0.256 0.994 328Protestant share 0.04 0.064 0 0.567 328Hussite church share 0.059 0.081 0 0.454 328Judaism share 0.006 0.007 0 0.07 328Other faith share 0.005 0.013 0 0.145 328Atheism share 0.056 0.069 0.001 0.366 32825-44 age share 0.293 0.026 0.159 0.414 32845-64 age share 0.194 0.017 0.091 0.27 32865 and higher age share 0.076 0.015 0.028 0.113 328Adult female share 0.53 0.019 0.357 0.587 328Population density (people per km2) 178.089 478.37 8.726 8007.671 328Population size (thousands) 32.544 54.74 5.724 936.608 328Log of income per capita (Kc) 6.756 0.381 5.83 8.627 328Taxpayers per 1000 people 55.968 17.686 23.3 128.8 328Unemployment rate 0.062 0.048 0.005 0.27 328Sudetenland dummy 0.47 0.5 0 1 328Major towns dummy 0.213 0.41 0 1 328CSDSD 0.115 0.089 0.001 0.438 328DSAP 0.052 0.074 0 0.321 328KSC 0.072 0.059 0.001 0.344 328NarSj 0.036 0.038 0 0.26 328RSZML 0.161 0.141 0.001 0.546 328SdP 0.221 0.263 0 0.782 328CSDSD1929 0.123 0.107 0 0.503 328CSND1929 0.032 0.037 0 0.279 328DSAP1929 0.096 0.133 0 0.516 328KSC1929 0.086 0.069 0 0.355 328RSZML1929 0.146 0.143 0 0.544 328SdP1929 0.092 0.113 0 0.515 284BdL1925 0.11 0.157 0 0.679 328CSDSD1925 0.09 0.088 0.001 0.457 328CSL1925 0.129 0.13 0 0.648 328CSNS1925 0.08 0.072 0.002 0.451 328DSAP1925 0.084 0.117 0 0.461 328KSC1925 0.102 0.081 0.001 0.467 328RSZML1925 0.152 0.144 0 0.552 328CSDSD1920 0.192 0.156 0 0.629 318CSND1920 0.053 0.044 0 0.293 318DSAP1920 0.160 0.200 0 0.726 318RSZML1920 0.159 0.138 0 0.547 298SdP1920 0.104 0.131 0 0.593 228

19

Table 2: Determinants of Czech and German radio availability

Czech signal strength German signal strengthGermans share -35.57∗∗∗ (10.13) -4.278 (6.613)Jews share 805.3∗∗ (405.9) -422.7∗ (243.2)Other ethnicity share -46.76∗∗∗ (15.17) -0.933 (13.64)BdL1925 -6.102 (8.839) -4.982 (6.326)CSDSD1925 -40.03∗∗∗ (14.31) -8.134 (8.845)CSL1925 -20.66 (14.93) -14.33 (8.699)CSNS1925 -61.57∗∗∗ (16.58) -19.55∗∗ (9.057)DSAP1925 -28.44∗∗∗ (10.50) -5.385 (8.960)KSC1925 -23.55∗ (13.38) -13.42∗ (7.974)RSZML1925 -54.12∗∗∗ (13.90) -8.581 (8.174)Sudetenland dummy 1.355 (2.286) 1.342 (1.275)Major towns dummy -2.546∗ (1.440) 0.198 (0.903)Population density 0.00471∗∗∗ (0.000937) 0.000110 (0.000716)Population size 0.0163∗∗∗ (0.00603) 0.000590 (0.00326)Log of income -8.811∗ (5.021) 3.937∗ (1.999)Unemployment rate -2.973 (19.24) 26.00∗∗ (11.65)Taxpayers per 1000 people 0.207∗∗ (0.0977) -0.107∗∗ (0.0434)25-44 age share 95.27∗∗∗ (32.61) 26.07 (24.68)45-64 age share -115.1∗∗ (52.12) -16.38 (37.61)65 and higher age share 104.6 (71.21) -5.776 (48.37)Adult female share -23.38 (33.81) 2.820 (29.03)Protestant share 14.31 (9.434) 3.312 (7.381)Hussite church share 4.819 (8.751) 11.65∗∗ (5.001)Judaism share -287.8∗ (153.4) 7.312 (129.4)Other faith share -27.50 (28.05) -36.17∗∗ (17.65)Atheism share -22.73∗ (13.19) 6.383 (8.675)Constant 101.4∗∗∗ (30.74) -9.446 (24.61)N 328 328adj. R2 0.384 0.201Note: Standard errors clustered by political county in parenthesis. *** p < 0.01, ** p < 0.05,* p < 0.1.

20

Table 3: Altonji-Elder-Taber test

CSDSD DSAP KSC NarSj RSZML SdPPredicted Czech signal -0.000140 -0.00265∗∗∗ 0.00270∗∗∗ 0.000307 -0.00637∗∗∗ 0.00186(based on all controls) (0.000768) (0.000674) (0.000704) (0.000749) (0.00103) (0.00121)

Predicted German signal 0.00811∗∗∗ 0.00245∗ 0.00580∗∗∗ 0.00508∗∗∗ -0.0142∗∗∗ 0.00143(based on all controls) (0.00197) (0.00129) (0.00175) (0.000976) (0.00195) (0.00175)ethnicity controls yes yes yes yes yes yesN 328 328 328 328 328 328adj. R2 0.580 0.725 0.233 0.414 0.790 0.941Note: Standard errors clustered by political county in parenthesis. *** p < 0.01, ** p < 0.05, * p < 0.1. Ethnicity controlsinclude shares of Czech & Slovaks, Germans, Jews and others. For predicting singal strengths we use voting, population,economic, demographic and religion controls. Voting controls include vote shares of BdL, CSDSD, CSL, CSNS, DSAP,KSC and RSZML in 1925. Population controls include Sudetenland and major towns dummies, population size andpopulation density. Economic controls include log of income per capita, unemployment rate and number of taxpayers per1000 people. Demographic controls include shares of people aged 25-44, 45-64 and 65+ as well as share of females in adultpopulation. Religion controls include shares of Catholicism, Protestant, Hussite church, Judaism, others and atheists.

Table 4: Main resultsCSDSD DSAP KSC NarSj RSZML SdP

Czech signal strength -0.00018 0.00059∗∗∗ 0.00031∗ -0.00039∗∗∗ 0.0001 -0.0014∗∗

(0.00017) (0.00018) (0.00017) (0.00014) (0.00017) (0.00057)

German signal strength -0.00061∗∗∗ -0.000174 0.0005∗ 0.00034∗ -0.00041∗ 0.0019∗∗

(0.00021) (0.00028) (0.0003) (0.00021) (0.00023) (0.00084)ethnicity controls yes yes yes yes yes yesvoting controls yes yes yes yes yes yespopulation controls yes yes yes yes yes yeseconomic controls yes yes yes yes yes yesdemographic controls yes yes yes yes yes yesreligion controls yes yes yes yes yes yesN 328 328 328 328 328 328adj. R2 0.928 0.912 0.839 0.744 0.967 0.951Note: Standard errors clustered by political county in parenthesis. *** p < 0.01, ** p < 0.05, * p < 0.1. Ethnicity controlsinclude shares of Czech & Slovaks, Germans, Jews and others. Voting controls include vote shares of BdL, CSDSD, CSL,CSNS, DSAP, KSC and RSZML in 1925. Population controls include Sudetenland and major towns dummies, populationsize and population density. Economic controls include log of income per capita, unemployment rate and number oftaxpayers per 1000 people. Demographic controls include shares of people aged 25-44, 45-64 and 65+ as well as share offemales in adult population. Religion controls include shares of Catholicism, Protestant, Hussite church, Judaism, othersand atheists.

21

Table 5: Regressions with partial controls

vote share of SdP in 1935 electionsCzech signal strength -0.00860∗∗∗ -0.000360 -0.000894∗ -0.000845∗ -0.00121∗∗ -0.00144∗∗ -0.00144∗∗

(0.00144) (0.000534) (0.000466) (0.000485) (0.000525) (0.000572) (0.000569)

German signal strength 0.0135∗∗∗ 0.00161∗ 0.00147∗ 0.00141∗ 0.00185∗∗ 0.00170∗∗ 0.00185∗∗

(0.00271) (0.000883) (0.000816) (0.000811) (0.000808) (0.000839) (0.000839)ethnicity controls no yes yes yes yes yes yesvoting controls no no yes yes yes yes yespopulation controls no no no yes yes yes yeseconomic controls no no no no yes yes yesdemographic controls no no no no no yes yesreligion controls no no no no no no yesN 328 328 328 328 328 328 328adj. R2 0.224 0.942 0.945 0.944 0.948 0.950 0.951Note: Standard errors clustered by political county in parenthesis. *** p < 0.01, ** p < 0.05, * p < 0.1. Ethnicity controls includeshares of Czech & Slovaks, Germans, Jews and others. Voting controls include vote shares of BdL, CSDSD, CSL, CSNS, DSAP, KSCand RSZML in 1925. Population controls include Sudetenland and major towns dummies, population size and population density.Economic controls include log of income per capita, unemployment rate and number of taxpayers per 1000 people. Demographiccontrols include shares of people aged 25-44, 45-64 and 65+ as well as share of females in adult population. Religion controls includeshares of Catholicism, Protestant, Hussite church, Judaism, others and atheists.

22

Table 6: Placebo test with 1920 electionsCSDSD DSAP CSND RSZML SdP

Czech signal strength 0.000780∗∗ -0.000828∗∗∗ -0.000504∗∗∗ -0.000365∗ 0.00296∗∗∗

(0.000321) (0.000275) (0.000124) (0.000220) (0.000688)

German signal strength 0.000452 0.000515 0.0000449 0.0000667 -0.000630(0.000430) (0.000428) (0.000216) (0.000414) (0.00100)

ethnicity controls yes yes yes yes yesvoting controls yes yes yes yes yespopulation controls yes yes yes yes yeseconomic controls yes yes yes yes yesdemographic controls yes yes yes yes yesreligion controls yes yes yes yes yesN 318 318 318 298 228adj. R2 0.937 0.976 0.766 0.937 0.787Note: Standard errors clustered by political county in parenthesis. *** p < 0.01, ** p < 0.05, * p < 0.1. Ethnicitycontrols include shares of Czech & Slovaks, Germans, Jews and others. Voting controls include vote shares ofBdL, CSDSD, CSL, CSNS, DSAP, KSC and RSZML in 1925. Population controls include Sudetenland and majortowns dummies, population size and population density. Economic controls include log of income per capita,unemployment rate and number of taxpayers per 1000 people. Demographic controls include shares of people aged25-44, 45-64 and 65+ as well as share of females in adult population. Religion controls include shares of Catholicism,Protestant, Hussite church, Judaism, others and atheists. As the vote share of SdP we use the combined vote sharesof DNP and DNSAP. Number of observations vary because DNP, DNSAP and RSZML did not run in all judicialcounties in 1920 elections. KSC did not yet exist in 1920.

Table 7: Placebo test with 1929 electionsCSDSD DSAP KSC CSND RSZML SdP

Czech signal strength -0.00016 0.00029 -0.00039∗ -0.00031 0.001∗ -0.00036(0.00027) (0.00023) (0.00023) (0.0002) (0.00056) (0.00092)

German signal strength -0.000087 0.00037 0.00022 0.00013 -0.000031 -0.000099(0.00035) (0.00028) (0.00029) (0.00034) (0.00051) (0.00098)

ethnicity controls yes yes yes yes yes yesvoting controls yes yes yes yes yes yespopulation controls yes yes yes yes yes yeseconomic controls yes yes yes yes yes yesdemographic controls yes yes yes yes yes yesreligion controls yes yes yes yes yes yesN 328 328 328 328 328 284adj. R2 0.921 0.948 0.813 0.485 0.807 0.489Note: Standard errors clustered by political county in parenthesis. *** p < 0.01, ** p < 0.05, * p < 0.1. Ethnicitycontrols include shares of Czech & Slovaks, Germans, Jews and others. Voting controls include vote shares ofBdL, CSDSD, CSL, CSNS, DSAP, KSC and RSZML in 1925. Population controls include Sudetenland and majortowns dummies, population size and population density. Economic controls include log of income per capita,unemployment rate and number of taxpayers per 1000 people. Demographic controls include shares of people aged25-44, 45-64 and 65+ as well as share of females in adult population. Religion controls include shares of Catholicism,Protestant, Hussite church, Judaism, others and atheists. As the vote share of SdP we use the combined vote sharesof DNP and DNSAP. There are less observations in case of SdP because DNP and DNSAP did not run in all judicialcounties in 1929 elections.

23