Content uploaded by Karsten Müller
Author content
All content in this area was uploaded by Karsten Müller on Feb 28, 2021
Content may be subject to copyright.
© The Author(s) 2020. Published by Oxford University Press on behalf of European Economic Association. All rights reserved.
For permissions, please e-mail: journals.permissions@oup.com
ORIGINAL UNEDITED MANUSCRIPT
FANNING THE FLAMES OF HATE:
SOCIAL MEDIA AND HATE CRIME
Karsten Müller
Princeton University, The
Julis-Rabinowitz Center for Public Policy
and Finance
Carlo Schwarz
Bocconi University, Department of
Economics
Abstract
This paper investigates the link between social media and hate crime. We show that anti-refugee
sentiment on Facebook predicts crimes against refugees in otherwise similar municipalities with
higher social media usage. To establish causality, we exploit exogenous variation in the timing of
major Facebook and internet outages. Consistent with a role for “echo chambers”, we nd that right-
wing social media posts contain narrower and more loaded content than news reports. Our results
suggest that social media can act as a propagation mechanism for violent crimes by enabling the
spread of extreme viewpoints. (JEL: D74, J15, Z10, D72, O35)
1. Introduction
Social media has come under increasing scrutiny in recent years. In the wake of
the 2016 presidential election in the United States, for example, relatively recent
phenomena such as fake news, social media echo chambers, and bot farms have been
subjects of widespread media coverage and public discourse (e.g New York Times
2016, 2017a). The role of online hate speech in particular has been at the center of an
intense and polarized debate. Despite public interest and calls for policy action, there
is little empirical evidence on how hateful social media content translates into real-life
behavior.
The editor in charge of this paper was Imran Rasul.
Acknowledgments: We would like to thank the editor, Imran Rasul, and four anonymous referees for their
comments, which greatly improved the paper. We are also grateful to Sascha Becker, Christopher Blattman,
Leonardo Bursztyn, Mirko Draca, Ruben Enikolopov, Thiemo Fetzer, Evan Fradkin, Matthew Gentzkow,
Andy Guess, Vardges Levonyan, Atif Mian, Magne Mogstad, Sharun Mukand, Hans-Joachim Voth, Fabian
Waldinger, Noam Yuchtman, and seminar participants at the NBER Summer Institute, University of
Chicago, EEA Conference 2018, Transatlantic Doctoral Conference (LBS), Oxford Internet Institute,
Geneva Academy of Humanitarian Law, Bruneck Political Economy Workshop, Leverhulme Causality
Conference at the University of Warwick, Spring Meeting of Young Economists 2019, the Royal Economic
Society 2019, and the UNHCR Conference on Forced Displacement for their helpful suggestions. We would
also like to thank the Amadeu Antonio Stiftung for sharing their data on refugee attacks with us. Müller
was supported by a Doctoral Training Centre scholarship granted by the ESRC [grant number 1500313].
Schwarz was supported by a Doctoral Scholarship from the Leverhulme Trust.
E-mail: karstenm@princeton.edu (Müller); carlo.schwarz@unibocconi.it (Schwarz)
Journal of the European Economic Association
Preprint prepared on 5 September 2020 using jeea.cls v1.0.
Downloaded from https://academic.oup.com/jeea/advance-article/doi/10.1093/jeea/jvaa045/5917396 by guest on 28 October 2020
ORIGINAL UNEDITED MANUSCRIPT
Müller and Schwarz Fanning the Flames of Hate 2
In this paper, we investigate the role of social media in the propagation of hate
crimes. Previous research has shown that traditional media can play a role in violent
outbursts or ethnic hatred (e.g Yanagizawa-Drott 2014; Adena et al. 2015; DellaVigna
et al. 2014). In contrast to traditional media, social media platforms allow users to easily
self-select into niche topics and extreme viewpoints. This preferential selection may
limit the spectrum of information people absorb and create “echo chambers” (Sunstein
2009, 2017), which reinforce similar ideas (see e.g. Bessi et al. 2015; Del Vicario et al.
2016; Schmidt et al. 2017). Social media has also become a widely-consumed news
source, particularly for young people: in Germany, for example, social media is among
the main news sources of 18 to 25 year olds (Hölig and Hasebrink 2016). In the US,
around half of all adults use social media to get news and two thirds of Facebook users
use it as a news source (Pew Research Center 2018). This suggests that social media
could be particularly effective in propagating hateful sentiments.
We study the link between anti-refugee sentiment on Facebook and hate crimes
against refugees in Germany. The German setting is motivated by the inux of around
one million refugees into the country between 2015 and 2016 (BAMF 2016), which
was accompanied by frequent violent crimes committed against them (see, for example,
recent video coverage by New York Times 2017b). Between January 2015 and early
2017 alone, the non-prot organization “Amadeu Antonio Stiftung” recorded around
3,300 anti-refugee incidents, including over 750 cases of arson or outright assault.
We posit that social media can reinforce anti-refugee sentiments, which may
push some potential perpetrators over the edge to carry out violent acts. Our empirical
strategy exploits differences in Facebook usage at the municipal level and weekly
variation in anti-refugee sentiment on social media. We create a novel measure for the
salience of anti-refugee hate speech on social media based on the Facebook page of the
“Alternative für Deutschland” (Alternative for Germany, AfD hereafter), a relatively
new right-wing party that became the third-strongest faction in the German parliament
following the 2017 federal election. The AfD has positioned itself as an anti-refugee
and anti-immigration party. With more than 300,000 followers, 175,000 posts, 290,000
comments, and 500,000 likes (as of early 2017), their Facebook page has a broader
reach than that of any other German party.1
This widespread reach makes the AfD’s Facebook page uniquely suited to
measure anti-refugee sentiment on social media. In contrast to established political
parties like Angela Merkel’s Christian Democratic Union (CDU) or the German Social
Democrats (SPD), the AfD allows users to directly post messages on its Facebook wall.
The AfD is also the only party that does not explicitly outline rules of conduct, e.g.
by threatening to remove racist, discriminating, or otherwise hateful comments. We
show that the content on the AfD page is consistently more focused on refugees than
that of traditional news reports and frequently contains loaded terms that civil rights
groups have identied as “hate speech”. These detailed data also allow us to construct
1. We provide a short history of the AfD in Appendix A in the online appendix.
Journal of the European Economic Association
Preprint prepared on 5 September 2020 using jeea.cls v1.0.
Downloaded from https://academic.oup.com/jeea/advance-article/doi/10.1093/jeea/jvaa045/5917396 by guest on 28 October 2020
ORIGINAL UNEDITED MANUSCRIPT
Müller and Schwarz Fanning the Flames of Hate 3
a measure of each town’s exposure to Germany-wide anti-refugee sentiment using the
share of the population that is active on the AfD Facebook page.
Using xed effects panel regressions, we nd that—during periods of high
salience of refugees on right-wing social media—anti-refugee hate crimes increase in
areas with higher Facebook usage. This correlation is especially pronounced for violent
incidents such as assault. Controlling for a large vector of municipality characteristics,
interacted with our salience measure, makes little difference for the magnitude and
statistical signicance of these estimates.
A concern is that our measures of exposure to right-wing social media may
be correlated with unobserved municipal characteristics that explain disproportionate
increases in hate crimes during times of high anti-refugee sentiment. To narrow down
the social media transmission channel, we provide quasi-experimental evidence using
the exact timing of country-wide Facebook outages and local internet disruptions,
which reduce the number of social media posts.
To begin, we study large, Germany-wide Facebook outages resulting from
programming or server problems at the platform. These outages disrupt users’ exposure
to this particular social media platform without affecting other online channels. We nd
that Facebook disruptions reduce local hate crimes, particularly in areas with many
AfD users. Further, during Facebook outages, higher anti-refugee sentiment is not
associated with a differential increase in hate crimes in areas with high Facebook usage.
These results suggest that social media might play a propagating role in translating
online content into ofine violence.
We also exploit the precise timing of hundreds of local internet disruptions
as a source of granular exogenous variation in access to social media. These
local disruptions reduce a particular town’s exposure to social media content while
leaving Germany-wide refugee salience unaffected. Notably, the frequency of internet
disruptions is geographically dispersed and largely unrelated to observable local
characteristics, including AfD likes on Facebook.
We nd that, while hate crimes increase in periods of higher refugee
salience, this correlation disappears for municipalities experiencing an internet outage.
Quantitatively, a typical internet disruption fully mediates the link between social
media and hate crime. Further, once we take into account social media transmission,
these internet outages themselves are no longer associated with anti-refugee incidents,
nor are their interactions with local internet usage or mobile internet access. These
results point to social media as propagation mechanism rather than other online
channels. It also makes it unlikely that we are capturing a “displacement effect” arising
from potential perpetrators xing their internet access.
We also analyze how other salient news events mediate the link of anti-refugee
Facebook posts with the number of violent incidents, building on Eisensee and
Strömberg (2007) and Durante and Zhuravskaya (2018). Specically, we look at the
European Soccer Championship, Brexit, and Donald Trump’s presidential election, all
of which crowded out the salience of refugees. Similar to our outage results, social
media exposure has a signicantly more muted relationship with hate crimes during
these events. The link we uncover appears to be specic to anti-refugee sentiment: other
Journal of the European Economic Association
Preprint prepared on 5 September 2020 using jeea.cls v1.0.
Downloaded from https://academic.oup.com/jeea/advance-article/doi/10.1093/jeea/jvaa045/5917396 by guest on 28 October 2020
ORIGINAL UNEDITED MANUSCRIPT
Müller and Schwarz Fanning the Flames of Hate 4
posts on the AfD Facebook page, e.g. those related to Muslims or the European Union,
do not have the same predictive power for anti-refugee hate crimes. Consistent with the
hypothesis that social networks can act as transmission channel, the correlation with
hate crime is larger in regions where AfD users show higher Facebook engagement.
When interpreting our results, we do not claim that social media itself causes
crimes against refugees out of thin air. Rather, our argument is that social media
can act as a propagating mechanism for hateful sentiments that likely have many
fundamental sources. We nd evidence for two potential channels. First, our results
are driven by refugee attacks committed by groups of perpetrators. This suggests that
social media may motivate collective action, consistent with existing evidence on other
political outcomes such as protests (e.g. Enikolopov et al. 2016). Second, we nd
evidence for a spillover channel. Hate crimes are considerably more common in weeks
when neighboring towns also experience them, and this is particularly true for towns
with many right-wing social media users when anti-refugee sentiment is elevated.
In contrast, we nd little evidence that social media provides useful information
to perpetrators. Our results are also unlikely to be explained by persuasion effects,
because we focus on high-frequency variation.
Related Literature. Our work provides evidence that social media may have effects
on real-life outcomes, as measured by hate crimes. We build on existing work on
media exposure and persuasion (see e.g. DellaVigna and Gentzkow 2010; DellaVigna
and Ferrara 2015). In addition to work on traditional media and violence cited above,
Dahl and DellaVigna (2009) show that—in contrast to experimental settings—violent
movies decrease violent crime in the eld due to displacement effects. Television
has also been associated with short-lived outbursts of domestic violence (Card and
Dahl 2011). In other research, Bhuller et al. (2013) demonstrate that exposure to
pornographic material on the internet is linked to increased sex crime. Bursztyn et al.
(2017) nd that media coverage of close elections increases voter turnout, while
Gavazza et al. (2018) show that broadband diffusion decreased voter turnout in the
United Kingdom (see also Gentzkow 2006; Manacorda and Tesei 2020). Enikolopov
et al. (2016) nd that social media exposure spurs protest participation in Russia by
reducing coordination costs.
We contribute to this literature by investigating the role of social media in
stirring up violence. Previous research has documented the prevalence of online hate
speech (Oksanen et al. 2014). Other work has shown that Google search data can be
used to measure racial animus (Stephens-Davidowitz 2014). In complementary work,
we study the effect of Twitter usage on anti-minority sentiments in the United States
(Müller and Schwarz 2018). Bursztyn et al. (2019) study the effect of social media on
xenophobia in Russia. In contrast to these papers, we focus on the short-run impact of
social media posts, rather than long-run effects that may work through persuasion or
changes in social norms.
Our paper also builds on research about the polarization of citizens (e.g Fiorina
and Abrams 2008). There is no consensus on whether social media increases or
decreases polarization: some authors argue that social media are divisive (Pariser 2011;
Journal of the European Economic Association
Preprint prepared on 5 September 2020 using jeea.cls v1.0.
Downloaded from https://academic.oup.com/jeea/advance-article/doi/10.1093/jeea/jvaa045/5917396 by guest on 28 October 2020
ORIGINAL UNEDITED MANUSCRIPT
Müller and Schwarz Fanning the Flames of Hate 5
Gabler 2016), while others nd that polarization decreases with social media usage
(Barberá 2014; Boxell et al. 2017). Our work suggests that—independent of whether
social media affects overall polarization or not—social media content can be associated
with violent crimes.
We also build on the literature on culture and violence. Summarizing a vast
body of research, Alesina and La Ferrara (2005) nd that cultural and and religious
fragmentation predict the likelihood of civil war across countries. Voigtlander and
Voth (2012) show that anti-Semitic violence in Germany is highly persistent: pogroms
during the era of the Black Death predict pogroms in the 1920s, Jewish deportations,
and synagogue attacks during the rise of the Nazi party. Similarly, Jha (2013) shows
that medieval interethnic complementarities in trade decrease the likelihood of modern
Hindu-Muslim riots. These papers, however, are largely silent on the existence of
volatile, short-lived bursts of sentiment leading to violent incidents. As such, our work
is also related to Fouka and Voth (2013), who show that monthly variation in public
acrimony between Greek and German politicians during the Greek debt crisis affected
German car purchases particularly in areas of Greece where German troops committed
war crimes during World War II. Our results also align with the ndings of Colussi et al.
(2016), who show that a higher salience of minority groups increases the likelihood of
hate crimes.
While traditional media such as television are regulated in most countries,
legislators are now beginning to address social media. Our work is thus particularly
topical in light of the political discussions in many countries about anti-hate speech
laws and censoring hate speech on social media. The German parliament, for example,
passed an anti online hate speech law (“Netzwerkdurchsetzungsgesetz”) on June 30,
2017, which threatens providers of online platforms such as Facebook with nes
up to EUR 50 million for failing to delete “criminal” content that is “obviously
unlawful”. The controversial law was the initiative of German Minister of Justice
Heiko Maas, who lamented social media platforms’ unwillingness to address “online
hate crime”.2The European Union has issued independent guidelines calling on social
media companies to remove illegal hate speech as well. In the United Kingdom, the
Crown Prosecution Service plans to increase prosecution of online hate crimes (The
Guardian 2017; BBC 2017). Our paper serves as a rst attempt to address this important
topic empirically.
The paper proceeds as follows. In Section 2 we introduce the data used in our
empirical analysis. Section 3 presents the results. Section 4 concludes.
2. Data
We construct a dataset on social media activity and anti-refugee hate crimes in
Germany. In total, we combine data from 12 different sources which we describe in
2. See, for example, the ofcial statement of the German parliament on bundestag.de.
Journal of the European Economic Association
Preprint prepared on 5 September 2020 using jeea.cls v1.0.
Downloaded from https://academic.oup.com/jeea/advance-article/doi/10.1093/jeea/jvaa045/5917396 by guest on 28 October 2020
ORIGINAL UNEDITED MANUSCRIPT
Müller and Schwarz Fanning the Flames of Hate 6
more detail in the following subsections: (1) Municipal-level data on anti-refugee hate
crimes; (2) Facebook data on posts, likes, and comments on the AfD page; (3) hand-
collected municipal-level data on Facebook user locations; (4) municipal-level data
on internet outages; (5) a hand-coded dataset on major weekly Facebook outages;
(6) municipal- and county-level socioeconomic data from the German Statistical
Ofce; (7) municipal-level voting data; (8) county-level data on broadband access;
(9) municipal-level data on newspaper sales; (10) data on the content of reporting
about refugees from Nexis; (11) city-level data on neo-Nazi murders and historical anti-
Semitism; and (12) weekly Google search data on major news events in our sample.
The nal panel dataset covers 4,466 German municipalities for the 111 weeks from
1st January 2015 to 13th February 2017. Summary statistics for the main variables
of interest can be found in Table 1 and Table B.3. The online appendix provides a
comprehensive overview of the data sources and variable denitions (see Table B.4).
Table 1. Summary statistics for main variables.
Level Obs Mean SD Min. Max.
Refugee Attacks
Refugee attacks Muni.-Week 495,726 0.007 0.099 0 8
Arson attacks Muni.-Week 495,726 0.000 0.022 0 2
Other property damage Muni.-Week 495,726 0.004 0.076 0 8
Assaults Muni.-Week 495,726 0.001 0.035 0 3
Protests Muni.-Week 495,726 0.001 0.030 0 5
Social Media Data
AfD users/Pop.Municipality 495,726 0.301 0.286 0 8
Refugee posts Week 495,726 84 61 2 259
Posts/AfD users Municipality 395,493 0.554 3.882 0 118
Comments/AfD users Municipality 395,493 1.1 7.3 0 270
Likes/AfD users Municipality 395,493 1.8 12.3 0 370
Auxiliary Variables
IInternet outage Muni.-Week 495,726 0.001 0.025 0.000 1.000
IFacebook outage Municipality 495,726 0.072 0.259 0.000 1.000
Baseline Controls
Ln(Population (2015)) Municipality 495,726 9 1 6 15
GDP/Worker County 493,617 63,095 9,846 46,835 136,763
Population density Municipality 495,726 282 382 7 4,653
AfD vote share (2017) (in %) Municipality 492,618 15 7 3 45
Share high school (in %) Municipality 495,726 29 8 0 58
Share broadband access (in %) Municipality 495,726 83 11 44 100
Share immigrants (in %) Municipality 483,072 14 8 2 50
Asylum Seekers/Pop. County 495,726 0.011 0.006 0.000 0.102
Notes: This table reports summary statistics for the main variables in the estimation sample. Variables
tagged with a are scaled by population (in 1,000).
Journal of the European Economic Association
Preprint prepared on 5 September 2020 using jeea.cls v1.0.
Downloaded from https://academic.oup.com/jeea/advance-article/doi/10.1093/jeea/jvaa045/5917396 by guest on 28 October 2020
ORIGINAL UNEDITED MANUSCRIPT
Müller and Schwarz Fanning the Flames of Hate 7
2.1. Anti-Refugee Incidents
The data on incidents targeting refugees were collected by the Amadeu Antonio
Foundation and Pro Asyl (a pro asylum NGO).3These data cover incidents including
anti-refugee grafti, arson of refugee homes, assault, and incidents during protests in
Germany between January 2015 and early 2017. This period is of particular interest
since it includes the beginning and height of the refugee crisis in Germany. All 3,335
anti-refugee aggressions feature a short description and are classied into four groups.
The most common cases are property damage to refugee homes (2,226 incidents),
followed by assault (534), incidents during anti-refugee protests (339), arson (225). 11
events are classied as suspected cases that were still under investigation. Table B.2 in
the online appendix lists examples for each class of anti-refugee activity.
All incidents are geo-coded with an exact longitude and latitude, which we use
to assign them to municipalities.4Section 2.1 shows the location of the anti-refugee
incidents in our observation period for each German municipality.
The data appear to be high quality. Each entry has a clearly indicated source.
Nearly half of the incidents in the dataset are reported by the federal government in
response to inquiries by the left-wing party “Die Linke”. Other sources include police
reports and national or local media outlets. We hand-checked a random sample of 100
incidents and found their coding accurately reected the information reported in the
respective source.
2.2. Facebook Data on Refugee Salience
We construct a proxy for the frequency of anti-refugee messaging on social media
based on the Facebook page of the AfD. We chose the AfD’s page because the party
is by far the most popular far-right political movement in Germany. At the time of
the refugee crisis, the AfD also had the highest number of Facebook followers of any
German party. This makes their page arguably the most important platform of exchange
about refugees among Germany’s right-wing social media users.
We start by using the Facebook Graph API to collect all status posts, comments,
and likes from the AfD Facebook page (see Appendix B.1 for an introduction to
Facebook). The API provides a unique identier for each post, allowing us to link posts
to comments and likes, as well as the users who posted, commented, or liked anything
on the page. Overall, we collected 176,153 posts, 290,854 comments, 510,268 likes,
and 93,806 individual user IDs.
3. These data are available at https://www.mut-gegen-rechte-gewalt.de/service/chronik-vorfaelle.
4. To assign coordinates to municipalities, we use the shape les provided by the ©GeoBasis-DE/BKG 2016
website. The shape le contains data for the 4,679 German municipalities (“Gemeindeverwaltungsverband”).
213 of these municipalities do not have inhabitants (e.g. forest areas) nor anti-refugee incidents. After
dropping these cases, we are left with 4,466 municipalities in our estimation sample. We use the level of the
“Gemeindeverwaltungsverband” since these exhibit smaller differences in their size and population than
the 11,165 German “Gemeinden” and are therefore more suitable for spatial analysis according to the data
provider (see link).
Journal of the European Economic Association
Preprint prepared on 5 September 2020 using jeea.cls v1.0.
Downloaded from https://academic.oup.com/jeea/advance-article/doi/10.1093/jeea/jvaa045/5917396 by guest on 28 October 2020
ORIGINAL UNEDITED MANUSCRIPT
Müller and Schwarz Fanning the Flames of Hate 8
1st Quintile
2nd Quintile
3rd Quintile
4th Quintile
5th Quintile
No Users
Anti-Refugee Incident
Figure 1. AfD Facebook usage per capita and anti-refugee incidents. This map plots the number of
Facebook users of the Alternative for Germany (AfD) page per capita for each of the 4,466 German
municipalities. The dots indicate the locations of the 3,335 anti-refugee incidents from the Amadeu
Antonio Foundation.
As our baseline measure for the salience of anti-refugee hate speech on social
media, we use the number of posts on the AfD Facebook page that contain the
word “Flüchtling” (refugee) in any given week. The narrative in these posts centers
around the idea that the “elites”—politicians and mainstream media outlets—have
betrayed “the people” by allowing “streams” of illegitimate “economic refugees” to
enter the country, who are described as being criminals and rapists for “cultural
reasons”. Table B.1 in the online appendix provides a few representative examples;
Section 3.5 provides a more in-depth analysis. A potential downside of this approach
is that we may inadvertently tag posts that do not express negative sentiments towards
refugees. However, a careful content analysis of posts and comments reveals that
Journal of the European Economic Association
Preprint prepared on 5 September 2020 using jeea.cls v1.0.
Downloaded from https://academic.oup.com/jeea/advance-article/doi/10.1093/jeea/jvaa045/5917396 by guest on 28 October 2020
ORIGINAL UNEDITED MANUSCRIPT
Müller and Schwarz Fanning the Flames of Hate 9
the overwhelming majority appear to agree with the positions of the AfD. This is
perhaps unsurprising given that only people who “like” the AfD Facebook page will
be informed about new posts. Critics, on the other hand, have a strong incentive not to
indicate publicly that they “like” the party.
We plot the total number of AfD Facebook page posts about refugees and the
number of anti-refugee incidents in Figure 2. Weeks with more refugee posts also tend
to have more anti-refugee events. Both series clearly spike during salient events related
to refugees, such as Angela Merkel’s widely reported statement “Wir schaffen das”
(“We can do this”) during a press conference on the challenges of the refugee situation.
A simple time series regression of refugee attacks on AfD posts yields a R2of 0.34
(unreported).
Reports about
sexual assaults
by refugees
in Cologne
Merkel speech:
"Wir schaffen das"
Two terrorist
attacks by
refugees
0
50
100
150
0
50
100
150
200
250
2015w1 2015w26 2016w1 2016w27 2017w1
Attacks on refugees (left axis) AfD Facebook posts about refugees (right axis)
Figure 2. Refugee posts on social media and anti-refugee incidents over time. This gure plots
the number of posts about refugees on the Facebook page of the “Alternative for Germany” and the
number of anti-refugee incidents in Germany over time.
2.3. Municipal-Level Facebook Measures
We construct a measure of exposure to right-wing social media at the municipal
level. Because survey data about German Facebook usage are, to our knowledge, only
available at the level of the 16 federal states, we hand-collect user location data by using
the unique user identiers provided by the Facebook Graph API. Due to Facebook’s
privacy policy, we are only able to collect this information for people who make it
publicly available.
Because we are interested in the transmission of right-wing social media
sentiment, we measure exposure to it on Facebook based on users of the AfD page.
Journal of the European Economic Association
Preprint prepared on 5 September 2020 using jeea.cls v1.0.
Downloaded from https://academic.oup.com/jeea/advance-article/doi/10.1093/jeea/jvaa045/5917396 by guest on 28 October 2020
ORIGINAL UNEDITED MANUSCRIPT
Müller and Schwarz Fanning the Flames of Hate 10
In total, we can identify 93,806 users who interacted with the page at least once.5We
were able to hand-collect and geocode a place of residence for 34,396 of these users.
Overall, we were able to identify at least one AfD Facebook page user for 3,563 of
the 4,466 municipalities.6In Section 2.1 we visualize the distribution of AfD users per
capita. Anti-refugee incidents are concentrated in areas with more right-wing social
media users. To illustrate this, Figure B.3 in the online appendix shows the share of
municipalities with at least one refugee attack, depending on whether we can identify
at least one AfD Facebook page user. Municipalities with AfD users are three times
as likely to experience an attack during our observation period. Out of the total 3,335
attacks on refugees in our sample, 3,171 occurred in municipalities with AfD Facebook
page users. A t-test rejects the null hypothesis of no difference between the mean of the
two groups with a value of 22:95. Using the location data for AfD users, we can also
assign posts, comments, and likes to municipalities. Based on these data, we construct
auxiliary measures of social media interactions, e.g. the number of local posts scaled
over the number of AfD users.7
2.4. Data on Internet and Facebook Outages
We collect data on local internet outages from Heise Online. Heise lists user reports of
internet problems by telephone area codes and includes start times and duration. We
use area codes to assign internet problems to municipalities; the start date and duration
allow us to count the number of problems for each municipality and week.8The
internet outage reports are geographically dispersed with no clear patterns of regional
clustering (see Figure C.2a). The outages are also dispersed over time Figure C.2b.
To validate the Heise data, we search for newspaper reports on major internet
disruptions. While the large-scale and short-lived outages discussed in the newspaper
reports are not representative of the more localized and longer-lasting outages we
exploit in our regressions, they do suggest that the Heise data provide a valid proxy
for internet disruptions. For all major disruptions we could identify in newspapers, the
Heise data suggest an increase in the number of outages specic to the internet provider
5. The Facebook API does not provide data on which users “like” a page but only on users who interact
with a page, e.g. by liking another user’s comment. As a result, the total number of user IDs we have is
smaller than the more than 300,000 people who had liked the AfD Facebook page as of 2017.
6. Note that the decision of users to disclose their location is unlikely to matter in our setting. This
is because we exploit variation within the same location over time, which abstracts from time-invariant
endogenous selection using municipality xed effects.
7. We nd that some users post and comment excessively, which leads to a few outliers in measuring how
active users are in a given municipality. We therefore winsorize the number of posts, comments, and likes
we can attribute to local users at the 99.9th percentile to avoid individual users driving the results.
8. If an area code spans multiple municipalities, we assign an internet outage to the municipality that
overlaps most with the area code. We prefer this over to assigning the outage to all municipalities within the
area code’s territory because some area codes include minor overlaps with many municipalities. Assigning
an internet outage to all of these municipalities would introduce substantial noise.
Journal of the European Economic Association
Preprint prepared on 5 September 2020 using jeea.cls v1.0.
Downloaded from https://academic.oup.com/jeea/advance-article/doi/10.1093/jeea/jvaa045/5917396 by guest on 28 October 2020
ORIGINAL UNEDITED MANUSCRIPT
Müller and Schwarz Fanning the Flames of Hate 11
experiencing the outage. Table C.1 lists several examples of newspaper reports on such
outages and the respective information in our data.9
We focus on major outages that fulll two criteria: (1) they have to last longer
than 24 hours, and (2) they affect a signicant part of the population (be in the
top quartile of the reported internet problems to population ratio). This gets around
the issue that some reports may reect individual users’ glitches rather than general
disruptions.10
We also collect information on major Facebook disruptions. To identify these,
we start by searching for newspaper reports of Facebook problems in our sample period.
In total, we nd reports on eight large outages (see Table C.2 for an overview and more
details). We then validate their precise timing using the number of weekly user-reported
Facebook problems on the website of “Allestörungen”, a portal for aggregating user
complaints on individual websites and apps. Perhaps unsurprisingly, the eight outages
widely reported on in the news media are also associated with spikes in user-reported
problems.
Using these data, we dene a dummy variable that is 1 for weeks with Facebook
outages and 0 otherwise. These outages have the advantage that they are specic to
Facebook; in fact, they are uncorrelated with the total number of weekly internet
outages in a given week from our Heise data. In contrast to the internet disruptions,
the downside is that Facebook outages are rare, shorter, and only generate weekly
variation.
2.5. Auxiliary and Control Variables
We obtain control variables from a host of sources, which are explained in more detail
in the online appendix. Socioeconomic data on the municipality and county level
are from the German Statistical Ofce, available via www.regionalstatistik.de. We
include information on each municipality’s population by age group, GDP per worker,
population density, the share of the population with a high school degree (“Abitur”), the
share of the population receiving social benets, the share working in manufacturing,
and the vote results for the 2017 German Federal Election. To control for “pull factors”
of anti-minority crimes, we also obtain the share of the population that are immigrants
and asylum seekers.
To measure the extent to which people use the internet, we use the share of
households in a county with broadband access as well as average mobile download
speeds, collected by the Federal Ministry of Transport and Digital Infrastructure
9. To interpret the number of outages, note that the Heise data reports an average of four reported internet
outages per provider per week. That means even an increase of 15 reported outages represents a large
increase.
10. In some cases, users do not seem to report the end date of the internet outage, which can lead to
unlikely durations of several months. We thus winsorize the maximum duration at 3 weeks, but this choice
is not material for our results. We scale outages over population because towns with more inhabitants
mechanically also report more disruptions. As we discuss below, our results are robust to using alternative
denitions of this cut-off.
Journal of the European Economic Association
Preprint prepared on 5 September 2020 using jeea.cls v1.0.
Downloaded from https://academic.oup.com/jeea/advance-article/doi/10.1093/jeea/jvaa045/5917396 by guest on 28 October 2020
ORIGINAL UNEDITED MANUSCRIPT
Müller and Schwarz Fanning the Flames of Hate 12
(BMVI).11 In addition, we use the number of registered :de internet domains per capita
in a county to measure internet afnity, which has a correlation of 0.48 with broadband
access.
To measure the local penetration of traditional media, we obtain data
for 2016/2017 newspaper sales from the “Zeitungsmarktforschung Gesellschaft
der deutschen Zeitungen (ZMG)” (Society for Market Research of German
Newspapers).12 Based on this data, we construct a measure of traditional newspaper
consumption as the number of newspaper sales per capita.
For our comparison of social and more traditional media, we collected the
number of total and refugee-related reports in German news media from Nexis UNI
(previously LexisNexis). We use this to construct the weekly share of news reports
about refugees. For further analysis, we obtained the full text of all refugee-related
reports using the Lexis bulk data API, as well as all Facebook data from the pages
of ve major German newspapers (Welt, Frankfurter Allgemeine Zeitung (FAZ),
Tageszeitung (TAZ), Süddeutsche Zeitung (SZ), and Bild).
We also include controls for the local prevalence of right-wing extremism. One
such measure is the number of murders committed by neo-Nazis in each municipality
from 1990 until 2016, which were collected by “Mut gegen rechte Gewalt” (Courage
Against Right-Wing Violence). We complement this proxy for contemporary right-
wing violence with data on the historic prevalence of anti-semitism collected by
Voigtlander and Voth (2012).13
Finally, we obtain Google trends data on overall interest in the search terms
“Brexit”, “Trump”, and “UEFA EM 2016” in Germany to proxy for distracting news
events. Google scales the weekly number of searches for these terms on a scale from 0
to 100, where 100 marks the week with the highest search interest in the preceding
5 years. The time series plots in Figure D.1 in the online appendix suggest these
measures are sound approximations for attention paid to Brexit, the Trump election,
and the UEFA European Championship (one of the most widely followed sports events
in Germany).
11. Broadband access is highly correlated with publicly available survey data on individuals’ internet use
from Eurostat; these data are only available on the state level (see Figure B.4 in the online appendix).
12. These data contain the number of print newspapers sold in each municipality with more than 3,000
inhabitants. Newspapers are listed if, in any given town, they (1) sell at least 50 copies and (2) have a market
share of at least 1%. To have a similar sample size across specications, we impute values for 1,120 towns
for which news paper sales data are not available, based on a municipality’s population, population density,
AfD vote share, and county xed effects. However, the results are almost equivalent without imputation
(available upon request).
13. From their dataset, we use the natural logarithm of one plus the number of deported Jews as well
as one plus the number of letters written to “Der Stürmer”, the antisemitic newspaper published by Nazi
politician Julius Streicher. Towns with no information are coded as zero. We do not use scaled variables
because the data from Voigtlander and Voth (2012) only cover a fraction of the municipalities in our sample.
Journal of the European Economic Association
Preprint prepared on 5 September 2020 using jeea.cls v1.0.
Downloaded from https://academic.oup.com/jeea/advance-article/doi/10.1093/jeea/jvaa045/5917396 by guest on 28 October 2020
ORIGINAL UNEDITED MANUSCRIPT
Müller and Schwarz Fanning the Flames of Hate 13
3. Empirical Strategy and Main Results
3.1. Empirical Strategy
We begin to investigate the link between social media and anti-refugee incidents
by estimating xed effects panel regressions akin to a Bartik-type approach
(Bartik 1991). In particular, we use the interaction of local right-wing Facebook
usage (AfD Users/Popi) and weekly refugee posts on the AfD Facebook page
(Refugee Postst) to measure the differential change of hate crimes conditional on anti-
refugee sentiment on social media. This empirical set-up creates variation by week and
municipality, which we exploit in the following regression model:
Refugee attackit DˇAfD Users/PopiRefugee Postst
CControlsiRefugee Postst
CWeek FEtCMunicipality FEiC"it ;
(1)
The dependent variable is a dummy for the incidence of a refugee attack in municipality
iin week t.ˇmeasures the differential change in anti-refugee incidents conditional on
Germany-wide posts about refugees on the AfD page—as a proxy of Germany-wide
anti-refugee sentiment on social media—and right-wing social media users per capita.
We control for a host of local characteristics interacted with the refugee post measure.
Because we include many xed effects and interaction terms, we estimate 1 using
Ordinary Least Squares, which yields the linear probability model. Standard errors
are clustered by municipality. We consider alternative specications of the dependent
variable and standard errors in robustness exercises.
This framework has three key features. First, it circumvents reverse causality,
because refugee incidents in one town are unlikely to change anti-refugee sentiment in
all other towns. Second, our measure of social media exposure is time-invariant and
thus not the result of whether a municipality experiences refugee attacks in a given
week.14 Third, a full set of xed effects controls for unobserved heterogeneity that
affects all towns at the same time (such as salient news events), as well as time-invariant
differences across towns (such as a history of anti-minority violence).
The main concern with estimating Equation (1) is that AfD Users/Pop. may be
correlated with other municipality characteristics that could explain differences in how
local anti-refugee attacks co-vary with the salience of refugees online. In that case,
we would not be capturing a pure social media “effect”. For example, the share of
AfD Facebook subscribers may partially pick up general right-wing attitudes, which
could lead to more anti-refugee attacks in times of high refugee salience. This concern
may also not be sufciently addressed by controlling for interactions of observable
municipality characteristics with the refugee salience measure.
14. In the robustness section below, we alternatively measure local social media penetration before the
start of the refugee crisis, at the cost of reducing the number of users for whom we have location data. This
adjustment makes little difference for the results.
Journal of the European Economic Association
Preprint prepared on 5 September 2020 using jeea.cls v1.0.
Downloaded from https://academic.oup.com/jeea/advance-article/doi/10.1093/jeea/jvaa045/5917396 by guest on 28 October 2020
ORIGINAL UNEDITED MANUSCRIPT
Müller and Schwarz Fanning the Flames of Hate 14
We therefore develop an identication strategy based on Facebook and internet
outages. These disruptions induce plausibly exogenous variation in people’s exposure
to social media while leaving other local characteristics unchanged. The rst part of
this empirical strategy exploits the timing of major server problems at Facebook, which
disrupt access to the platform. In the second part, we build on the insight that German
internet infrastructure is trailing behind that of many other European Countries (e.g.
Latvia) and the OECD average (see Financial Times 2017; OECD 2016). As a result,
prolonged internet outages are relatively common. Because around 50% of worldwide
Facebook users accessed the platform with their computers, many users are exposed to
disruptions in internet access. In Germany, this share is likely to be even higher because
of the relatively slow adoption of mobile internet.15
Local internet outages are widely dispersed geographically: Figure C.2a
visualizes the distribution of disruptions per capita across Germany. The outages
are also not particularly clustered in a particular time period (see Figure C.2b).
Crucially, the frequency of internet problems is uncorrelated with the share of the
population on the AfD Facebook page. As such, internet disruptions provide exogenous
variation that is not already captured by our variable on local Facebook usage. The
number of reported internet problems is also uncorrelated with the total number of
refugee attacks in a given municipality. In fact, regressing the frequency of internet
outages on a host of municipality characteristics in Figure 3 suggests that they are
largely uncorrelated with observable factors: the estimated coefcients are nearly all
statistically indistinguishable from zero and quantitatively small. Taken together, our
interpretation is that whether an internet outage occurs in a given town and week is as
good as randomly assigned with regard to unobserved other factors that might drive
hate crimes.
We analyze the effect of Facebook and internet outages in a exible empirical
framework. We begin by asking whether these outages reduce anti-refugee attacks, and
whether they do so particularly in areas with a higher concentration of AfD Facebook
users. We then study whether these disruptions also decrease our baseline correlation
of local exposure to anti-refugee sentiment and hate crimes. More formally, the most
saturated regressions have the following triple difference form:
Refugee attackit DˇAfD Users/PopiRefugee Postst
COutageit AfD Users/PopiRefugee Postst
Cı1Outageit Cı2Outagei t Refugee Postst
Cı3Outageit AfD Users/Popi
C1ControlsiRefugee Postst
C2ControlsiOutageit
CWeek FEtCMunicipality FEiC"it ;
(2)
15. Data on Facebook usage patterns reported on Statista.com and on mobile internet usage in Germany
on (also on Statista.com) support this assessment.
Journal of the European Economic Association
Preprint prepared on 5 September 2020 using jeea.cls v1.0.
Downloaded from https://academic.oup.com/jeea/advance-article/doi/10.1093/jeea/jvaa045/5917396 by guest on 28 October 2020
ORIGINAL UNEDITED MANUSCRIPT
Müller and Schwarz Fanning the Flames of Hate 15
AfD users/Pop.
Population
GDP/worker
Population density
AfD vote share
Share high school degree
Share broadband access
Share immigrants
Asylum Seekers/Pop.
Registered domains/Pop.
Mobile Internet Speed
News paper sales/Pop.
Nazi murders/Pop.
NPD vote share
Deported Jews
Stürmer letters
Average age
Share benefit recipients
Share non-Christians
Manufacturing share
CDU vote share
SPD vote share
Left vote share
Green vote share
FDP vote share
Voter turnout
Share aged 0-24
Share aged 25-49
Share aged 50-74
Share aged 75+
-0.2 -0.1 0.0 0.1 0.2
Coefficients (in SD)
Figure 3. Balancedness — internet outages and local characteristics. This gure plots the
coefcients of the regression I nt er net out ages iD˛CX0°C"i, where the dependent variable is
the total number of internet outages in a municipality (based on our baseline denition) and Xis a
vector of local characteristics for which we plot the estimates. To make the magnitudes comparable,
we standardize all variables to have a mean of zero and standard deviation of one. 95% condence
intervals are based on standard errors clustered by municipality.
For the Facebook outages, which only vary by week, we replace Outagei t with
Outaget.16 For the initial tests, we focus on the estimates for ı1and ı3while
excluding the coefcients ˇ,,ı2, and 1. That is, we ask whether outages reduce
anti-refugee incidents, and whether they reduce them more in areas with more AfD
Facebook users. In the fully interacted regressions, the main coefcient of interest
captures the correlation of anti-refugee attacks and local exposure to anti-refugee
sentiment on social media, depending on whether an outage occurs. Put differently,
we test whether outages break the correlation between real-life incidents and refugee
salience, particularly for areas with high right-wing Facebook penetration. The vector
ControlsiOutageit controls for the differential effect of outages based on observable
characteristics, such as internet afnity.
The identifying assumption of this approach is that Facebook and internet
outages only affect anti-refugee incidents through their effect on social media exposure.
This assumption is plausible for Facebook outages. In the case of internet outages,
for which we have variation at the municipality-week level, one may be worried
16. Note that, as a result, the estimates of
ı1
and
ı2
in Equation (2) are absorbed by the week xed effects.
Journal of the European Economic Association
Preprint prepared on 5 September 2020 using jeea.cls v1.0.
Downloaded from https://academic.oup.com/jeea/advance-article/doi/10.1093/jeea/jvaa045/5917396 by guest on 28 October 2020
ORIGINAL UNEDITED MANUSCRIPT
Müller and Schwarz Fanning the Flames of Hate 16
about alternative online channels. We discuss these and other potential threats to
identication in the next section.
Exploiting variation in Facebook and internet outages also allow us to address
the concern that towns with a stronger right-wing presence may show differential trends
whenever the nationwide sentiment towards refugees changes. This is because these
relatively short-lived outages are unlikely to affect the presence of deep-rooted right-
wing attitudes in a municipality; absent online channels, the outages should thus not
have an impact on real-life outcomes. The frameworkin Equation (2) further addresses
reverse causality concerns. If we were merely capturing that local incidents drive posts
on social media, Facebook and internet outages should not reduce the number of hate
crimes. Instead, they should only reduce social media activity, keeping the number of
anti-refugee incidents unchanged.
3.2. Panel Regression Results
We illustrate the intuition behind our regression framework in Figure 4. The gure
shows a binned scatter plot of anti-refugee attacks and anti-refugee sentiment, split by
the degree of exposure to right-wing social media. Higher refugee salience is associated
with a higher probability of anti-refugee attacks in both sub-samples, but the positive
slope is far more pronounced for towns with an above median AfD user to population
ratio (Panel (a)). Our baseline regression coefcient picks up the difference in slopes
between municipalities with high and low Facebook usage.
Table 2 presents the regression results from estimating Equation (1) with varying
sets of control variables (interacted with refugee salience). The coefcient on the
interaction of local Facebook usage and Germany-wide refugee posts is positive and
highly signicant in all specications. Column 1 shows the panel regressions with the
baseline control variables, which yields a coefcient 0:024 on the interaction term. This
correlation does not appear to be driven by support for the AfD alone: the result holds
although we control for the AfD vote share in the 2017 federal election. This highlights
a distinction between our social media measure and general support for the party.
To get a sense of the magnitudes, consider as a case study the cities of Bochum
and Hannover, which are about one standard deviation apart in the ratio of AfD users
to population (in 1000s) (0:29). Holding average anti-refugee sentiment in our data
constant (84 posts), this means a one standard deviation higher right-wing social media
usage is associated with a 10% higher probability of an anti-refugee incident relative to
the mean. Table D.1 in the online appendix shows that this correlation is largely driven
by cases of assault.
In columns 2 through 6, we introduce a richer set of controls that accounts for
local right-wing attitudes, general media exposure, more socio-economic factors, and
the vote shares of all major parties in the 2017 election (see Table B.3 for an overview of
the control variables). In column 7, we add all interacted controls jointly. The inclusion
of these covariates makes little difference to our main estimate. This is a rst indication
that the correlation between social media exposure and anti-refugee incidents is not
driven by observable municipality differences unrelated to Facebook usage.
Journal of the European Economic Association
Preprint prepared on 5 September 2020 using jeea.cls v1.0.
Downloaded from https://academic.oup.com/jeea/advance-article/doi/10.1093/jeea/jvaa045/5917396 by guest on 28 October 2020
ORIGINAL UNEDITED MANUSCRIPT
Müller and Schwarz Fanning the Flames of Hate 17
0.0
0.5
1.0
1.5
2.0
Probability of refugee attack (in %)
0 50 100 150 200 250
Refugee posts
a) AfD users/Pop. ≥ Median
0.0
0.5
1.0
1.5
2.0
Probability of refugee attack (in %)
0 50 100 150 200 250
Refugee posts
b) AfD users/Pop. < Median
Figure 4. Exposure to refugee sentiment on facebook and hate crimes. This gure plots the average
number of anti-refugee attacks against our measure of anti-refugee sentiment for municipalities below
and above the median of Af D Us er s=P op. Refugee attacks are binned by 20 quantiles of refugee
posts and residualized with respect to population.
Table 2. Baseline correlations — Facebook posts and hate crime.
Additional interacted controls
(1) (2) (3) (4) (5) (6) (7)
Right Socio- 2017 Age
Baseline Wing Media economic vote structure All
controls controls controls controls controls controls controls
AfD users/Pop. Refugee posts 0.024*** 0.020** 0.023** 0.024** 0.021** 0.023** 0.016**
(0.009) (0.008) (0.009) (0.009) (0.009) (0.009) (0.008)
Observations 479,964 479,964 479,964 474,303 479,964 476,856 474,303
R-squared 0.082 0.083 0.082 0.083 0.083 0.083 0.084
Municipalities 4324 4324 4324 4273 4324 4296 4273
Municipality FE Yes Yes Yes Yes Yes Yes Yes
Week FE Yes Yes Yes Yes Yes Yes Yes
Baseline controls [8] Posts Yes Yes Yes Yes Yes Yes Yes
Right-wing controls [4] Posts Yes Yes
Media controls [4] Posts Yes Yes
Socio-econ. controls [4] Posts Yes Yes
Election controls [7] Posts Yes Yes
Age controls [4] Posts Yes Yes
Notes: This table presents the estimated coefcients from a regression of hate crimes against refugees on the interaction
of local social media usage and anti-refugee sentiment as in Equation (1). The dependent variable is a dummy for
the incidence of a refugee attack. AfD users/Pop. is the ratio of people with any activity on the AfD Facebook page
to population. Refugee posts is the Germany-wide number of posts on the AfD’s Facebook wall containing the word
refugee (“Flüchtling”). All control variables are interacted with the Ref ugee post s measure; see text for a description
of the controls. Robust standard errors in all specications are clustered by municipality. ***, **, and * indicate
statistical signicance at the 0.01, 0.05, and 0.1 levels, respectively.
Journal of the European Economic Association
Preprint prepared on 5 September 2020 using jeea.cls v1.0.
Downloaded from https://academic.oup.com/jeea/advance-article/doi/10.1093/jeea/jvaa045/5917396 by guest on 28 October 2020
ORIGINAL UNEDITED MANUSCRIPT
Müller and Schwarz Fanning the Flames of Hate 18
3.3. Quasi-Experimental Evidence: Facebook and Internet Outages
To isolate the importance of social media, we next draw on internet and Facebook
outages as sources of quasi-experimental variation. To count as a severe internet
disruption, our baseline measure has to fulll two criteria: (1) it has to last at least
24 hours, and (2) it has to affect a signicant part of the population, i.e. be in the
top quartile of reported internet disruptions per capita, which vary by municipality
and week (see section Section 2 for more details). This gives us 313 severe internet
outages.17
Internet Outages. Are local internet outages severe enough to decrease a
municipality’s exposure to social media? We investigate this question by using a
sample of posts from the AfD Facebook page for which we know the users’ locations.18
Figure 5a plots the local number of posts against the intensity of local internet outages.
Local Facebook activity falls with outage intensity and is close to 0 as soon as we
observe more than 0.25 outage reports per 10,000 inhabitants. Figure C.3 shows that
we observe signicantly fewer posts and comments on Facebook for municipalities that
experience an internet disruption. These results lend credence to the idea that exposure
to social media content is reduced in the affected municipalities and not compensated
by users accessing Facebook with their mobile phones.
If internet outages indeed reduce local social media exposure, we would expect
them to mediate the capacity of social media to propagate anti-refugee incidents. As
described in Section 3.1, we test this hypothesis by interacting the main terms of
interest AfD Users/PopiRefugee Poststwith Internet Problemsit , our dummy for
severe internet disruptions. We graphically illustrate the results in Figure 5b. The
binned scatter plot is almost identical to Figure 4, except that we plot a separate slope
for municipalities that experience an internet outage. This reveals a striking pattern:
while anti-refugee attacks increase with anti-refugee posts, this relationship disappears
in municipalities that experience an internet outage. This holds true for municipalities
with high and low Facebook usage.
Figure 5b implies that internet outages have a substantial attenuating effect.
Consider the pattern in panel (a). Without outages, there is a strong correlation of
refugee posts and attacks. During outages, the correlation is essentially zero. This
means that the outage effect is larger than the baseline estimate for AfD Users/Pop.
Refugee posts, which is given by the slope difference of the dotted lines in panels (a)
and (b). We interpret this as evidence that cutting of users from social media completely
has large effects.
17. In the online appendix, we show our results are robust to alternative denitions. We also exploit the
eight major Facebook outages, which only vary by week. We discuss the results and their interpretation in
turn.
18. These posts and comments are a sub-sample by users who publicly disclosed their location in their
Facebook proles.
Journal of the European Economic Association
Preprint prepared on 5 September 2020 using jeea.cls v1.0.
Downloaded from https://academic.oup.com/jeea/advance-article/doi/10.1093/jeea/jvaa045/5917396 by guest on 28 October 2020
ORIGINAL UNEDITED MANUSCRIPT
Müller and Schwarz Fanning the Flames of Hate 19
(a) Internet Outages Reduce Local Facebook Activity
0
10
20
30
0.00 0.05 0.10 0.15 0.20 0.25
Internet outage intensity (reports per 1,000 inhabitants)
Number of local posts
(b) Internet Outages Reduce Local Anti-Refugee Incidents
0.0
0.5
1.0
1.5
2.0
Probability of refugee attack (in %)
0 50 100 150 200 250
Refugee posts
a) AfD users/Pop. ≥ Median
0.0
0.5
1.0
1.5
2.0
Probability of refugee attack (in %)
0 50 100 150 200 250
Refugee posts
b) AfD users/Pop. < Median
No Outage Outage
Figure 5. Quasi-experimental results from internet outages. Panel (a) shows a binned scatter plot
of local posts on the AfD Facebook page as a function of the reports on internet outages in a given
week. Panel (b) plots the average number of anti-refugee attacks against our measure of anti-refugee
sentiment for municipalities above and below the median of Af D U se rs =Pop. Refugee attacks are
binned by 20 quantiles of refugee posts. We additionally split towns by whether they experience an
internet outage in a given week (gray squares). The number of anti-refugee attacks is residualized
with respect to population; hence, the number of attacks can be slightly below 0 in some bins.
We next estimate versions of Equation (2) and report the regression results
in Table 3. Column 1 shows that internet outages reduce anti-refugee violence. The
coefcient of 0:003 implies that, during such outages, the probability of a refugee
attack is 53% lower relative to the dependent variable mean (0:006). In Figure 6, we
investigate the timing of this drop in incidents. Because the outages are relatively rare
in the municipality-week panel, the estimates are necessarily noisy. Nonetheless, we
Journal of the European Economic Association
Preprint prepared on 5 September 2020 using jeea.cls v1.0.
Downloaded from https://academic.oup.com/jeea/advance-article/doi/10.1093/jeea/jvaa045/5917396 by guest on 28 October 2020
ORIGINAL UNEDITED MANUSCRIPT
Müller and Schwarz Fanning the Flames of Hate 20
can see a reduction in anti-refugee incidents that is sharply concentrated in the week
of the internet outage.
-0.02
-0.01
0.00
0.01
0.02
-2 -1 0 1 2
Weeks since major internet outage
Change in probability of anti-refugee incident
Figure 6. Internet outage event study . This gure plots estimates the estimates for ıfrom the event
study regression At t ack sit DP2
tD2ıwDtOut ageit CF ix ed Eff ec ts C"i t , where Outage
refers to internet outages in municipality iin week t. 95% condence intervals are based on standard
errors clustered by municipality.
Column 2 in Table 3 implies that this effect is driven by periods of high
sentiment; it may also be driven by areas with many AfD Facebook users (column
3) but the coefcient is not statistically signicant. In columns 4 through 6, we
estimate the full triple-difference model. Here, we estimate the effect of outages in
areas with high social media use at times of high anti-refugee sentiment. The estimates
suggest that internet problems reduce social media’s impact on anti-refugee violence.
While the coefcient of refugee posts and social media exposure is similar to our
baseline correlations, the triple interaction term with internet outages is negative
and statistically signicant in all three specications. Quantitatively, internet outages
appear to mitigate the entire effect of social media. In line with the graphical evidence
in Figure 5b, we nd that the triple interaction coefcient is larger than the baseline
coefcient. Put differently, for a given level of anti-refugee sentiment, there are fewer
attacks in municipalities with high Facebook usage during an internet outage than in
municipalities with low Facebook usage without an outage.
Could it be that the effect of internet outages is merely coincidental? As an
alternative way of assessing statistical signicance, we perform a randomization test.
Instead of the actual internet disruptions, we randomly dene 313 municipality-week
pairs as placebo outages. We then estimate the same regression using 500 different sets
of placebo outages. This allows us to evaluate the probability of nding a statistically
signicant coefcient in our dataset. Using this procedure, we nd that more than 99%
Journal of the European Economic Association
Preprint prepared on 5 September 2020 using jeea.cls v1.0.
Downloaded from https://academic.oup.com/jeea/advance-article/doi/10.1093/jeea/jvaa045/5917396 by guest on 28 October 2020
ORIGINAL UNEDITED MANUSCRIPT
Müller and Schwarz Fanning the Flames of Hate 21
Table 3. Local internet outages and social media transmission.
(1) (2) (3) (4) (5) (6)
Baseline Interaction
AfD users/Pop. Refugee posts 0.024*** 0.016** 0.016**
(0.009) (0.008) (0.008)
AfD users/Pop. Posts Outage -0.181*** -0.184*** -0.172***
(0.058) (0.058) (0.057)
Outage Interaction
Outage -0.003*** -0.000 -0.003** -0.001 -0.002 -0.007
(0.001) (0.001) (0.001) (0.002) (0.002) (0.008)
Refugee posts Outage -0.005*** -0.000 0.001 0.000
(0.001) (0.002) (0.002) (0.002)
AfD users/Pop. Outage -2.685 4.441 4.455 4.391
(3.464) (4.384) (4.054) (4.058)
Internet Usage Interaction
Share broadband access Outage -0.000
(0.000)
Internet domains/Pop. Outage 0.021*
(0.012)
Mobile Broadband Speed Outage 0.000
(0.000)
Observations 479,964 479,964 479,964 479,964 474,303 474,303
R-squared 0.082 0.082 0.082 0.082 0.084 0.084
Municipalities 4324 4324 4324 4324 4273 4273
Municipality FE Yes Yes Yes Yes Yes Yes
Week FE Yes Yes Yes Yes Yes Yes
Baseline controls [8] Posts Yes Yes Yes Yes Yes Yes
All other controls [22] Posts Yes Yes
Notes: This table presents the estimated coefcients from a regression of hate crimes against refugees on the interaction
of local social media usage and anti-refugee sentiment as in Equation (1). The dependent variable is a dummy for the
incidence of a refugee attack. AfD users/Pop. is the ratio of people with any activity on the AfD Facebook page to
population. Refugee posts is the Germany-wide number of posts on the AfD’s Facebook wall containing the word
refugee (“Flüchtling”). Internet outages are dened as municipality-weeks that are in the top quartile of the ratio
of reported internet outages to population. The coefcient of “Refugee posts Outage” is multiplied by 100 for
readability. Columns 1-4 include the baseline controls. Columns 5 and 6 include all controls as in column 7 of table
2, interacted with Refugee posts (unreported). Column 6 further adds the interaction of broadband access and internet
domains/pop. with local internet outages. Robust standard errors in all specications are clustered by municipality.
***, **, and * indicate statistical signicance at the 0.01, 0.05, and 0.1 levels, respectively.
of the placebo triple interaction coefcients exhibit a lower t-statistic than our estimate.
Our ndings are thus unlikely to be purely coincidental. We show the full distribution
of t-statistics from this randomization test in Figure C.6a in the online appendix.
The identifying assumption for internet outages in our framework is that they
only have an effect on anti-refugee hate crime through the reduced exposure to social
media. Could it be that we observe reduced hate crimes because users are cut off from
the internet generally, not from social media in particular? Two pieces of evidence
support the idea that we capture a social media channel.
First, when we include interactions of internet disruptions with measures of
internet usage (broadband access, per capita internet domains, mobile internet access),
our main coefcient is unaffected (see column 6 in Table 3). The coefcients of the
internet usage interactions are generally statistically insignicant or have the opposite
of the expected sign. This is at least some indication that we are not merely capturing
Journal of the European Economic Association
Preprint prepared on 5 September 2020 using jeea.cls v1.0.
Downloaded from https://academic.oup.com/jeea/advance-article/doi/10.1093/jeea/jvaa045/5917396 by guest on 28 October 2020
ORIGINAL UNEDITED MANUSCRIPT
Müller and Schwarz Fanning the Flames of Hate 22
general internet usage. It also suggests that our ndings are unlikely to capture that
people are busy xing internet access problems. If we were merely capturing such
displacement effects, one would expect it to more strongly affect people in areas with
high internet usage, which does not seem to be the case in the data. Second, after
including the other interaction terms in columns 4 through 6, the coefcient on internet
outages is no longer statistically signicant. This result also supports the idea that
internet outages reduce hate crime by limiting access to social media.
Another concern could be that hate crimes are less likely to be reported during
internet outages. We believe this is unlikely to explain our ndings because we analyze
incidents that happened years in the past. While internet outages might hamper the
ow of information, it seems highly unlikely that incidents such as assault or property
damages are never reported due to a temporary internet disruption. As further evidence,
we limit our analysis to ofcial reports by the police or the German parliament, for
which social media reporting is an unlikely concern. This yields similar results (see
column 1 of Table C.4).
We also run a number of tests to rule out that our Germany-wide measure
of refugee posts is affected by local internet outages. As stated above, this appears
unlikely because we focus on local disruptions to the internet; Table C.3 in the
online appendix shows that the total number of internet outages in a given week
is uncorrelated with the total number of Facebook posts. The outage results are
also robust to using a leave-one-out measure of refugee posts (column 2), Germany-
wide posts in the previous week (column 3), and an alternative measure based on
Google search intensity for the word refugee (Flüchtling) in column 4. The implied
magnitudes are almost equivalent.19 This suggests that the outage effect is driven by
exposure rather than the production of anti-refugee content. In Table C.6, we show
additional robustness checks for alternative transformations of the dependent variable.
The ndings remain robust throughout. Table C.7 shows that the results also hold using
alternative denitions of the outage dummy.
Facebook Outages. As further evidence for the social media transmission
mechanism, we use eight major Germany-wide Facebook outages as a source of
exogenous variation specic to social media access. Table C.2 outlines the details
of each of the eight outages and links to relevant press reports. By denition, these
outages are Facebook-specic and therefore do not affect other potential channels of
online transmission.
Table C.3 in the online appendix shows that these outages are large enough to
disrupt weekly activity on right-wing social media. Column 1 and 2 show that, during
weeks with Facebook outages, there are on average 11% fewer new total posts and 24%
19. To see this, consider the effect implied by dividing the triple interaction coefcients by the standard
deviation of these salience metrics. This suggests that internet outages have a mediating effect of
9:6
,
10:5
,
and 11:0 for the AfD posts about refugees, the leave-one-out measure, and Google trends, respectively.
Journal of the European Economic Association
Preprint prepared on 5 September 2020 using jeea.cls v1.0.
Downloaded from https://academic.oup.com/jeea/advance-article/doi/10.1093/jeea/jvaa045/5917396 by guest on 28 October 2020
ORIGINAL UNEDITED MANUSCRIPT
Müller and Schwarz Fanning the Flames of Hate 23
fewer posts about refugees on the AfD page.20 There is no evidence of such an effect
in the week before. Column 5 shows that Facebook outages are also uncorrelated with
the total number of weekly internet disruptions (tD 0:41).
We next present the results of interacting Facebook disruptions analogous
to the internet outages in Table 4. The results again reveal a clear pattern. The
coefcient of 0:001 in column 1 shows that the probability of an anti-refugee
incident is around 18% lower in weeks with major Facebook outages (relative to the
unconditional probability of an attack). Figure C.4 suggests that the timing of this
effect is concentrated in the week of the Facebook outage, without signicant effects
in the week before or after the outage. Because we solely rely on the weekly variation
from the few major Facebook outages, the estimates are noisier than those for internet
outages. Column 2 shows that, intuitively, this effect is also larger in areas with many
users on the AfD Facebook page. The coefcient of 2:222 suggests that Facebook
outages reduce the probability of a hate crime by 12% more for a one standard deviation
increase in AfD users / Pop.21 This is additional evidence that social media per se might
affect hate crimes.
Next, we introduce the triple interaction of Facebook outages with social
media usage and our refugee salience measure. The triple interaction is negative and
statistically signicant in all three specications in columns 3 through 5. Quantitatively,
we nd that Facebook disruptions fully undo the baseline correlation of refugee attacks
and exposure to social media sentiment. For example, consider that the coefcient
of AfD users/Pop. and Refugee Posts is 0:027 in column 4 but 0:04 on the triple
interaction. This implies that, in weeks of major Facebook outages, heightened refugee
sentiment is not associated with a differential increase of anti-refugee attacks in
municipalities with higher Facebook usage.
It is worth noting that we would expect the Facebook outage coefcients to differ
in magnitude from the internet outage coefcients. This is because Facebook outages
eliminate the differential exposure between areas with high and low social media
usage to anti-refugee posts. In contrast, internet outages further exploit variation within
municipalities. Because within-municipality variation induced by internet outages
appears to matter more in our setting, we nd smaller coefcients for Facebook
outages.
We again perform a randomization test to assess the statistical signicance of
the Facebook outage results. We randomly assign placebo Facebook outages to eight
weeks in our data, excluding the weeks in which we identied Facebook outages. We
then estimate the same regression using 500 different sets of placebo outages. Using
this procedure, we nd that 92% of the placebo triple interaction coefcients exhibit
smaller t-statistics. We show the full distribution of t-statistics from this randomization
20. The average number of refugee posts in the time series is around
84
. The coefcient estimate of
19:880 implies an effect of Facebook outages on posts of 19:880=84 0:24 relative to the mean.
21. In unreported results, we also nd that the interaction of Facebook outages with refugee posts has a
statistically signicant negative coefcient.
Journal of the European Economic Association
Preprint prepared on 5 September 2020 using jeea.cls v1.0.
Downloaded from https://academic.oup.com/jeea/advance-article/doi/10.1093/jeea/jvaa045/5917396 by guest on 28 October 2020
ORIGINAL UNEDITED MANUSCRIPT
Müller and Schwarz Fanning the Flames of Hate 24
test in Figure C.6b in the online appendix. This conrms that our ndings are unlikely
to be a matter of coincidence.
Taken together, the evidence here suggests that the relationship of anti-refugee
sentiments online and hate crimes is attenuated by Facebook and internet outages.
These results are most consistent with a causal propagation effect of social media.
Table 4. Facebook outages and social media transmission.
(1) (2) (3) (4) (5) (6)
Baseline Interaction
AfD users/Pop. Refugee posts 0.027*** 0.027*** 0.021** 0.021**
(0.010) (0.010) (0.009) (0.009)
AfD users/Pop. Posts Outage -0.040* -0.040* -0.046** -0.046**
(0.021) (0.021) (0.022) (0.022)
Additional Outage Coeffcients
Outage -0.001***
(0.000)
AfD users/Pop. Outage -2.222* 1.164 1.164 1.367 3.230
(1.273) (1.833) (1.833) (1.862) (1.969)
Observations 479,964 479,964 479,964 479,964 474,303 474,303
R-squared 0.079 0.082 0.082 0.082 0.084 0.084
Municipalities 4324 4324 4324 4324 4273 4273
Municipality FE Yes Yes Yes Yes Yes Yes
Week FE Yes Yes Yes Yes Yes
Baseline controls [8] Posts Yes Yes Yes Yes Yes Yes
All other controls [22] Posts Yes Yes
All controls [30] Outages Yes
Notes: This table presents the estimated coefcients from a regression of hate crimes against refugees on the
interaction of local social media usage and anti-refugee sentiment as in Equation (1). The dependent variable is
a dummy for the incidence of a refugee attack. AfD users/Pop.is the ratio of people with any activity on the AfD
Facebook page to population. Refugee posts is the Germany-wide number of posts on the AfD’s Facebook wall
containing the word refugee (“Flüchtling”). Facebook outages refer to weeks in which Facebook experienced
considerable disruptions; see the online appendix for more details on how these are dened. Note that the other
interaction terms Out age,R ef ugee p osts and O utage Ref ugee pos ts are absorbed by the week xed
effects in columns 3-5. Columns 1-3 include the baseline controls. Columns 4 and 5 include all controls as in
column 7 of table 2, interacted with Refugee posts. Column 5 adds the interaction of these control variables
with Facebook outages. Robust standard errors in all specications are clustered by municipality. ***, **, and
* indicate statistical signicance at the 0.01, 0.05, and 0.1 levels, respectively.
In the online appendix, we conduct additional robustness exercises for our
outage results. In Table C.5, we show a range of different standard errors. We also
assess our results’ robustness to different transformations of the refugee attack variable
and estimation methods in Table C.6. Our results are similar when we use the number
of attacks, log(1+refugee attacks) or the ratio of refugee attacks to asylum seekers
as dependent variable. In all cases, the estimated coefcients are highly statistically
signicant.
3.4. Additional Results
Other Posts on the AfD Facebook Page. If the channel we uncover is indeed
specic to refugees, we would expect a weaker correlation between refugee attacks
Journal of the European Economic Association
Preprint prepared on 5 September 2020 using jeea.cls v1.0.
Downloaded from https://academic.oup.com/jeea/advance-article/doi/10.1093/jeea/jvaa045/5917396 by guest on 28 October 2020
ORIGINAL UNEDITED MANUSCRIPT
Müller and Schwarz Fanning the Flames of Hate 25
and posts about other topics on the AfD Facebook page. We test this hypothesis in
Table D.2, where we plot the baseline estimation with refugee posts in column 1 for
convenience. We also report coefcients for standardized post measures (with a mean
of zero and standard deviation of one) in square brackets to compare coefcient sizes
across the different posts. Next, we estimate Equation (1) using all posts except those
containing the word refugee (“Flüchtling”) in column 2. The estimate is statistically
indistinguishable from zero. We also repeat our baseline test using posts containing the
words “Muslim”, “Islam”, or “EU”—the latter is motivated by the AfD’s long-standing
criticism of the European Union. For all these terms, we nd no signicant relationship
between the number of posts and the number of attacks; all estimated coefcients are
considerably smaller in standardized terms compared to the baseline measure. This
shows the specicity of our refugee measure: the correlation we capture does not
appear to be an artifact of general anti-minority sentiment, but rather a predictable
result of increased animosities towards refugees on social media in particular weeks.
Intensive Margin of Facebook Usage. If social media works as the propagating
mechanism for hate speech, we would also expect its effect to increase with
how frequently users interact with the AfD Facebook page. We explore this issue
empirically in Table D.3, where we interact our main interaction term with the total
number of local posts on the AfD wall and the number of comments and likes on AfD
posts, all scaled over the number of AfD users in a municipality.22 These measures of
usage intensity are not systematically correlated with local Facebook penetration, city
size, or population density. As such, they create additional variation in social media
engagement across towns.
The results suggest that local engagement on Facebook matters: all three
triple interaction terms are positive and statistically signicant. Consistent with the
hypothesis that social media enables hateful sentiment to spread, a higher reach per
AfD user increases the correlation of social media exposure with hate crimes. These
interactions work on top of our baseline interaction term, which remains similar in
magnitude and highly statistically signicant throughout. The smallest coefcient on
the triple interaction term of 0:001 in column 3 implies that a one standard deviation
increase in likes per user (around 12) increases the baseline coefcient by 25%.23
Distracting News Events. As an additional piece of analysis, we investigate the role of
news shocks on the transmission of online hate speech to real-world actions. We build
on the evidence in Durante and Zhuravskaya (2018), who show that the Israeli army is
more likely to strike against Palestinian targets when US media outlets are distracted by
other news events. In our case, we hypothesize that other important news events might
22. Note that we can only construct these measures on the intensive margin of municipalities where we
can identify at least one AfD user. Our baseline results also hold in this sub-sample, which we show in
Table E.2 in the online appendix.
23. To see this, consider that the total implied estimate including interaction is calculated as
0:001 12
0:012, which is about 25% than the baseline coefcient of 0:049.
Journal of the European Economic Association
Preprint prepared on 5 September 2020 using jeea.cls v1.0.
Downloaded from https://academic.oup.com/jeea/advance-article/doi/10.1093/jeea/jvaa045/5917396 by guest on 28 October 2020
ORIGINAL UNEDITED MANUSCRIPT
Müller and Schwarz Fanning the Flames of Hate 26
distract people from the topic of refugees. This is somewhat analogous to Facebook
outages in that we exploit additional exogenous weekly variation: if major news events
act as a distraction, they should reduce the correlation of exposure to refugee salience
with hate crimes.
To measure these news shocks, we obtain Google Trends data on weekly search
interest on the terms “Brexit”, “Trump”, and “UEFA Euro 2016’. Figure D.1 shows that
these spike around the respective events. In Table D.4, we show that they are indeed
associated with a crowding out of refugee salience: the share of posts about refugees
is markedly lower during these key events. As an example, the spike in search interest
for Brexit (100 on the Google search index) is associated with an almost 30% drop in
the share of refugee posts (relative to the mean).
We next investigate whether, as a result, refugee salience has a weaker link
with hate crimes in the weeks these major events attracted particular news attention.
If this is the case, we would expect that these events decrease the correlation of social
media transmission with refugee attacks. As before, we implement this by including
the Google trends measures as a further interaction in our panel regressions.
Table 5 plots the results. For each of the events in columns 1 to 3, we nd a
signicant negative coefcient on the number of anti-refugee incidents for the triple
interaction with distracting news. The negative sign of the coefcient indicates that,
during weeks of major news events, changes in anti-refugee incidents correlate less
with heightened refugee salience. As the salience of other events crowds that of
refugees, there are smaller increases of hate crimes in municipalities with more AfD
social media users.
Journal of the European Economic Association
Preprint prepared on 5 September 2020 using jeea.cls v1.0.
Downloaded from https://academic.oup.com/jeea/advance-article/doi/10.1093/jeea/jvaa045/5917396 by guest on 28 October 2020
ORIGINAL UNEDITED MANUSCRIPT
Müller and Schwarz Fanning the Flames of Hate 27
Table 5. News shock salience and hate crime propagation.
(1) (2) (3)
Brexit Trump UEFA EM 2016
AfD users/Pop. Refugee posts 0.071*** 0.096*** 0.067***
(0.018) (0.022) (0.017)
AfD users/Pop. Posts News shock -0.019** -0.009*** -0.002**
(0.008) (0.003) (0.001)
Observations 495,726 495,726 495,726
R-squared 0.078 0.079 0.078
Municipalities 4466 4466 4466
Municipality FE Yes Yes Yes
Week FE Yes Yes Yes
Notes: This table presents the estimated coefcients from a regression of hate crimes
against refugees on the interaction of local social media usage and anti-refugee sen-
timent as in Equation (1). The dependent variable is a dummy for the incidence of
a refugee attack. AfD users/Pop. is the ratio of people with any activity on the AfD
Facebook page to population. Refugee posts is the Germany-wide number of posts
on the AfD’s Facebook wall containing the word refugee (“Flüchtling”). The news
shocks refer to the Google searches as indicated in the text. Robust standard errors
in all specications are clustered by municipality. ***, **, and * indicate statistical
signicance at the 0.01%, 0.05%, and 0.1% levels, respectively.
Journal of the European Economic Association
Preprint prepared on 5 September 2020 using jeea.cls v1.0.
Downloaded from https://academic.oup.com/jeea/advance-article/doi/10.1093/jeea/jvaa045/5917396 by guest on 28 October 2020
ORIGINAL UNEDITED MANUSCRIPT
Müller and Schwarz Fanning the Flames of Hate 28
3.5. Differences Between Social Media And Traditional Media
How does social media differ from traditional media? And could such differences
partially explain our results? Existing work has highlighted the ability of users to
self-select and interact on social media (e.g. Schmidt et al. 2017). In the following,
we highlight three aspects of far-right social media in Germany that may make it a
particularly effective transmission mechanism for anti-refugee sentiment compared to
mainstream news sources.
First, Figure 7a shows that the share of content about refugees is consistently
higher on the AfD’s Facebook page compared to traditional news outlets in the Nexis
data. The share of refugee mentions on Facebook is also far more volatile and spikes
coincide more clearly with salient news events like Merkel’s “Wir schaffen das” speech
or the Cologne New Year’s Eve incidents. In both of these examples, the share of
refugee posts on right-wing social media is nearly 100% higher than the share of
news stories on refugees, which is consistent with the idea that the topics discussed
on Facebook are considerably narrower than in traditional media.
In Figure D.3a in the online appendix, we show that this also holds true in a like-
for-like comparison of the share of refugee posts on the AfD’s Facebook page relative
to the Facebook pages of ve major German news outlets. AfD users post twice as
much about refugees compared to the next-ranked newspaper. This suggests that the
narrowness of content is unlikely to be explained only be the editorial constraints (e.g.
space limits in newspapers) of traditional media outlets. Instead, self-selection of like-
minded people into the AfD Facebook page likely also play a role. Combined with
the interactive nature of social media, this result points towards an anti-refugee group
dynamic on the AfD’s Facebook page.
Second, as argued by Sunstein (2017), self-selection of like-minded people can
lead to the expression of more extreme viewpoints. To shed light on this hypothesis
empirically, we compare the full text of news reports about refugees with posts on the
AfD Facebook page. Existing reports on far-right hate speech on social media highlight
three characteristics as typical (see for example Dinar et al. 2016; Kreißel et al. 2018;
Ott and Gür-Seker 2019): (1) a belief to speak for the “true will” of the people, i.e. the
in-group (citizens) compared to the out-group (refugees); (2) an opposition to “elites”,
in particular politicians and the media, who supposedly mislead or betray the people
in an undemocratic way; and (3) a legitimization of discrimination against refugees by
highlighting crimes by refugees, an alleged incompatibility of cultural differences, and
negative repercussions for vulnerable “locals” (e.g. women, children or pensions).
We nd evidence for all three of these features of right-wing hate speech on
the AfD’s Facebook page. Our approach is to investigate which words occur with
a higher probability in posts on the AfD page relative to news reports in the Lexis
corpus.24 We lter words using the word stems of the German terms for people, elite,
24. We calculate word probabilities for each corpus by dividing the number of times a word is
mentioned (
Wordi
) by the total number of words in the corpus (
PWordsi
), e.g.
P(WordNews
i/D
Journal of the European Economic Association
Preprint prepared on 5 September 2020 using jeea.cls v1.0.
Downloaded from https://academic.oup.com/jeea/advance-article/doi/10.1093/jeea/jvaa045/5917396 by guest on 28 October 2020
ORIGINAL UNEDITED MANUSCRIPT
Müller and Schwarz Fanning the Flames of Hate 29
(a) Share of Refugee Post over Time
AfD Facebook page
News reports
0.00
0.05
0.10
0.15
0.20
2015w1 2015w26 2016w1 2016w27 2017w1
Share of Posts/News About Refugees
(b) Individual Posting Behavior, by Length of Exposure
0.0
0.2
0.4
0.6
0 20 40 60 80 100
Weeks since first post
Number of refugee posts
Figure 7. Highlighting social media echo chambers. Panel (a) plots the share of posts/reports about
refugees on the AfD Facebook page and major German news outlets from Nexis. Panel (b) plots the
10-week moving average of the number of refugee posts per person as a function of a user’s time
spent on the AfD Facebook page, proxied by the time since the rst post. The shaded area indicates
95% condence intervals.
democratic, press, crime, foreign, culture, refugee, betrayal, and several vulnerable
groups (pensioners, children, women, homeless).
The results of this exercise in Table 6 reveal a clear pattern (see also Table D.5
in the online appendix). As one example, the term “Volksbetrug” (betrayal of the
WordNews
i=PWordsNews
i
. The relative probability is the ratios between the two calculated the two
probabilities, i.e. P(WordFacebook
i/=P(WordNews
i/.
Journal of the European Economic Association
Preprint prepared on 5 September 2020 using jeea.cls v1.0.
Downloaded from https://academic.oup.com/jeea/advance-article/doi/10.1093/jeea/jvaa045/5917396 by guest on 28 October 2020
ORIGINAL UNEDITED MANUSCRIPT
Müller and Schwarz Fanning the Flames of Hate 30
people) is 1715 times more likely to appear on the AfD page than in traditional news
outlets. Criticism of “elites” and the media are also far more frequent. Another main
difference is how often crimes by refugees are discussed, based on the use of loaded
terms like “Flüchtlingskriminalität” (refugee crime). We see expressed fears about
“Fremdkulturen” (foreign cultures) and “Burkafrauen” (burka women). This analysis
clearly shows that far-right ideas that have widely been interpreted as hate speech are
far more pervasive on the AfD page than in traditional media reports.
We nd similar results using a text analysis approach using machine learning.
In particular, we train a L1 regularized logistic regression model classier that predicts
whether a text comes from the AfD Facebook page or a traditional media outlet. The
classier thereby identies the words with the highest predictive ability for posts on the
AfD Facebook page. Figure D.4 shows a word cloud of the 100 words that best separate
social media from traditional media content, based on the model with the highest cross-
validated out-of-sample F1 scores.25 The size of the words represents the magnitude
of the coefcients as a measure of variable importance. Consistent with the ndings in
Table 6, critiques of establishment parties and the economic or social costs of refugees
are among the words that most uniquely identify posts on the AfD page.
Third, we investigate how individuals’ posting behavior varies with the length
of exposure to far-right social media content. We construct a balanced panel of users’
activity on the AfD’s Facebook page. In Figure 7b, we show users’ average number of
posts about refugees since their rst post on the page. To avoid that a changing sample
composition drives our results, we restrict the analysis to the approximately 60% of
users who rst interacted with the AfD page before June 2015 and thus have been
active on it for at least 100 weeks. The results are similar without this restriction.
The frequency of refugee posts strongly increases with users’ duration on
Facebook: within the rst year, the average user on the AfD page goes from close
to zero to posting at least once about refugees every 2 weeks.26 This result suggests
that the AfD page does not merely attract already active Facebook users with right-
wing views, but may increase the willingness of people to express anti-refugee views
over time.
This analysis also highlights an important distinction compared to existing
research on media and violence. Yanagizawa-Drott (2014) Adena et al. (2015), and
DellaVigna et al. (2014) all investigate the effect of nationalistic propaganda in
settings of high ethnic tensions. In our setting, there is no nationalistic anti-minority
propaganda in traditional media outlets. Rather, we nd that social media provides
an alternative forum to exchange and spread extreme rhetoric and viewpoints for the
fringe elements of society.
25. Note that the model was run in German and the words translated by the authors afterwards. For more
details on the machine learning model, see the notes to Figure D.4.
26. The same holds true for the total number of posts (see Figure D.3b in the Online Appendix).
Journal of the European Economic Association
Preprint prepared on 5 September 2020 using jeea.cls v1.0.
Downloaded from https://academic.oup.com/jeea/advance-article/doi/10.1093/jeea/jvaa045/5917396 by guest on 28 October 2020
ORIGINAL UNEDITED MANUSCRIPT
Müller and Schwarz Fanning the Flames of Hate 31
Table 6. Relative word frequencies on the AfD Facebook page.
Rank Word Translation Relativ prob.
Panel A: Flücht (refugee)
1 Flüchtlingsenklaven refugee enclave 780
2 Flüchtlingslüge refugee lie 693
3 Flüchtlingsirrsinn refugee insanity 650
4 Flüchtlingsmaa refugee maa 520
5 Flüchtlingsbefürworter refugee supporter 520
Panel B: Krimi (crime)
1 Regierungskriminalität goverment crime 1300
2 Diskriminierungsgesetze anti-discrimination laws 520
3 Schwerstkriminellen dangerous criminals 260
4 Fluechtlingskriminalität refugee crimes 260
5 Kriminalitätssteigerung increase in crime 260
Panel C: Presse (media)
1 Freie Presse free press 390
2 Propagandapresse propaganda press 260
3 Presseempfang press meeting 260
4 Meinungspresse opinionated media 260
5 Nazipresse nazi media 260
Panel D: Volk (people)
1 Volksbetrug betrayal of the people 1715
2 volksfeindlich hostile to the people 780
3 volksverdummenden brainwashing the people 520
4 Volksverhetzungsparagraphen law against incitement 520
5 Volksprotesten protest by the people 260
Panel E: Verrat (betrayal)
1 Volksverrats betrayal of the people 130
2 Vaterlandsverrat betrayal of the fatherland 43
3 Volksverrat betrayal of the people 43
4 Hochverrat high treason 36
5 verratenen betrayed 32
Notes: This table plots the relative probability of words mentioned on the AfD Facebook page
compared to reports by major German news outlets on Nexis. We report the results by groups
of word stems identied as likely to reecting right-wing hate speech on social media by
previous work in Dinar et al. (2016).
3.6. Mechanisms
In theory, multiple mechanisms could be consistent with social media playing a
propagating role in real-life hate crimes. We discuss four mechanisms: information
exchange, persuasion, collective action, and local spillovers. We provide suggestive
evidence that collective action and local spillovers likely play a role in our setting.
First, social media might facilitate the exchange of information. In our
setting, relevant information for potential perpetrators could, for example, include the
Journal of the European Economic Association
Preprint prepared on 5 September 2020 using jeea.cls v1.0.
Downloaded from https://academic.oup.com/jeea/advance-article/doi/10.1093/jeea/jvaa045/5917396 by guest on 28 October 2020
ORIGINAL UNEDITED MANUSCRIPT
Müller and Schwarz Fanning the Flames of Hate 32
locations of refugee homes and meeting points for demonstrations. We analyze the
content of the refugee posts on the AfD Facebook to identify any post that might
contain location information. To do so, we tag posts that either contain a zip code,
mention the word “straße” (street), “weg” (path), “Flüchtlingsheim”, “Asylantenheim”,
“Flüchtlingsunterkunft” (all three translate to refugee home) or refer to a name of a
German town or village.27 We then manually check the content of tagged posts. This
analysis suggests that while some locations like Berlin and Cologne are frequently
mentioned in the posts as references to politicians and crimes committed by refugees,
we nd no mention about specic local information. We found no instance of zip codes
or exact addresses. It hence appears unlikely that this channel is the primary driver
behind our ndings.
A second mechanism could be a persuasion channel, implying that social media
persuades potential perpetrators that refugees may be dangerous or undeserving, which
may then push some people over the edge. We believe that the timing in our setting
makes this channel unlikely. In contrast to other work in Müller and Schwarz (2018)
and Bursztyn et al. (2019), we focus entirely on high-frequency variation in social
media posts and refugee violence. To the extent that social media changes people’s
attitudes, this is unlikely to happen in a single week and revert back after anti-refugee
salience has subsided. This is particularly true for the results on Facebook and internet
outages: it seems unlikely that being cut off from social media during such disruptions
reduces hate crimes because potential perpetrators become less xenophobic for a single
week.
Third, social media could motivate collective action. Existing evidence in
Enikolopov et al. (2016) and Manacorda and Tesei (2020) suggests that social media
and mobile internet increase the incidence of protests. In our setting, users could
coordinate to carry out hate crimes or learn about others’ willingness to carry them
out via social media. To investigate this, we rerun the panel regressions in Equation (1)
but limit refugee attacks to those undertaken by multiple perpetrators.28 In line with
the collective action hypothesis, Table 7 suggests that our panel regression results
are predominantly accounted for by cases with four or more perpetrators. We nd no
relationship for incidents with fewer than 4 perpetrators. Within the sub-sample where
we can identify the number of perpetrators, these attacks account for a similar number
of total incidents compared to the cases with more than 4 perpetrators. Hence, this
nding is unlikely to be the result of limited statistical power.
Fourth, and somewhat relatedly, it could be that social media enables local
spillovers, e.g. through “copy-cat” incidents. This mechanism suggests that potential
perpetrators may use social media to learn about other attacks taking place, which could
inspire them to carry out additional hate crimes. Because friendship networks on social
media are clustered geographically (Bailey et al. 2018), this should be particularly
27. We base the search on a comprehensive list of 2,061 German towns and 11,000 municipalities from
the German statistical ofce, which covers villages with as little as 20 inhabitants.
28. We were able to hand-code the number of perpetrators for 28% of the hate crimes.
Journal of the European Economic Association
Preprint prepared on 5 September 2020 using jeea.cls v1.0.
Downloaded from https://academic.oup.com/jeea/advance-article/doi/10.1093/jeea/jvaa045/5917396 by guest on 28 October 2020
ORIGINAL UNEDITED MANUSCRIPT
Müller and Schwarz Fanning the Flames of Hate 33
Table 7. Mechanism — Anti-refugee incidents, by number of perpetrators.
(1) (2) (3) (4)
Known
perp. sample 1 perp. <4 perp. 4 perp.
AfD users/Pop. Refugee posts 0.010** 0.003 0.004 0.007**
(0.005) (0.002) (0.003) (0.003)
Observations 479,964 479,964 479,964 479,964
R-squared 0.081 0.037 0.046 0.055
Municipalities 4,324 4,324 4,324 4,324
Share of attacks 1 0.245 0.494 0.534
Mean of DV 0.002 0.000 0.001 0.001
Municipality FE Yes Yes Yes Yes
Week FE Yes Yes Yes Yes
Baseline controls [8] Posts Yes Yes Yes Yes
Notes: This table presents the estimated coefcients from a regression of hate crimes
against refugees on the interaction of local social media usage and anti-refugee
sentiment as in Equation (1), where we vary the denition of the dependent variable
based on the number of perpetrators. All control variables are interacted with the
Ref ugee p ost s measure. Robust standard errors in all specications are clustered by
municipality. ***, **, and * indicate statistical signicance at the 0.01, 0.05, and 0.1
levels, respectively.
pronounced for attacks happening nearby. We thus again rerun the panel regressions
in Equation (1) but now include a dummy variable if neighboring municipalities
experience an attack in a given week.29
Table D.6 suggests that hate crimes happening in the same week nearby are
associated with more anti-refugee incidents. This correlation strongly interacts with
the popularity of right-wing social media, particularly when anti-refugee sentiment is
elevated. In other words, having an attack in a neighbouring municipality is associated
with a stronger correlation of exposure to right-wing social media and the probability
of an anti-refugee incident.30
Overall, our results appear to be most consistent with the idea that short-run
bursts in anti-refugee sentiment on social media can translate into real-life hate crimes
by enabling coordination online, both through group actions and local spillovers.
29. This is akin to the common correlated effects (CCE) estimator proposed by Pesaran (2006) to hold
common shocks constant.
30. Note that, although they are suggestive, we do not interpret these estimates as causal “peer effects”,
because we cannot distinguish them from common shocks (see Manski 1993).
Journal of the European Economic Association
Preprint prepared on 5 September 2020 using jeea.cls v1.0.
Downloaded from https://academic.oup.com/jeea/advance-article/doi/10.1093/jeea/jvaa045/5917396 by guest on 28 October 2020
ORIGINAL UNEDITED MANUSCRIPT
Müller and Schwarz Fanning the Flames of Hate 34
3.7. How Many Refugee Attacks Are Caused By Online Hate Speech?
We conduct a back-of-the-envelope calculation of how many attacks against refugees
would have taken place with lower anti-refugee sentiment on right-wing social media.
Given that we rely on high-frequency variation, this question is difcult to address. As
our estimates are likely to pick up two separate facets of exposure to social media.
On one hand, it could be that exposure to anti-refugee sentiment on social media
merely affects the exact timing when refugee attacks occur without changing their total
number. On the other hand, the time series of hate crimes and refugee posts on social
media in Figure 2 exhibits prolonged overall increases in the number of anti-refugee
incidents with the onset of the refugee crisis. These increases are not easy to explain
if anti-refugee sentiment exclusively affects the timing of incidents. In our empirical
setting, we cannot distinguish between these possibilities.
Despite this important caveat, we still believe it is instructive to assume social
media indeed increases the number of hate crimes to illustrate the magnitudes of the
results. We calculate the predicted number of attacks, based on the coefcient estimate
of 0:024 from a regression with the baseline control variables (see column 1 in Table 2).
Multiplying this coefcient with AfD Users/Pop. and Ref ugee pos t s gives us the
estimated effect on anti-refugee attacks. We sum over all observations to get the total
predicted number of anti-refugee attacks as a result of social media. This calculation
implies that in absence of social media transmission on social media would result in
289 (10%) fewer anti-refugee incidents.
4. Conclusion
Social media has become a powerful tool for sharing and disseminating information. In
this paper, we investigate whether social media can play a role in propagating violent
hate crimes. Our ndings suggest that social media has not only become a fertile soil for
the spread of hateful ideas but also motivates real-life action. By combining detailed
local data on Facebook usage with user-generated content, we can shed light on the
link between online posts and anti-refugee incidents in Germany. Plausibly exogenous
variation in disruptions to users’ Facebook or internet access supports the view that
some of the correlations we document reect a causal effect.
Existing research shows local cultural attitudes towards foreigners are
enormously persistent (e.g. Becker and Pascali 2019; Becker et al. 2016; Voigtlander
and Voth 2012, 2015). We extend this literature by showing that volatile, short-lived
bursts in sentiment within a given location have substantial effects on people’s behavior
and that social media may play a role in their propagation. Our ndings are particularly
timely in light of recent policy debates about whether and how to “regulate” hate speech
on social media. Such legislation may come at a high price: since the lines between
what constitutes free speech and hate speech can be blurred, regulation can open the
door to censorship. Our work does, however, suggest that policymakers ignore online
hate speech at their peril. Future research should investigate effective ways to tackle
Journal of the European Economic Association
Preprint prepared on 5 September 2020 using jeea.cls v1.0.
Downloaded from https://academic.oup.com/jeea/advance-article/doi/10.1093/jeea/jvaa045/5917396 by guest on 28 October 2020
ORIGINAL UNEDITED MANUSCRIPT
Müller and Schwarz Fanning the Flames of Hate 35
online hate speech. By quantifying the extent of the challenge, our paper takes a rst
step towards identifying potential harm arising from extended social media usage.
References
Adena, Maja, Ruben Enikolopov, Maria Petrova, Veronica Santarosa, and Ekaterina Zhuravskaya
(2015). “Radio and the Rise of The Nazis in Prewar Germany.” The Quarterly Journal of
Economics, 130(4), 1885–1939.
Alesina, Alberto and Eliana La Ferrara (2005). “Ethnic Diversity and Economic Performance.”
Journal of Economic Literature, 43, 721–761.
Bailey, Michael, Rachel Cao, Theresa Kuchler, Johannes Stroebel, and Arlene Wong (2018). “Social
Connectedness: Measurement, Determinants, and Effects.” Journal of Economic Perspectives,
32(3), 259–80.
BAMF (2016). “Aktuelle Zahlen zu Asyl.” Bundesamt für Migration und Flüchtlinge.
Barberá, Pablo (2014). “How Social Media Reduces Mass Political Polarization: Evidence from
Germany, Spain, and the US.” Job Market Paper, New York University, 46.
Bartik, Timothy J. (1991). Who Benefits from State and Local Economic Development Policies?
No. wbsle in Books from Upjohn Press, W.E. Upjohn Institute for Employment Research, URL
https://ideas.repec.org/b/upj/ubooks/wbsle.html.
BBC (2017). “Social Media Warned to Crack Down on Hate Speech.”
Becker, Sascha O. and Luigi Pascali (2019). “Religion, Division of Labor, and Conict: Anti-
semitism in Germany over 600 Years.” American Economic Review, 109(5), 1764–1804.
Becker, Sascha O., Steven Pfaff, and Jared Rubin (2016). “Causes and Consequences of the Protestant
Reformation.” Explorations in Economic History, 62, 1 – 25.
Bessi, Alessandro, Fabiana Zollo, Michela Del Vicario, Antonio Scala, Guido Caldarelli, and Walter
Quattrociocchi (2015). “Trend of Narratives in the Age of Misinformation.” PLOS ONE, 10(8),
1–16.
Bhuller, Manudeep, Tarjei Havnes, Edwin Leuven, and Magne Mogstad (2013). “Broadband Internet:
An Information Superhighway to Sex Crime?” Review of Economic Studies, 80(4), 1237–1266.
Boxell, Levi, Matthew Gentzkow, and Jesse M. Shapiro (2017). “Greater Internet Use Is Not
Associated With Faster Growth in Political Polarization Among US Demographic Groups.”
Proceedings of the National Academy of Sciences, 114(40), 10612–10617.
Bursztyn, Leonardo, Davide Cantoni, Patricia Funk, and Noam Yuchtman (2017). “Polls, the Press,
and Political Participation: The Effects of Anticipated Election Closeness on Voter Turnout.”
NBER Working Papers 23490, National Bureau of Economic Research, Inc, URL https://ideas.
repec.org/p/nbr/nberwo/23490.html.
Bursztyn, Leonardo, Georgy Egorov, Ruben Enikolopov, and Maria Petrova (2019). “Social Media
and Xenophobia: Evidence from Russia.” Working Paper 26567, National Bureau of Economic
Research, URL http://www.nber.org/papers/w26567.
Card, David and Gordon B. Dahl (2011). “Family Violence and Football: The Effect of Unexpected
Emotional Cues on Violent Behavior.” The Quarterly Journal of Economics, 126(1), 103–143.
Colussi, Tommaso, Ingo E. Isphording, and Nico Pestel (2016). “Minority Salience and Political
Extremism.” IZA Discussion Papers 10417, Institute of Labor Economics (IZA), URL https:
//ideas.repec.org/p/iza/izadps/dp10417.html.
Dahl, Gordon and Stefano DellaVigna (2009). “Does Movie Violence Increase Violent Crime?” The
Quarterly Journal of Economics, 124(2), 677–734.
Del Vicario, Michela, Alessandro Bessi, Fabiana Zollo, Fabio Petroni, Antonio Scala, Guido
Caldarelli, H. Eugene Stanley, and Walter Quattrociocchi (2016). “The Spreading of
Misinformation Online.” Proceedings of the National Academy of Sciences, 113(3), 554–559.
Journal of the European Economic Association
Preprint prepared on 5 September 2020 using jeea.cls v1.0.
Downloaded from https://academic.oup.com/jeea/advance-article/doi/10.1093/jeea/jvaa045/5917396 by guest on 28 October 2020
ORIGINAL UNEDITED MANUSCRIPT
Müller and Schwarz Fanning the Flames of Hate 36
DellaVigna, Stefano, Ruben Enikolopov, Vera Mironova, Maria Petrova, and Ekaterina Zhuravskaya
(2014). “Cross-Border Media and Nationalism: Evidence from Serbian Radio in Croatia.”
American Economic Journal: Applied Economics, 6(3), 103–32.
DellaVigna, Stefano and Eliana La Ferrara (2015). “Economic and Social Impacts of the Media.”
NBER Working Papers 21360, National Bureau of Economic Research, Inc, URL https://ideas.
repec.org/p/nbr/nberwo/21360.html.
DellaVigna, Stefano and Matthew Gentzkow (2010). “Persuasion: Empirical Evidence.” Annual
Review of Economics, 2(1), 643–669.
Dinar, Christina, Theresa Mair, Simone Rafael, Jan Rathje, and Julia Schramm (2016). “Hetze gegen
Flüchtlinge in Sozialen Medien.” Amadeu Antonio Stiftung.
Durante, Ruben and Ekaterina Zhuravskaya (2018). “Attack When the World Is Not Watching? US
News and the Israeli-Palestinian Conict.” Journal of Political Economy, 126(3), 1085–1133.
Eisensee, Thomas and David Strömberg (2007). “News Droughts, News Floods, and U. S. Disaster
Relief.” The Quarterly Journal of Economics, 122(2), 693–728.
Enikolopov, Ruben, Alexey Makarin, and Maria Petrova (2016). “Social Media and Protest
Participation: Evidence from Russia.” CEPR Discussion Papers 11254, C.E.P.R. Discussion
Papers, URL https://ideas.repec.org/p/cpr/ceprdp/11254.html.
Financial Times (2017). “Powerhouse Germany Badly Trailing Rivals in Broadband.”
Fiorina, Morris P. and Samuel J. Abrams (2008). “Political Polarization in the American Public.”
Annual Review of Political Science, 11, 563–588.
Fouka, Vasiliki and Hans-Joachim Voth (2013). “Reprisals Remembered: German-Greek Conict
and Car Sales during the Euro Crisis.” CEPR Discussion Papers 9704, C.E.P.R. Discussion Papers,
URL https://ideas.repec.org/p/cpr/ceprdp/9704.html.
Gabler, N (2016). “The Internet and Social Media Are Increasingly Divisive and Undermining of
Democracy.” Alternet.
Gavazza, Alessandro, Mattia Nardotto, and Tommaso Valletti(2018). “Internet and Politics: Evidence
from U.K. Local Elections and Local Government Policies.” The Review of Economic Studies,
86(5), 2092–2135.
Gentzkow, Matthew (2006). “Television and Voter Turnout.” The Quarterly Journal of Economics,
121(3), 931–972.
Hölig, Sascha and Uwe Hasebrink (2016). Reuters Institute Digital News Survey 2017: Ergebnisse
für Deutschland,Arbeitspapiere des Hans-Bredow-Instituts, vol. Nr. 38. Verlag Hans-Bredow-
Institut, Hamburg.
Jha, Saumitra (2013). “Trade, Institutions, and Ethnic Tolerance: Evidence from South Asia.”
American Political Science Review, 107(4), 806–832.
Kreißel, Philip, Julia Ebner, Alexander Urban, and Jakob Guhl (2018). “Hass auf Knopfdruck:
Rechtsextreme Trollfabriken und das Ökosystem koordinierter Hasskampagnen im Netz.”
Institute for Strategic Dialogue.
Manacorda, Marco and Andrea Tesei (2020). “Liberation Technology: Mobile Phones and Political
Mobilization in Africa.” Econometrica, 88(2), 533–567.
Manski, Charles F. (1993). “Identication of Endogenous Social Effects: The Reection Problem.”
Review of Economic Studies, 60(3), 531–542.
Müller, Karsten and Carlo Schwarz (2018). “From Hashtag to Hate Crime: Twitter and Anti-Minority
Sentiment.” Available at SSRN: https://ssrn.com/abstract=3149103.
New York Times (2016). “How Facebook Warps Our Worlds, By Frank Bruni.”
New York Times (2017a). “How Fiction Becomes Fact on Social Media, By Benedict Carey.”
New York Times (2017b). “Seeking Asylum in Germany, and Finding Hatred, By Ainara Tiefenthäler,
Shane O’neill and Andrew Michael Ellis .”
OECD (2016). “Broadband Statistics.”
Oksanen, Atte, James Hawdon, Emma Holkeri, Matti Näsi, and Pekka Räsänen (2014). “Exposure
to Online Hate Among Young Social Media Users.” Soul of Society: A Focus on the Lives of
Children & Youth, 18, 253–273.
Journal of the European Economic Association
Preprint prepared on 5 September 2020 using jeea.cls v1.0.
Downloaded from https://academic.oup.com/jeea/advance-article/doi/10.1093/jeea/jvaa045/5917396 by guest on 28 October 2020
ORIGINAL UNEDITED MANUSCRIPT
Müller and Schwarz Fanning the Flames of Hate 37
Ott, Christine and Derya Gür-Seker (2019). Rechtspopulismus und Social Media: Wie Wortgebräuche
in Social Media sprachkritisch betrachtet werden können, pp. 279–318. Peter Lang AG, URL
http://www.jstor.org/stable/j.ctvnp0hrq.12.
Pariser, Eli (2011). The Filter Bubble: What the Internet Is Hiding From You. Penguin UK.
Pesaran, M. Hashem (2006). “Estimation and Inference in Large Heterogeneous Panels with a
Multifactor Error Structure.” Econometrica, 74(4), 967–1012.
Pew Research Center (2018). “News Use Across Social Media Platforms 2018.” Tech. rep.
Schmidt, Ana Lucía, Fabiana Zollo, Michela Del Vicario, Alessandro Bessi, Antonio Scala,
Guido Caldarelli, H. Eugene Stanley, and Walter Quattrociocchi (2017). “Anatomy of News
Consumption on Facebook.” Proceedings of the National Academy of Sciences, 114(12), 3035–
3039.
Stephens-Davidowitz, Seth (2014). “The Cost of Racial Animus on a Black Candidate: Evidence
using Google Search Data.” Journal of Public Economics, 118(C), 26–40.
Sunstein, Cass R. (2009). Republic. com 2.0. Princeton University Press.
Sunstein, Cass R. (2017). # Republic: Divided Democracy in the Age of Social Media. Princeton
University Press.
The Guardian (2017). “CPS to Crack Down on Social Media Hate Crime, says Alison Saunders, by
Vikram Dodd.”
Voigtlander, N. and H.-J. Voth (2012). “Persecution Perpetuated: The Medieval Origins of Anti-
Semitic Violence in Nazi Germany.” The Quarterly Journal of Economics, 127(3), 1339–1392.
Voigtlander, Nico and Hans-Joachim Voth (2015). “Nazi Indoctrination and Anti-Semitic Beliefs in
Germany.” Proceedings of the National Academy of Sciences of the United States of America,
112(26), 7931–7936.
Yanagizawa-Drott, David (2014). “Propaganda and Conict: Evidence from the Rwandan Genocide.”
The Quarterly Journal of Economics, 129(4), 1947–1994.
Journal of the European Economic Association
Preprint prepared on 5 September 2020 using jeea.cls v1.0.
Downloaded from https://academic.oup.com/jeea/advance-article/doi/10.1093/jeea/jvaa045/5917396 by guest on 28 October 2020