S1537592714003132 55..73

Anti-Americanism and

Anti-Interventionism in Arabic

Twitter Discourses

Amaney A. Jamal, Robert O. Keohane, David Romney, and Dustin Tingley

Systematic investigation of attitudes expressed in Arabic on Twitter towards the United States and Iran during 2012–13 shows

how the analysis of social media can illuminate the politics of contemporary political discourses and generates an informative

analysis of anti-Americanism in the Middle East. We not only analyze overall attitudes, but using a novel events-based analytical

strategy, we examine reactions to speciﬁc events, including the removal of Mohamed Morsi in Egypt, the Innocence of Muslims

video, and reactions to possible U.S. intervention in Syria. We also examine the Boston Marathon bombings of April 2013, in which

the United States suffered damage from human beings, and Hurricane Sandy, in which it suffered damage from nature. Our

ﬁndings reinforce evidence from polling that anti-Americanism is pervasive and intense, but they also suggest that this animus is

directed less toward American society than toward the impingement of the United States on other countries. Arabic Twitter

discourses about Iran are at least as negative as discourses about the United States, and less ambivalent. Anti-Americanism may be

a speciﬁc manifestation of a more general phenomenon: resentment toward powerful countries perceived as interfering in national

and regional affairs.

ne aspect of glo balism—a state of the world

involving networks of interdepende nce at multi-

continental distances—is what could be called

“social globalism,” entailing long-distance transnational

transmission of ideas , information, and images.

Social

globalism implies discord, since it br ings groups with

different interests and valu es into contact with one

another.

Contemporary social media enable individuals

who identify wit h different groups to express their views

in public in relatively s afe ways. The result is a discordant

set of discourses—contentious and not always deeply

reﬂective, but revealing about values, perspectives, and

emotions of large numbers of people who have politically

relevant views and are ready to express them. We use

the plural, “discourses,” because it is not clear that

participants in social media are occupying a common

public sphere, speaking to one another. There may be

distinct discourses, with distinct populations, speaking to

A permanent link to supplementary materials provided by the authors precedes the references section.

Amaney A. Jamal is the Edward S. Sanford Professor of Politics at Princeton University ([email protected]). She

specializes on the politics of development and democratization in the Middle East. Robert O. Keohane is Professor of

International Affairs at the Woodrow Wilson School of Public and International Affairs, Princeton University

([email protected]). He is the author of After Hegemony: Cooperation and Discord in the World Political Economy

(1984) and Power and Governance in a Partially Globalized World (2002). He has served as president of the International

Studies Association and the American Political Science Association. David Romney is a doctoral candidate at Harvard University

whose primary research interests are the psychology of intergroup relations, ethnic and religious conﬂict, and the Middle East. His

research interests include experimental political science, social media and the internet, and Southeast Asia (dromney@fas.

harvard.edu). Dustin Tingley is the Paul Sack Associate Professor of Political Economy in the Government Department at

Harvard University. His research interests include international relations, international political economy, experimental

approaches to political science, and statistical methodology ([email protected]). The authors wish to thank, for written

comments, the editor; ﬁve anonymous reviewers; and multiple colleagues: Carla Beth Abdo, Giacomo Chiozza, Henry Farrell,

Peter Katzenstein, Gary King, Marc Lynch, Helen Milner, Rich Nielsen, and Brandon Stewart. We also thank seminar

participants at Harvard University, Princeton University, and NYU. Access to Crimson Hexagon via the Social Research Grant

Program is acknowledged. Harvard University and Princeton University provided support for this research.

doi:10.1017/S1537592714003132

March 2015

Vol. 13/No. 1 55

Articles

one another and only glancing at other communities,

perhaps negatively.

Discourses on social media are typically focused on con-

temporary topics—often events that have just occurred—

and therefore illuminate how people in this partially

globalized world interpret phenomena that are inherently

ambiguous. These discourses expand the public sphere by

enabling ordinary people to comment, in real time and

for a potentially global audience, on world events. They

also provide opportunities for political scientists who are

interested in new and interactive patterns of globalization

to explore them directly, by monitoring them and seeking

to analyze their content. Students of politics have been

quick to observe these discourses, and to analyze them.

One theme of many of these discourses, especially in

the contemporary Middle East, is anti-Americanism. Since

World War II the United States has engaged in extensive

military intervention in the Middle East—aregionin

which anti-American views have become predominant.

Monitoring the Arabic-language social media provides

a fascinating window into the cross-currents of opinion

with respect to the United States in the Middle East, and

therefore indicates how American policy and the policies of

other countries generate animosity, ambivalence, and

contentious discussion. We can learn a great deal about

anti-Americanism, and more broadly about politics in an

era of global social media, by analyzing these discourses.

Social scientists have so ught to analyze whether anti-

Americanism is principally po litical—hating America

because of what it does, particularly its interventions—

or principally social: hating America for what it is and

what it represents.

Some of the best wor k on this subject,

analyzing the 2004 Pew Global Attitudes Survey in detail,

shows both that the Middle East stands out for its negative

views of th e Unit ed States and th at even in the Middle

East, anti-Americanism is differentiated by issue. Atti-

tudes are most negative toward the diffusion of U.S.

customs abroad, toward the war on terror, an d toward the

effects of American foreign policy on inequality between

rich and poor countries. Attitudes toward U.S. political

institutions and popular culture are somewhat negative

but more balanced; views of U.S. science are quite

positive. In general, what Giacomo Chioz za calls “socio-

cultural anti-Americanism” seems to be much les s intense

than political anti-Americanism.

We take advantage here of new technology to analyze

the social media site, Twitter, in order to gain

informationthatisnotbasedonpollsbutonauton-

omous expressions of opinion by individua ls abou t

anti-Americanism in the Middle East. This technology

enablesustoobserveresponsestoeventsonTwitter

and to analyze the resulting discourses. In doing so, we

generate multi ple empirical observations about Arab

attitudes, creating new information not only about

anti-Americanism but also about how Arabic-speaking

publics react to events and to other states in the region,

notably Iran. Our analysis helps to resolve a major

ambiguity in the existing literature on anti-Americanism:

whether Middle Eastern animus toward the United States

is directed toward the nature of its society—what the

United States “is”—or toward the effects of its policies and

social practices on other societies—what the United States

“does.”

The Twitter discourse is highly political and

focused on the impact of the United States abroad more

than on criticisms of American society. The specter of U.S.

intervention hangs over the Twitter discourse.

Our ﬁndings about the effects of the impingement of

the United States, and American society, on Arab societies

led us to ask whether Arab publics express similar resent-

ments toward Iran. We ﬁnd that Arabic Twitter partic-

ipants strongly dislike Iranian policy and do not express

positive views toward Iranian society—unlike their more

ambivalent views of the United States. Our interpretation

of this data is that what is often labeled “anti-Americanism”

reﬂects, to a considerable extent, fear of alien intrusions and

hegemonic inﬂuence, from whatever source, into one’s

own society. It may well reﬂect a desire for political and

social autonomy rather than dislike for America per se.

Participants on Twitter do not constitute a random

sample of any identiﬁable population: we do not claim

that our data reﬂect Arab attitudes in a perfectly repre-

sentative sense. However, Twitter participants are not

a tiny portion of the Arabic-speaking population; accord-

ing to the Third Wave of the Arab Barometer, close to

40 percent of the Arab public is now online and of this

population, 30 percent is on Twitter. According to

a report by the Dubai School of Government, the

number of active Twitter users in the Arab World

reached 3.7 million in 2013, up from 2 million in

2012. Saudi Arabia records the highest number of active

Twitter users at 1.9 million. Egypt (519,000) and the

UAE (401,000) come in second and third.

In these

countries Arabic is the preferred language of those who

tweet, with 73.6 percent of all activity in Arabic. No

systematic analysis has been done to establish the proﬁle of

Arabic Twitter users, but it seems clear that they are

relatively young. Sometimes the Arab youth population

has been portrayed as more cosmopolitan, more moderate,

less religious, and more pro-democratic than the Arab

population as a whole.

But careful analysis of the Arab

Barometer Second Wave data (2011–2012) indicates that

Arab youth are not necessarily more democratic than older

cohorts and are actually slightly more supportive of

political Islam.

Little evidence therefore exists to support

the possible objection that our analyses of Twitter data are

biased in a pro-democratic, moderate, or secular direction.

Twitter discourses are distinctive, and not necessarily

representative of mass public attitudes, but they are

increasingly important as a medium of expression and

communication, particularly for citizens who are especially

56 Perspectives on Politics

Articles | Anti-Americanism and Anti-Interventionism in Arabic Twitter Discourses

interested in politics. Arabic-language social media dis-

courses affect participants’ expectations about how other

Twitter participants will respond to their own posts. They

are therefore likely to affect participants’ own expressions

of views both through persuasion and socialization and by

shaping their incentives with respect to their own con-

tributions.

Hence, these discourses are politically im-

portant in their own right.

Insofar as public politics is

migrating to the Worldwide Web, political scientists need

to study social media.

Another major advantage of our focus on social media

is that we can examine reactions to speciﬁc events. As we

will show, these speciﬁc reactions are often more in-

formative than aggregate ﬁndings, whether from polls or

our own Twitter-based analysis, because they reﬂect

interpretations of events as they unfold, thereby pro-

viding evidence of pre-existing attitudes, and because

reactions to events are interactive, generating a social and

political discourse. In situations characterized by ambi-

guity, people who are active on Twitter create a social

reality, which may correspond more or less closely to

what analysts later decide actually occurred. The tech-

nology that we employ therefore has wide implications

for studying social and political discourses around the

world, on a variety of topics. Its relevance is not limited

to the Middle East or to the analysis of anti-Americanism.

We hope that our analysis of anti-Americanism will help

to stimulate creative and rigorous social media analysis in

political science.

We ﬁrs t explain how we use social media analysis to

analyze discourses about America and intervention, in

the Middle East and pose our critical questions. Next we

analyze general levels of anti-Americanism in the Middle

East.Wethenmovetoourcorecontributionwherewe

examine Twitter data following speciﬁcevents.Weﬁrst

look at Arabic Twitter responses to actions or inactions

by the United States that affected Arabs including the

July 2013 coup against Mohamed Morsi’s government

in Egypt and the Syrian civil war. We then turn to

responses to the publicity, between September 1 and

November 10, 2012, over Innocence of Muslims,avideo

that was widely viewed in the Arab and Muslim world

and discussed on Twitter. Next, we shift to events where

America was the target, either of human action (the

Boston Marathon bombing of April 2013) or of nature

(as in Hurricane Sandy, October 2012).

Reactions by the Arabic Twitter public to these

events show that in terms of volume of trafﬁc, political

anti-Americanism is much stronger than s ocial anti-

Americanism. But a more important distinction is

between attit udes toward American domestic politics

and society, which reveal considerable ambivalence,

and attitudes toward how the United States affects

politics a nd society in Arab countries, which are

intensely negative. Evidence of Arab hostility

to Iran, which we then examine, provides crucial

evidence that anti-Americanism is one aspect of a more

generic phenomenon: publics in a variety of countries

dislike the impingement of powerful countries on their

societies.

Analyzing Soci al Media Discourses

Studying so cial media such as Twitte r implies studying

discourses—how people talk to one another electroni-

cally. Expressions of opinion on social media are not

necessarily representative of the views of the publics of

countries in which they or iginate, but they are views that

individuals have decided—unprompted —to express.

Fur thermore, people w ho express these views have been

able to see others’ views on Twitter and can t herefore

react to the changing social media universe as they

per ceive i t. Data derived from monitoring Twitt er trafﬁc

therefo re enable us not only to answer questions simi lar to

those posed by analysts relying on public opinion dat a,

but al so to answer questions abo ut interactions among

those people expressing opinions.

In carrying out this

analysis we are able to learn about anti- A mericanism in

the Middle East on the basis of an analysis of Twitter

feeds, without identifying individuals.

The existing literature on the Middle East is divided

about the depth and nature of Arab animosity to the

United States. One s chool of thought holds that levels of

anti-Americanism are inherent to the culture and

identity of Arabs and Muslims.

High levels of anti-

Americanism are viewed as reactions to western and

liberal values, which some in the region view as antithet-

ical to Islamic precepts. In this vein, some studies show

that levels of anti-Americanism increase when secular-

religious tensions grow.

A different school of thought

holds that levels of anti-Americanism reﬂect negative

views of U.S. policies in the region rather than of the

United States as a society.

Many of these scholars have

shown that there is much admiration for basic American

values and culture.

Looking at responses t o events sheds light on these

questions. We can investigate responses to American

intervention or possible interventions in the Middle

East; we can examine Arabic responses to social events

where American society could be seen as critical of

Arabic society; and we can look at responses to events

where U.S. society itself is a tar get. We ﬁnd that Arabic-

language political discourses are permeated with anti-

Americanism, particularly when issues of intervention

arise; but views toward American society are more

complex. The volume of social expression is much lower

than that of political opinion. When Americans are the

victims rather than the perpetrators of harmful actions

there is some animus toward American society but this

animus is not as intense as Arab hostility toward

American policy abroad.

March 2015

Vol. 13/No. 1 57

Research Design and Methodology

We use Arabic Twitter posts gathered and stored by

Crimson Hexagon, a social media analytics company,

employing a supervised text analysis model developed by

Daniel Hopkins and Gary King.

Text analysis methods

can be either “supervised” or “ unsupervised.”

The

difference lies in who (or what) determines the topics by

which to classify texts. Using an unsupervised method,

topics are not determined ex ante; rather, the statistical

method itself helps determine them.

In contrast, using

a supervised method, the researcher hand-codes a

“training-set” of documents into pre-determined catego-

ries, and on the basis of these training documents the rest

are classiﬁed by an algorithm.

In our study we seek to

know the proportion of the population of tweets that ﬁtin

speciﬁc categories. The supervised ReadMe algorithm is

highly suited to this calculation. Rather than estimating

topic proportions based on the categorization of individual

documents, ReadMe estimates these proportions using

words within each text. Importantly, this means that

ReadMe does not individually classify tweets. But we do

get population-based estimates, which are the quantity of

interest here, and we present both proportions and

estimated volume, which is the product of the proportions

and overall volume.

We gain access to these techniques and data through

Crimson Hexagon,

a social media analytics company

founded in 2007. Crimson Hexagon (CH) combines the

text analysis method developed by Hopkins and King with

a vast collection of social media data in an easy-to-use

online platform. In particular, we use here, among other

things, the universe of Arabic language Twitter data from

2012–2013.

The analysis proceeds as follows. First, the user

determines the date range of interest and sources to draw

upon, imposing language and geographical restrictions as

desired.

After these basic parameters are set up, she

determines a set of keywords on which to draw texts

satisfying the other restrictions. CH offers the use of

Boolean operators to create a complex set of keywords (or

phrases) used to include or exclude texts.

After setting up the basic parameters and the keyword

restrictions—forming what CH calls a “monitor”—the

user is then ready to begin training. Because Crimson

Hexagon uses essentially the Hopkins-King algorithm—

a supervised text analysis method—a set of pre-deﬁned

categories and training texts is required. To aid with the

supervised document classiﬁcation process, CH feeds the

user, one-by-one, tweets that match the setup restrictions.

The user places each tweet into the appropriate category,

continuing this process until she feels that the monitor has

been sufﬁciently trained.

Once training is complete, the user runs the algorithm

and accesses the analysis results. The main results consist

of daily estimated frequency data for each of the

categories. The user also has limited access to the raw

tweets themselves, either through Crimson Hexagon’s

“bulk export” function, which is limited to 10,000 tweets

per day, or through example tweets, which it uses a pro-

prietary classiﬁer to select. They are the classiﬁer’s

“best guess” of good examples of each category. These

are helpful for getting a rough sense at whether the training

has picked up on what the analyst is interested in, and they

were used in our monitor review process.

The greatest advantage of Crimson Hexagon is that, as

a Twitter-certiﬁed product with access to the “Twitter ﬁre

hose,” it provides every public tweet ever posted on

Twitter—in any language and from any location—that

matches the user-determined restrictions; and its commer-

cial customers demand a high level of ﬁltering out of spam

and bots that could otherwise contaminate our analysis.

Two weaknesses of using Crimson Hexagon should be

mentioned. CH limits access to the underlying source

texts; we have compensated by reading thousands of tweets

manually. And CH does not enable us to distinguish

between original tweets and re-tweets, so our implicit

assumption is that re-tweets usually reﬂect sympathy with

the original tweet.

Thus, with a reliable set of Arabic keywords referring

to the United States, we can analyze every genuine Arabic

tweet referring to the United States that has ever been

sent. Of course no method comes without limitations.

Ferreting out sarcasm is well beyond the Read-Me

algorithm. And any text method requires extensive ex

post checking and transparency, which we attempt to do

in our extensive online supplementary materials.

Only

small percentages of Twitter users indicate their country,

making cross-national analysis difﬁcult.

Analyzing General Levels of

Anti-Americanism in the Middle East

As mentioned earlier, examination of public opinion data

from the Arab world has long revealed high levels of

anti-Americanism.

Because Twitter posts result from

choices to express oneself (rather than a response to a

question someone might never have thought about), data

from Twitter enable us to discover whether self-expressed

views—the Arabic discourses—are similarly hostile to the

United States.

Table 1 reports o n two different sets of public opinion

polls—the 2012 Pew Global Attitudes Poll and the

2011/2012 Arab Barometer—in seven Arab countries:

Egypt,Jordan,Lebanon,Tunisia,Iraq,Algeriaand

Yemen. First, we look at two sets of questions in the

Pew Global Attitude Poll: favorability scores toward the

United States and agreement with the statement that it is

good that U.S. ideas and customs are spreading to the

region . Second, we turn to two Arab Baromete r ques-

tions: whether respondents believe that “armed

58 Perspectives on Politics

Articles | Anti-Americanism and Anti-Interventionism in Arabic Twitter Discourses

operations” are justiﬁed against the United States because

of its interference in the region, and whether respondents

believe there are positive features linked to American and

Western culture.

Favorability scores are quite low—remarkably low in

some countries such as Egypt and Jordan, while even in

Lebanon and Tunisia these scores are not predominantly

positive. There is widespread opposition to the spread of

Western values. Yet, quite remarkably in terms of the

strongly negative views about American foreign policy and

the spread of western values, 50 percent to 83 percent of

respondents say that American culture has positive aspects.

That is, in the polling data we see indications of the same

distinction we will document for the Twitter data between

Arab attitudes toward the impingement of American

policy and society on Arab countries and toward American

society as such.

Systematic polling data in the region is quite difﬁcult

to secure. The Pew Global Attitudes poll, for example,

does not surve y in many Arab countries and in the

2012 wave it only gathered data in four countries:

Egypt, Lebanon, Jordan and Tunisia. The Arab

Barometer data has expan ded its number o f countries

in recent waves but problems remain . For example,

some countries have not allowed social scientists to ask

questions about attitudes toward the United States.

Clearly, countries are worried about the negative

ramiﬁcations of “exposing” the degree of anti-American

sentiment in t heir countries. Morocco (2006), Egypt

(2006), and Saudi Arabia (2013) are examples of

countries restrictin g pu blic opin ion polling.

General Attitudes toward America in

the Twitter Universe

Our ﬁrst monitor, called the “General Anti-Americanism”

(General AA) monitor, looks at general trends in the

Arabic Twitter conversation about the United States. We

sought to make this monitor as broad as possible, choosing

a large date range: from the earliest date for which Twitter

data is available in Crimson Hexagon to the current date.

However, we quickly found that data after January 1, 2012

are the most reliable, and all of our ﬁgures are based on

tweets after this date. We also aimed for breadth in our

keyword criteria, producing a large set of keywords.

Between January 1, 2012, and December 31, 2013, an

astounding number of 33,009,354 tweets matched these

date and keyword criteria and were classiﬁed in one of our

six substantive categories.

The goal of our general monitor is to get a general

sense of Arabic-language tweets referring to the United

States. Furthermore, this analysis helps give an indication

of instances of when an “event” has happened that

requires additional scrutiny. We developed the six training

categories shown in Table 2 based on two distinctions that

are important in the anti-Americanism literature: the topic

of discussion (political vs. social) and its valence (negative,

neutral, or positive).

Which categories are mo st prominent? Is anti-

Americanism rampant in th eArabicTwitteruniverse?

Does it span both political and social categories, or is it

circumscribed to political anti-Americanism? Table 2

plots the estimated number of tweets for each category.

The results are striking. Although the ratio of negative to

positive tweets is over 3:1 in both social and political

categories, the volume of political trafﬁc is nearly four

times as great as social tra fﬁ c.

Table 1

Arab public opinion toward the United States

Egypt Jordan Lebanon Tunisia Iraq Algeria Yemen

Favorability opinion of the United States, 2012

(Pew Global Attitudes 2012, % favorable)

19 12 48 45 NA NA NA

It’s good American ideas and customs are spreading

here (Pew Global Attitudes 2012, % agree)

11 10 41 25 NA NA NA

The United States’ interference in the region justiﬁes

armed operations against the United States

everywhere (Arab Barometer 2011/2012, % agree)

63 58 44 64 70 53 47

American and western culture have positive aspects

(Arab Barometer 2011/2012, % agree)

63 55 75 83 67 50 66

Table 2

Estimated number of tweets per category

for general attitudes monitor

Category Frequency Percentage

Positive social 957,922 3

Neutral social 2,986,740 9

Negative social 3,140,021 10

Positive political 4,389,598 13

Neutral political 6,759,360 20

Negative political 14,775,713 45

March 2015

Vol. 13/No. 1 59

In ﬁgure 1 we plot over time each of the categories

presented in table 2. The small black triangles along the

horizontal axis represent a set of key events during our time

period, many of which we analyze in detail in the next

section. One thing that stands out is that for these events

there are often dramatic spikes in Twitter volume. This is

perhaps not surprising given the nature of Twitter.

However, it raises an important question: when we remove

high-category days, is the volume of negative political

trafﬁc still substantially higher than the other categories?

Figure 2 shows that even when we eliminate days where

there are spikes in Twitter activity, negative political trafﬁc

remains dominant. The top left plot is a histogram of total

on-topic volume at the daily level. The population median

is given with a vertical line. This clearly shows the presence

of spikes. The top right plot shows the same histogram,

where we have removed all days above the ﬁftieth

percentile. The bottom left plots the percent of total on-

topic volume that is politically positive or negative (y-axis)

against the percentile used to remove days that had a total

volume above that percentile. For example, at an x-axis

value of 1, all days are used, and at the x-axis of .5, we are

only using days captured in the histogram in the top right

(those days with a volume below the median). We see that

once we restrict our sample to days without events, we get

a similar picture. Negative political volume stays close to

40 percent of the entire sample. We see a similar story if we

take the ratio of negative (political and social) to positive

tweets in the bottom right of ﬁgure 2. Looking at both the

universe of tweets and only using days that did not have

spikes in volume, we arrive at the same conclusion that

negative political tweets dominate the Arabic language

conversation about U.S. politics and society.

However, in spite of the predominance of negative

political tweets over others, one political event stands out

as prompting positive responses: the 2012 U.S. presiden-

tial election. The spike in positive trafﬁc around this time

(refer to ﬁgure 1) was the largest spike in trafﬁc for our

entire monitor. When we closely inspected this we found it

was comprised principally of favorable reactions to Pres-

ident Obama’s re-election and admiration for the in-

stitution of free and fair elections, coupled with the desire

for similar institutions in one’s own country. However,

this positivity towards Obama’s re-election disappeared

very quickly, resulting in a very low percentage for this

category overall.

Figure 1

Total Arabic tweets by political and social category using all tweets in the world, plotted

across time

Note: The triangles indicate key events during this time period.

60 Perspectives on Politics

Articles | Anti-Americanism and Anti-Interventionism in Arabic Twitter Discourses

In summary, the conversation on Twitter, in Arabic,

about the Unites States is especially negative towards U.S.

policy; the conversation about U.S. society is also mostly

negative but with some positive elements and much

smaller in volume than the conversation about U.S.

policy. These results hold whether we look at all Twitter

trafﬁc or only those days that reﬂect more of a “baseline.”

Analyzing Responses to Speciﬁc

Events

We now analyze a set of events in more detail than was

possible in the general monitor. We structure our analysis

by looking both at events involving U.S. actions towards

the Arabic speaking world, and at events in which the

United States was the target. This distinction is impor-

tant because we might expect different levels of trafﬁc and

different levels of positive and negative reactions to these

two types of events. We might expect, because of

generally negative attitudes towards the United States,

that actions the United States takes towards the Arabic

speaking world would generate large amounts of negative

trafﬁc. On the other hand, we might expect different

reactions when the United States is the target, perhaps

even expressions of satisfaction. We ﬁrst discuss political

events in which the United States was seen as the

impinging actor; we then analyze the responses to the

Innocence of Muslims video emanating from American

society; and next we turn to events in which the United

States and the American people were affected by actions of

people or natural forces such as Hurricane Sandy.

Two of our monitors target political events: a monitor

that tracks the U.S.-focused trafﬁc surrounding Morsi’s

removal from ofﬁce and the subsequent pro- and anti-

military protests in Egypt; and a monitor that tracks

U.S.-focused trafﬁc surrounding the Syrian civil war. These

two situations differ from each other in many ways, but

their similarities invite interesting comparisons. In both

situations, two camps dominated the conversation: on the

Figure 2

Analysis excluding spikes in Twitter volume

Notes: The top left ﬁgure plots the histogram of total on-topic daily volume with a vertical line at the median. The top right ﬁgure plots the same

distribution but using observations below the population median. The bottom left ﬁgure plots the percentage of total on-topic volume that is

negative or positive political as days are removed for being over a given percentile (x-axis) in total volume. The bottom right ﬁgure plots

a similar line, but where the sum of negative tweets is divided by the sum of positive tweets.

March 2015

Vol. 13/No. 1 61

one hand, the “pro-regime” camps (those who supported

Assad in the Syria monitor and those who supported the

military in the Morsi monitor), and on the other, the “anti-

regime” camps (those who supported the Syrian revolution

and those who supported Morsi). Also, looking at these two

events together allows us to understand situations in which

the U.S. government is perceived to be able to inﬂuence the

course of events due to its inﬂuence in the region. This is in

contrast to situations where the United States is the target,

which we will discuss.

Morsi and Egypt

The Morsi monitor seeks to look at the Arabic conver-

sation surrounding the removal of Mohammed Morsi

from power on July 3, 2013 and the events that followed.

For this monitor, we were interested in looking at how

Arabic tweeters were talking about the United States in

relation to the events that were going on in Egypt. How

did the Arabic tweeter public respond to these events?

For this monitor tweets were only selected for analysis

if they included both a reference to the United States and

a reference to Egypt.

We ran the monitor from June 27,

2013 (approximately a week before Morsi was forced out

of ofﬁce) to September 8, 2013 (when it seemed that news

coverage of the situation in Egypt was giving way to

coverage of the crisis in Syria).

In designing the categories for our analysis we wanted

to allow for both sides of the dispute to take on both pro-

and anti-American positions, and to allow for both

general pro and anti sentiments towards the United

States that were not in the context of supporting one

or the other domestic group. With these desires in mind,

we came up the categories listed in table 3, where we

report the results. To facilitate comparability among the

event monitors, in all cases we exclude tweets that

contained only neutral news reports, though our sub-

stantive conclusions do not change if we include these.

The ﬁrst thing to notice is that positive views of the

United States were a very small proportion of Twitter

trafﬁc: 4 percent overall. The top three c ategories

encompassed tweets that expressed opposition to the

United States. The largest c ategory was an ti-American

and anti-military, the next largest was general anti-

Americanism, and third was anti-Ame rican and an ti-

Muslim Brotherhood. No matter which side of the

domestic dispute an individual was on, he or she was

likely to be opposed to the United States. Because of

the small number of positive tweets, it is hard to

determine how much of the positive trafﬁc was coming

from anti-military versus anti-Muslim Brotherhood

tweeters. In summary, this analysis clearly shows sharp

divisions within Egy ptian society on the question of

the Muslim Brotherhood, but a common opposition to

the United States. Rather than an enemy of an enemy

being a friend, the United States is consistently cast as

an enemy.

Syria

Our analysis of Syria is similar to the Morsi analysis in

that it seeks to look at how Arabic tweeters are talking

about the United States in relation to one particular event

or topic—in this case the Syrian Civil War, which began

on March 15, 2011. The explosion of discussion on

Twitter about possible U.S. intervention in Syria following

alleged chemical weapons attacks by the Assad regime

provides a good opportunity to look into reactions towards

U.S. foreign policy. Since this story in 2013 was one of

U.S. inaction, we are asking here about levels of anti-

American political opinion in an event where the United

States had not used force but was widely perceived as being

able to inﬂuence the course of events in Syria.

We use tweets if they included a reference to the

United States and a reference to Syria, following a similar

procedure to that used in the Morsi monitor. We trained

this analysis from January 1, 2012, to December 31,

2013. Modeling our categories on the Morsi monitor,

but with slight adaptions based on what we found when

looking at the Syria trafﬁc, we obtained the ﬁve categories

found in table 4, where we report the results.

Pro-U.S. trafﬁc is between about 1 percent and 4

percent of total non-news trafﬁc, depending on the time

period. All of the pro-U.S. trafﬁc comes from the anti-

regime camp, and the amount of negative trafﬁc coming

from the pro-regime camp far outweighs the negative

trafﬁc from the revolutionaries. This is to be expected,

given that the United States was seen as on the side of the

Syrian revolution. However, even for the anti-regime

tweeters there were 350 percent more anti-U.S. tweets

than pro-U.S. tweets before the chemical weapons in-

cident, and about 1,200 percent more after the event. In

addition, anti-regime trafﬁc that was ambiguous about the

United States was large prior to the chemical weapons

attack but declined sharply following the attacks, whereas

general negative trafﬁc against the United States increased

after the attacks. The right side of table 4 presents ﬁgures

for the post-August 20 period.

Table 3

Total estimated number of tweets in each

category for Egypt, omitting 515,257 posts

with News/Neutral content

Category Frequency Percentage

Pro-U.S. 66,024 4

Anti-U.S. and anti-Muslim

Brotherhood

220,906 13

General anti-American 500,311 29

Anti-American and

anti-military

918,541 54

62 Perspectives on Politics

Articles | Anti-Americanism and Anti-Interventionism in Arabic Twitter Discourses

Twitter responses to the Egypt and Syria situations

dramatically demonstrate the breadth and depth of negative

views toward American foreign policy in the Arabic Twitter

universe. It is not surprising that supporters of the Muslim

Brotherhood and of the Assad regime (not necessarily the

same people) are negative toward the United States, since

the United States was cool toward the Brotherhood and has

been very negative toward the Assad regime. But what is

dramatic is the extent to which Arabic Twitter users who

favor the Egyptian military are not positive toward the

United States, and the extent to which those who favor the

Syrian revolutionaries are predominantly negative toward

the United States. This is particularly striking for the Syria

monitor, in which 87 percent of tweeters who expressed

political views were antagonistic toward the United States,

despite the fact that the United States opposed the Assad

regime, which was also opposed by many Arabic tweeters.

In the Egypt monitor, the proportion of political tweets

that was anti-American reached 96 percent. We see a similar

pattern if we only look at Twitter participants who were

more or less on the same side as the United States: that is,

they favored the Egyptian military or were opposed to the

Assad regime. Once again, the fact that the United States

was the enemy of their enemy did not make them regard it

as their friend.

Arab Responses to Critical Views from American

Society: Innocence of Muslims

Written and produced by an Egyptian-born Coptic

Christian residing in the United States, the 14-minute

long Innocence of Muslims ﬁlm was widely considered an

attack on Muslim society. After the ﬁrst Arabic-dubbed

version of the ﬁlm appeared early September 2012, there

were widespread demonstrations and riots in the Middle

East, many with an anti-American tone, suggesting that

the ﬁlm’s views were attributed by many people to the

United States. Its appearance also led to an immense

outpouring of commentary on Twitter. It was important

to create a monitor for this event since it could be

considered an attack on Muslim society by American

society, and thus responses to it could help indicate the

extent to which perceptions of American society fuel anti-

Americanism. If perceptions of American society do fuel

anti-American sentiment, we would expect to see direct

expressions of anger directed at American society in

response to this event. We ﬁnd, however, a much more

complex pattern of responses.

We developed a short and speciﬁc list of base words to

help us capture tweets related to this event.

examined the two-month period right before and after

the movie began making headlines in the Middle East,

beginning on September 1, 2012 and continuing until

November 10, 2012. As we trained the monitor, we found

that responses were not easily mapped onto a s imple

pro- or anti-Americanism scale; such a scale did not

capture the Twitter discussion about the ﬁlm, which

focused on what the best course of action for Muslims to

address the ﬁlm would be. For this reason, we modi ﬁed

our categories so that they cap tured different sentiments

expressed about the ﬁlm. This process le d to t he

discovery of ﬁve cat egories that help us convey the

substantive patterns we did ﬁnd. These categories are

listed in table 5.

Reviewing these results, we ﬁnd that the biggest (non-

news) category was the category encouraging people to

ignore the ﬁlm. The second largest category included posts

supporting individual action against the ﬁlm. The third

largest category reﬂects posts arguing that Islam is stronger

than the ﬁlm. All of these reactions have negative

implications but none of them represents a clear con-

demnation of American s ociety in general. We did not

ﬁnd tweets with statements such as “This means that all

Americans hate us” or “All of American society and

people should be condemned.” In this monitor, we see

a clear negative reaction but we do not see a predomi-

nance of direct attacks on American society, unlike the

direct attacks on American policy in the general, Morsi,

and Syria monitors. But there is no doubt that both in

this monitor, and the general anti-American monitor for

Table 4

Total estimated number of tweets in each category for Syria, split by date just prior to

chemical weapons attacks

Jan. 1, 2012–Aug.19, 2013 Aug. 20, 2013–Dec. 31, 2013

Category Frequency Percentage Frequency Percentage

Pro U.S., anti-regime 67,704 4 7,856 1

Ambiguous about U.S., anti-regime 147,006 9 39,077 4

Negative U.S., anti-regime 252,958 15 94,898 10

Negative U.S., general 526,917 32 401,854 42

Negative U.S., pro-regime 663,838 40 416,112 43

Notes: News/Neutral posts are omitted.

March 2015

Vol. 13/No. 1 63

this time period, expressions of sentiment a re prepon-

derantly negative.

Arab Responses to Harms to America

Thepreviousanalysesalldealtwitheventswherethe

Arab world was the target. We n ow shift our attention

to two events in which the United States was a targ et,

providing an important symmetry to our paper.

We st udy th e Bos ton marat hon bombings and

Hurricane Sandy.

Boston Marathon Bombing. On April 15, 2013 a set of

bombs exploded at the Boston Marathon, killing three

people and injuring an estimated 264 others. The accused

perpetrators purportedly carried out the attack for the sake

of Islam. This monitor shows us how Arabic tweeters

responded to this attack on the United States. How much

sympathy was there for victims, and were responses

principally focused on U.S. politics and policies or on

U.S. society?

We developed a short and speciﬁc list of base words to

help us capture tweets related to this event, as detailed in

the on-line supplementary materials. As with the In-

nocence of Muslims monitor responses did not easily

map onto a pro- or anti-Americanism scale, so we

grouped responses by the topics that seemed prominent

during our training. These categories are listed in table 6,

where we report the results. This monitor was run from

April 13 to June 10, 2013.

Figure 3 presents the Boston monitor results over

time for each c ategory, using the total number of

tweets. The ﬁrst small black triangle marks the date of

thebombingandthesecondisthedateoftheﬁnal

shoot-out with the alleged perpetrators. Table 6

presents the total number of tweets for each cat egory,

aggregated ove r time.

Several interesting patterns are present in the data.

First, the ratio of tweets claiming that media attention to

the event was undeserved to tweets expressing sympathy

for the victims was almost 4:1. The low levels of

sympathy towards innocent victims, and high levels of

tweets dismissing the importance of their plight, suggest

substantial negative sentiments toward the United States.

Indeed, the amount of sympathy expressed for Arabs

facing discrimination was about a third the size of

sympathy for the victims, and the “Backlash” category,

indicating concern about a backlash against people in the

United States of Arab origin, had over 90,000 tweets.

On the other hand, we did not observe tweets celebrating

the attack on U.S. society. While many individuals argued

that the attention was undeserved, they did so by pointing

to deaths in the Arab world (especially Syria), often by

arguing that behind these deaths are American policies.

While overall sympathy for victims was a relatively small

percentage of the conversation, it is American policies and

interference in the Arab world that drew the ire of many

tweeters.

We reﬂected a great deal about whether these

negative views of the United States reﬂect “bias,” as

distinguished from opinion.

Opinion reﬂects attitudes

and reactions to events but no major cognitive distor-

tions. People with negative opinions of other people,

groups, societies, or institutions dislike what they actu-

ally do, or stand for symbolically. Bias, however, re ﬂ ects

distorted perceptions of reality: biased people misper-

ceive the actions of the people, groups, societies, or

institutions that they dislike.

Although in general we found it difﬁc ult to in fer bias

from the information we gathere d, the clearest possible

evidence of bias comes in the Boston Marathon

monitor, speciﬁcally in the Conspiracy category. There

were nearly twice as ma ny tweets speculating about

a conspiracy involving U.S. security agencies than

tweets expressing sympathy for the victims.

We interpret the ﬁndings from the Boston monitor as

suggesting a signiﬁcant level of bias against the United

States. Bias is based on a deeply negative view toward the

United States that strongly colors respondent’s percep-

tions of observed U.S. actions and that may create negative

beliefs about unobserved actions by the United States and

Table 5

Total estimated number of tweets in each

category for the world

Category Frequency Percentage

Demanding action from

political leaders

10,735 2

Anger towards producers 26,376 4

Evidence Islam is stronger

than ﬁlm

80,383 14

Supporting individual action

against ﬁlm

214,337 36

Encouraging people to

ignore ﬁlm

260,124 44

Note: News/Neutral content posts (433,590) were omitted.

Table 6

Estimated number of tweets per category

for Boston Marathon bombing monitor

Category Frequency Percentage

Sympathy for Arabs/

Muslims

7,999 3

Sympathy for victims 23,887 9

Conspiracy theories 39,742 15

Backlash 85,588 33

Not important 104,678 40

Note: News/Neutral content posts (262,761) were omitted.

64 Perspectives on Politics

Articles | Anti-Americanism and Anti-Interventionism in Arabic Twitter Discourses

the motivations of American decision-makers. Although

there was never any credible evidence of a con spiracy

beyond the plans of the two bombers, we see almost

twice as many tweets that are coded as expressing belief

in a conspiracy than tweets expressing sympathy for the

victims of the bombings. H owever, our inability to

discover clea r evidence of bias, as oppose d to negative

opinion, in the other monitors makes us difﬁde nt

about drawing strong general conclusions about bias.

To understand anti-Americanism it is important to ask

whether the results would have been different were this

event to have happened in another country. Our best effort

to ﬁnd a case as similar as possible comes from the Arabic

Twitter conversation surrounding the death of British

Army soldier Lee Rigby, who was attacked and killed by

Islamist terrorists on May 22, 2013, in Woolwich,

London. The results for London look very different than

those for Boston. While we do not present the detailed

results here,

relative to the total volume of activity there

was much more sympathy for the victim; the view that this

event is not important in light of attacks on Muslims in the

Middle East is not expressed; and there was no discussion of

a possible conspiracy. Apart from news reports, tweets fall

into three categories: expressions of sympathy with victims

(60 percent), expressions of sympathy towards Arabs/

Muslims (20 percent), and expressions of concern about

backlash against Muslims (20 percent). In London there

were three times as many “Sympathy for Victims” tweets

than “Backlash” tweets. The immediate response of Arabic

tweeters was overwhelmingly one of sympathy towards

Arabs and Muslims, with little to no explicit support for the

attackers. There was so little discussion of a possible

conspiracy that it was not included as one of our categories:

the facts of the murder were accepted. This evidence

suggests that hostility to the United States is greater than

hostility to the United Kingdom.

Hurricane Sandy. In late October 2012, Hurricane

Sandy caused extensive damage to the Northeastern

United States. We explore this event—with results pre-

sented in table 7—to search for high levels of antipathy for

American society that we have not heretofore detected.

In the Hurricane Sandy monitor most tweets were

merely news items, leaving just over 310,000 tweets to

analyze. Of these meaningful tweets, 10 percent (about

32,000) expressed the view that Hurricane Sandy is

punishment for the Innocence of Muslims ﬁlm or similar

Figure 3

Estimated number of tweets per category over time for Boston Marathon bombing monitor

March 2015

Vol. 13/No. 1 65

prejudice against Muslims. In addition, about 2 percent

(almost 13,000) were generally negative and 16 percent

(almost 50,000) expressed particular concern about Arabs

or Muslim Americans and residents. Over 30 percent

(about 99,000) said that the Hurricane Sandy events were

not important. In contrast, approximately 25 percent of

tweeters (almost 80,000) commented favorably on the

U.S. government’s handling of the disaster, often as

a contrast to the incompetence of Arab governments,

and about 10 percent (over 30,000) condemned the view

that Hurricane Sandy is punishment for the Innocence of

Muslims ﬁlm. In other words, of the Arabic Twitter

participants who expressed an opinion, somewhat more than

one-third expressed views that can be interpreted as generally

favorable toward American society, somewhat less than one-

third expressed explicitly negative views, and 30 percent

regarded the events as not important. Even if we interpret

“not important” as implying that harm to the United States is

not bad, the difference between this monitor and those

related to directly political events is striking.

The Boston Marathon and Hurricane Sandy monitors

present something of a contrast. An overwhelming 88

percent of non-news responses to the Boston Marathon

bombing are negative toward the United States: these

events are seen as not important (in view of harm done to

Arabs in the Middle East directly or indirectly by

American policy); the greatest concern is for the welfare

of Muslim Americans/residents; or a conspiracy of

American intelligence agencies is viewed as responsible

for the allegations. In contrast, in the Hurricane Sandy

monitor over one-third of Twitter participants, apart

from those merely re-tweeting news, expressed views

quite favorable toward aspects of American society. One

suggestion is that the Boston Marathon events were

framed in an explicitly political way, evoking negative

responses, while the Hurricane Sandy events were not.

Summary

What conclusions can we draw about the nature of anti-

Americanism from our ﬁndings? We ﬁnd that two

dimensions help us distinguish the volume and valence

of Arabic-language responses: social vs. political content;

and the U.S. impact abroad vs. U.S. domestic activities.

Table 8 presents the most important results for previous

surveys and our own analysis, placing negative responses in

bold type. The table excludes neutral content and focuses

on positive and negative reactions to events or topics. We

ﬁnd that events and activities in cell A—those with a focus

on the political impact of the United States abroad—

generate the largest volume and most negative responses.

On the other hand, those in cell D— with a focus on U.S.

domestic social activities and events—generate the

smallest volume and most positive responses, although

there are criticisms as well.

Table 8 shows that U.S. policy is the predominant

target of anti-Americanism. Discussions of American

society are more ambiguous. Arabic Twitter discourses

on how American society affects the Arab world are

particularly negative. Several pieces of evidence point

toward this conclusion. First, public opinion polls (table

1) reveal overwhelming opposition to the proposition that

it is good that American customs and ideas are spreading in

the Arab world, but majority agreement that there are

nevertheless positive aspects of U.S. society. Second, in our

analysis of general anti-Americanism presented in table 2,

the vast majority of tweets that were classi ﬁ ed as “negative

social” dealt with concerns about U.S. society impinging

on Arabic society, especially dealing with women’s issues.

Finally, the dramatic response towards the Innocence of

Muslims ﬁlm is best understood as a response towards the

impingement of U.S. society on Arabic society.

Lacking throughout our analyses is a discourse that is

targeted towards the nature of U.S. norms and practices

within the United States itself. People using social media

in the Arabic-speaking world strongly dislike the impact

of the United States, politically and socially, in their

societies, but are less hostile to the United States as

a society, with its very different customs and practices.

Further evidence from public opinion polling reinfor-

ces the signiﬁcance of our distinction between views of

American society and its impingement on the Arab

world, although the argument is sometimes implicit.

Recall from our introductory discussion that Chiozza

found the most negative views of the United States to

center on the diffusion of American customs abroad, the

war on terror, and U.S. foreign economic policy—all

issues in which the U.S. inﬂuence on the rest of the world

is central. American political institutions, popular culture,

and U.S. business all received a mix of evaluations, while

evaluations of U.S. science were positive. In general, topics

in this second set do not evoke thoughts about how United

Table 7

Estimated number of tweets per category

for Hurricane Sandy monitor

Category Frequency Percentage

Negative towards U.S.

government

6,143 2

Negative general 6,594 2

Concern for Americans 7,869 3

Defending Americans 31,561 10

Negative mentioning

Innocence of Muslims

32,125 10

Concern for Arabs/Muslims 49,276 16

Positive (towards U.S.

government)

78,824 25

Not important 99,131 32

Note: News/Neutral content posts (451,728) were omitted.

66 Perspectives on Politics

Articles | Anti-Americanism and Anti-Interventionism in Arabic Twitter Discourses

States power impinges on countries in the Middle East.

Marc Lynch ascribes increases in anti-Americanism in the

Middle East between 1999 and 2004 largely to the fact

that “the American presence in the Arab world measurably

increased” during this period.

Lars Berger ﬁnds, on the

basis of a poll taken in 2008 in Egypt, Indonesia, and

Pakistan, that fewer than ten percent of respondents

approve of attacks on civilians in the United States, and

that this view is strongly correlated with dislike for American

culture but not with o pposition to American foreign

policy—opposition that is much more widely shared.

Berger therefore provides empirical support for a distinc-

tion that Amaney Jamal made between criticism of U.S.

foreign policy and rejection of U.S. domestic practices. In

her analysis, Jamal argues that fear is an important

component of anti-Americani sm.

Even Arab liberals,

she claims, can become anti-American if they worry

about U.S. actions, and the antipathy of so vereign

nationalists to the United States is accentuated by fear.

Fear, of course, is only activated if the subject is thinking

about the impact of the United States on him or her, and

for most of our Twitter participants, that means foreign

actions by the United States.

These two points—that political anti-Americanism is

more intense than social anti-Americanism but that the

latter is still strong, and that objections are stronger to the

impingement of American society on the Arab world than

to American society itself—are both important. They

suggest that changes in United States policy alone are

unlikely to transform Arab attitudes toward the United

States, and that much will depend on how Arab attitudes

themselves change.

Is the United States Unique?

The literature on anti-Americanism has been written

overwhelmingly by Americans with an exclusive focus on

the United States, and this paper is no exception. The

United States has actively intervened in world politics for

the last 73 years, so it is not surprising that it generates

strong feelings. But in what sense should we view these

feelings as distinctively “anti-American” rather than being

merely expressions of a more generic phenomenon:

resentment of interference in one’s own affairs by a power-

ful state with a distinctly different set of cultural values as

well as political interests? As suggested at the outset of this

paper, we need to raise our sights a bit and look at

countries other than the United States to see whether anti-

Americanism is a highly speciﬁc or even a unique phe-

nomenon, centered on the United States, or one aspect of

a more general phenomenon: opposition to interference

and inﬂuence by powerful states with different cultural

values and political interests.

We approach this question ﬁrst by asking how much

Arabic tweeters talk about the United States, as opposed

Table 8

Summary of ﬁndings

Impact of the United States Abroad U.S. Domestic Activities

Political A. Overwhelmingly negative and very large

• U.S. alliance with Israel

• Iraq War

• Involvement in Syria/Egypt

• Association of U.S. with disliked Arab leaders

• U.S. foreign policy in general

B. Mixed and medium to small trafﬁc

• Admiration for U.S. political institutions,

including elections

• Favorable U.S.-M.E. leaders comparisons

• Criticism of treatment of Muslims and Arabs in

United States

• Negative views of the United States in the

Boston Marathon monitor, including

conspiracy theory: FBI and CIA

Social C. Negative and medium to small trafﬁc

• Salience of and negative responses to

Innocence of Muslims video

• Responses to Pew survey question about

spread of American ideas and customs

• Negative social comments outweigh positive

social comments (3:1) in general AA monitor

D. Mostly positive and small trafﬁc

• Support in Pew poll for view that U.S. society

has “positive aspects”

• Sympathy by some tweeters for Sandy/Boston

victims

• Some positive interest in U.S. culture in General

AA monitor

• Criticism of treatment of Muslims and Arabs in

United States

Note:

Most of this interest is classiﬁed as neutral.

March 2015

Vol. 13/No. 1 67

to other countries that are important either globally or in

the Middle East. To address this question, we compared

trafﬁc about the United States to trafﬁc about six other

countries: Iran, Israel, Turkey, India, Russia, China, and

the United Kingdom. The results of this analysis are

found in table 9. The United States is the most important

focus of this Twitter trafﬁc, an entire quarter of which was

about the United States. However, other inﬂuential non-

Arab countries that regularly get involved in Middle

Eastern affairs commanded a large share of this trafﬁcas

well: Iran with 17.51 percent, Israel with 17.28 percent,

and Turkey with 13.84 percent.

The large number of tweets about Iran enables us to

ask whether negative attitudes toward the United States

are unique or whether, on the contrary, they extend to

other powerful states with different cultural values and

political interests. To answer this question we conducted

an analysis on Iran that is similar to the one we described

earlier on general anti-Americanism. While the United

States and Iran are both regarded as enemies, Arabic

speaking tweeters appear to have distinct and cognitively

consistent reasons to dislike both states, since both states

are seen to interfere with their countries. So it does not

require cognitive distortion to dislike both the United

States and Iran.

As table 10 shows, non-neutral trafﬁc referring to Iran

was overwhelmingly negative. Indeed, we found so few

tweets that were either positive, or related to Iranian

society, that we were unable even to estimate the propor-

tions of these categories. Hence there is indeed a contrast

between feelings toward Iran and the United States, but it

seems to favor the United States. Our ﬁndings reinforce the

view taken by Peter J. Katzenstein and Robert O. Keohane,

and Giacomo Chiozza, on the basis of public opinion

polling, that views toward the United States elsewhere in

the world are highly ambivalent. This ambivalence

reﬂects the “polyvalence” of American symbols: “they

embody a variety of values with different meanings to

different people and indeed even to the same individ-

ual.

In both the polling data and even more in our

Twitter data, one observes admiration for American

popular culture, helping to create such ambivalence.

There is no such Arab admiration for Iranian popular

culture, and no discernible ambivalence.

Anti-Americanism,

Anti-Interventionism, and the Politics

of Social Media Discourses

Anti-Americanism is an important political phenomenon

and the subject of an extensive literature, with essentially

all of the previous quantitative work relying exclusively

on public opinion surveys and seeking to ascertain

attitudes. In contrast, we focus on social media, specif-

ically Twitter Arabic language feeds. We monitor millions

of Arabic-language messages during 2012–13 on Twitter,

enabling us to delineate the discourses that take place on

social media in the Arabic-speaking world. Our aggregate

monitor ﬁnds that discourses with respect to both political

and social issues are overwhelmingly negative but that the

volume of political trafﬁc is four to ﬁve times as great as the

volume of social trafﬁc. Consistent with the ﬁndings of

public opinion polls, distrust of the United States

Government in the Middle East is deep.

OuranalysisofTwitterdatahasallowedustorevisit

many of the salient debates on anti-Americanism. First,

with a h igh degree of conﬁdence, we show that the

primary focus of anti-Americanism in the Arabic Twitter

Table 9

Volume of Arabic Twitter trafﬁc for seven countries, Jan. 1, 2012–Dec. 31, 2013

Country Keywords Total Trafﬁc Percentage

United States of America America, American(s), the United States of America 40,845,963 25.88

Iran Iran, Iranian(s) 27,634,417 17.51

Israel Israel, Israeli(s), Zionism, Zionist(s), Judaism, Jew(s),

Jewish

27,266,590 17.28

Turkey Turkey, Turkish, Turk(s) 18,062,640 11.45

India India, Indian(s) 11,927,620 7.56

Russia Russia, Russian(s), Soviet(s) 11,248,828 7.13

China China, Chinese 11,123,497 7.05

United Kingdom Britain, British, United Kingdom 9,707,759 6.15

Table 10

Total estimated number of tweets in each

category for Iran monitor

Category Frequency Percentage

Negative political 12,141,844 28

Neutral political 30,510,853 72

Note: The number of tweets here is higher than in table 9

because references to “Shia” or “Shiite” were included in the

Iran monitor.

68 Perspectives on Politics

Articles | Anti-Americanism and Anti-Interventionism in Arabic Twitter Discourses

universe is more about international politics than about

domestic social norms. This ﬁnding has two important

implications. First, much of the literature on anti-Amer-

icanism as driven by domestic social norms has posited

that Arab societies, or even Muslim societies more

generally, dislike the United States (and other Western

countries) for the values that they express internally. If

cultural distaste dictated high levels of anti-Americanism,

then only fundamental cultural or normative change could

provide a remedy to such intolerance. Direct policies

would have to be aimed at inducing or supporting cultural

shifts at the societal level. Such possible policies would

include civil society initiatives to promote a democratic

ethos, curricular development that emphasizes tolerance

and western liberal values, the encouragement of “moder-

ate” voices to assume the pulpit, and the strengthening of

youth “liberal” voices on Arab streets. Their purpose would

be to encourage the evolution of society in more liberal-

democratic directions, generating also a more pro-American

stance among Arab societies.

However, our results, especially as summarized in

table 8, indicate that levels of anti-Americanism are

primarily driven by the perceived impingement of America

on the Middle East, and speciﬁcally by United States

intervention in the region. We have further demonstrated

that anti-American sentiment focuses around events in

which the United States is seen as playing a major role.

Overwhelmingly, citizens of the Arab world distrust the

intentions of the United States: regardless of what the

United States does, Arab publics will evaluate the United

States through a deeply suspicious lens. As long as the

United States continues to intervene militarily in the

Middle East, its actions will generate resentment. Iranian

intervention generates similar resentment.

Our analysis of attitudes toward Iran has led us to

interrogate the very concept of anti-Americanism. The

literature on anti-Americanism has been written over-

whelmingly by Americans with an exclusive focus on the

United States, and we began our analysis with the same

mind-set. The United States has actively intervened in

world politics for the last 73 years, so it is not surprising

that it generates strong feelings. But in what sense should

we view these feelings as distinctively “anti-American”

rather than being merely expressions of resentment

towards interference in one’s own affairs by a powerful

state with a distin ctly different set of cult ural values

and political interests? The concept of a distinct

“anti-Americanism” has been rei ﬁed by a polling

technology that asked people around the world what they

think of the United States, not their views of other powerful

countries that are perceived to intervene in their affairs.

In light of our ﬁndings about Iran, political scientists

should examine social media discourses and public opinion

toward countries other than the United States to see

whether anti-Americanism is a highly speciﬁc or even

a unique phenomenon, centered on the United States; or

merely one aspect of a more general phenomenon:

opposition to interference and inﬂuence by powerful states

with different cultural values and political interests. Our

analysis of Twitter data and public opinion analysis

suggests that the latter interpretation is likely to be correct.

Arab anti-Americanism appears to be a speciﬁc version

of a more general phenomenon: dislike and distrust of the

impingement of other societies with different political

and social values on the Arab Middle East. Seventy-ﬁve

years ago these views may have been common toward

Britain and France; but as the inﬂuence of these states has

diminished, publics around the world no longer think

of tweaking the tail of the British lion or calling out

French republican s for their imperialist hypocrisy.

Anti-Americanism arose in the wake of the rise of

America to world power, and was strongest in areas

such as the Middle East where intervention by the

United States was particularly intrusive; regional antago-

nism to Iran is high in the wake of active Iranian

intervention in Syria and neighboring countries. If China

begins to intervene on a global basis, it may be next.

Implications for Political Science and

World Politics

Our analysis also raises more general questions, not

limited to the study of anti-Americanism, which are

relevant to the study of what we have called social

globalism. Social media create a remarkable window not

only into attitudes, as expressed by individuals, but

into discourses. We can well imagine analyses that

would try to identify different discourses on a s et of

topics, and their connections, and their disconnects.

Twitter users with more moderate or extreme views

may have differ ent networks of whom they follow and

who follows them.

These discourses are socially constructed a nd

have their own dynamics. There may be considerable

path-dependence, with initial themes helping to de-

termine future themes: this is another set of questions

worth analy zing. Soci al scient ists sho uld build on

excellent existing work about how these discourses

affect politics in the street or policy-making in govern-

ments.

To what ex tent do these discourses shape so cial

movements and how closely are they responded to by

policy-makers? Furthermore, how does s ocial media

engagement relate to other forms of expression, such

as communications to leaders, commentary in newspapers,

or informal discussions?

More broadly, political discussion

includes people saying they like something, repeating what

they heard, and replying to each other. But these actions on

Twitter might engender substantively different dynamics

from face-to-face interaction. The politics of social media

discourses have created a subject that is ripe for innovative

social scientiﬁcanalysis.

March 2015

Vol. 13/No. 1 69

Beyond the opportunities that social media discourses

provide for social science, they raise major questions for

the study of contemporary world politics. As we note in

the introduction, the fact that voices of ordinary people

can be heard, worldwide, is a new feature of contempo-

rary globalism. It enables people with minority views

within their own communities to ﬁnd a broader com-

munity of sympathizers and to engage in debate. It also

enables militant organizations, such as the Islamic State of

Iraq and Syria (ISIS), to recruit, on a global basis, seeking

to entice people who have never known anyone from ISIS

to join these organizations. These organizations become

focal points—both on social media and in the world of

inter-personal politics and military action—for the expres-

sion of dissent and alienation. Prevailing high levels of

anti-Americanism on social media may provide validation

for decisions by some individuals to act against the United

States, by joining ISIS or otherwise.

Military and organizational strength—whether on the

part of Iran or the United States—does not enable modern

states to control expressions on social media sites abroad,

although some states such as China are quite effective in

controlling domestic media sites.

On the contrary, the

exertion of military power by these states to control events or

inﬂuence political outcomes generates opposition in society

that at least partially counterbalances their hard power. The

United States or Iran may be able to inﬂuence the state-run

media of sympathetic countries, but the extensive horizontal

communication on social media reduces the dominance of

state-run media and the efﬁcacy of strategies that rely on

them. Old-style intervention threatens to generate hostility

that reverberates through social media, which ampliﬁes

reactions to every military action.

The dynamic and expansive character of social global-

ism, fostered by the internet and social media, does not

imply that states have suddenly “lost control.” As Stephen

Krasner has argued, states remain the most important

actors in world politics and have historically been resilient,

with state activity increasing along with economic and

social interdependence.

But social media, and the

discordant discourses that they engender, do generate

new challenges for states, and particularly the United

States, as they seek to develop strategies for effective action

in world politics.

Online Supplemental Materials: “Anti-

Americanism and Anti-Interventionism

in Arabic Twitter Discourses”

• Overview: Aggregate Twitter Volume

• The Hopkins-King “ReadMe” Method

• London Monitor Results

• Monitor Training Details

http://dx.doi.org/10.7910/DVN/28171

Notes

1 Keohane and Nye 2000. The authors differentiate

economic, military, environmental, and social

globalism.

2 Keohane 2001.

3 Howard, 2011, Gainous and Wagner 2014, Gainous

and Wagner 2013, Lynch 2011, Tarrow 2013,

Bennett and Segerberg 2014, Norris 2001.

4 Katzenstein and Keohane 2007.

5 Chiozza 2009, table 3.1, 55; ﬁgures 4.2 and 4.3, 98

and 101.

6 Katzenstein and Keohane 2007.

7 Dubai School of Government 2013.

8 Esposito 2011.

9 Jamal 2012a.

10 See Mossberger et al. 2008.

11 Tarrow 2013; Gainous and Wagner 2014; O’Conner

et al. 2010.

12 We note that since tweets are attributed to their

authors, some tweeters may exhibit social desirability

bias—expressing views that they want to be seen to

have, rather than their own private views.

13 Joffe 2006, Ajami 2003, Lewis 1990, Huntington

1996.

14 Blaydes and Linzer 2012.

15 Gerges 1996, Jamal 2012b, Telhami 2002, Tessler

2003, Baxter and Akbarzadeh 2008, Makdisi 2002,

Khalidi 2004, Dawisha 1985, Esposito 1998, Haddad

1996.

16 Chiozza 2007, Tessler 2003.

17 Hopkins and King 2010.

18 See Grimmer and Stewart 2013. For a more focused

introduction to computer assisted methods of text

analysis for comparative politics, see Lucas et al.

forthcoming.

19 Blei 2012; Blei and Lafferty 2007; Blei; Ng, and

Jordan, 2003; Robert et al. 2014.

20 Laver, Benoit, and Garry 2003; Hopkins and King

2010.

21 See http://www.crimsonhexagon.com/.

22 For details please refer to the online supplementary

materials.

23 The user is limited to 5,000 characters for keyword

restrictions, requiring the user to make judgments

about which keywords are more or less important;

these judgment calls are discussed in the online

supplementary materials.

24 We placed tweets that do not ﬁtintoanycategory

into an “irrelevant” category, so that the frequencies

for the on-topic categories are not affected by

this trafﬁc. Refer to the online supplementary

materials.

25 See Grimmer and Stewart 2013.

26 Jamal 2012b; Chiozza 2007; Lynch 2007.

70 Perspectives on Politics

Articles | Anti-Americanism and Anti-Interventionism in Arabic Twitter Discourses

27 We used veriﬁcation methods as discussed in the

online supplementary materials. As explained there, if

our results are biased, they probably overestimate the

volume of the Positive Political category—which is,

relatively speaking, quite small.

28 This is evident over time as well. Throughout our time

period the negative political conversation almost

always dominates the other categories.

29 These monitors differ from the General AA monitor in

three key ways. The time range for each of these

monitors, except the Syria monitor, is signiﬁcantly

shorter. The category creation process for the event

monitors was often inductive rather than deﬁned ex

ante, as was the case in the general monitor. We

wanted to look at the conversation surrounding these

events in detail rather than applying a speciﬁc pre-

deﬁned set of categories to them. Finally, we either

used sets of keywords that were speciﬁc to the event

(Boston, Sandy, and Innocence of Muslims monitors)

or a combination of U.S. keywords from the General

AA monitor and event-speciﬁc keywords (Egypt and

Syria monitors) rather than just U.S. keywords.

30 For the United States, we used the general list of anti-

American keywords with some minor alterations; for

Egypt, we used four terms: “Egypt/Egyptian,”

“Morsi,”“The Brotherhood,” and “Sisi” (the

Commander-in-chief of Egypt’s armed forces). Refer

to the online supplementary materials for details.

31 Our ability quickly to analyze Twitter reactions to the

Syrian crisis contrasts with the fact that, as far as we

know, there have been no polls conducted in Arabic

anywhere on this issue.

32 We also ran a Libya monitor from February 1 to

October 31, 2011, during the NATO campaign that

overthrew the Qaddaﬁ regime; details for this monitor

can be found in the online supplementary materials.

Caution is required since the number of tweets was

dramatically lower then: apart from news and neutral

items, fewer than 20,000 tweets mentioned Libya and

the United States. The ratio of negative to positive

comments was over 3:1, indicating high negativism

toward U.S. intervention even though the United

States was aligned with some Arab countries, such as

Qatar in opposing the Qaddaﬁ dictatorship. Yet this

ratio is much lower than the 30:1 ratio in our Syria

monitor. U.S. alignment with Arab regimes seems to

have ameliorated anti-American sentiment but not to

have reversed it.

33 For the list, refer to the online supplementary materials.

34 Although there is little directly negative trafﬁc in the

Innocence of Muslims monitor, the General AA

monitor shows a spike in negative trafﬁc during this

time period. We checked this discrepancy by creating

a monitor using the same keywords as the General AA

monitor, around the spike between September 10–17,

in which we found quite a bit of negative trafﬁc, much

of which is political.

35 Katzenstein and Keohane 2007.

36 Refer to the online supplementary materials for these

results.

37 Lynch 2007, 223.

38 Berger 2014.

39 Jamal 2012b.

40 See also Katzenstein and Keohane 2007, 34–35.

41 Our inference that Arabic anti-Americanism princi-

pally reﬂects fear and distrust of the impact of the

United States on the Arab world is admittedly a matter

of interpretation, and subject to considerable uncer-

tainty, but it seems to us a plausible interpretation and

consistent with polling data.

42 Katzenstein and Keohane 2007, 317.

43 Tarrow 2013; Gainous and Wagner 2014; Howard

2011.

44 Cramer Walsh 2007, Lee 2002, Lynch 2011.

45 King, Pan, and Roberts 2013.

46 Krasner 1999, 223.

References

Ajami, Fouad. 2003. “The Falseness of Anti-Americanism.”

Foreign Policy 138: 52–61.

Almond, Gabriel A. 1950. The American People and Foreign

Policy. New York: Harcourt, Brace and Company.

Baxter, Kylie, and Shahram Akbarzadeh. 2008. US Foreign

Policy in the Middle East: The Roots of Anti-Americanism.

London: Routledge.

Bennett, Lance, and Alexandra Segerberg. The Logic of

Connective Action: Digital Media and the Personalization

of Contentious Politics. New York: Cambridge

University Press.

Berger, Lars. 2014. “Foreign Policies or Culture: What

Shapes Muslim Public Opinion on Political Violence

against the United States?” 2014. Journal of Peace

Research 51(6): 782–796.

Blaydes, Lisa, and Drew A. Linzer. 2012. “Elite

Competition, Religiosity, and Anti-Americanism in

the Islamic World.” Am erican Political Science Review

106(2): 225–43.

Blei, David M. 2012. “Probabilistic Topic Models.”

Communications of the ACM 55(April): 77–84.

Blei, David M., and John D. Lafferty. 2007. “A Correlated

Topic Model of Science.” Annals of Applied Statistics

1(June): 17–35.

Blei, David M., Y. Ng Andrew, and Michael I. Jordan.

2003. “Latent Dirichlet Allocation.” Journal of Machine

Learning 3(March): 993–1022.

Chiozza, Giacomo. 2007. “Disaggregating

Anti-Americanism: An Analysis of Individual Attitudes

toward the United States.” In Anti-Americanisms in

World Politics, ed. Peter J. Katzenstein and Robert O.

Keohane. Ithaca, NY: Cornell University Press.

March 2015

Vol. 13/No. 1 71

. 2009. Anti-Americanism and the American World

Order. Baltimore, MD: Johns Hopkins University

Press.

Cramer Walsh, Katherine 2007. Talking about Politics:

Informal Groups and Social Identity in American Life.

Chicago, IL: University of Chicago Press.

Dawisha, Adeed. 1985. “Anti-Americanism in the

Arab World: Memories of the Past and Attitudes of

the Present.” In Anti-Americanism in the

Third World: Implications for U.S. Foreign Policy, ed.

Alvin Z. Rubinstein and Donald Eugene Smith.

New York: Praeger.

Dubai School of Government. 2013. “Arab Social Media

Report,” http://www.arabso cialmediareport.com/

UserManagement/PDF/ASMR_5_Report_Final.pdf

Esposito, John L. 1998. Islam and Politics. Syracuse:

Syracuse University Press.

. 2011. “Arab Youth Want Democracy, not

Theocracy.” CNN, February 28, http://www.cnn.com/

2011/OPINION/02/28/protests.democracy.islam/.

Gainous, Jason, and Kevin Wagner. 2013. “Digital

Uprising: The Internet Revolution in the Middle

East. Journal o f Information Technology and Politics

10: 3.

. 2014. Tweeting to Power: The Social Media

Revolution in American Politics. London: Oxford

University Press.

Gerges, Fawaz A. 1996. “The 1967 Arab-Israeli W ar:

U.S. Actions and Arab Perceptions.” In The Middle

East and the United States: A Historical and Political

Reassessment,ed.DavidW.Lesch.Boulder,CO:

Westview Press.

Grimmer, Justin, and Brandon M. Stewart. 2013. “Text as

Data: The Promise and Pitfalls of Automatic Content

Analysis Methods for Political Texts.” Political Analysis

21 (January): 1–31.

Haddad, Yvonne Yazbeck. 1996. “Islamist Perceptions o f

U.S. Policy in the Middle East.” In The Middle East

and the United States: A Historical and Political

Reassessm ent, ed. David W. Lesch. Boulder, CO:

Westview Press.

Hopkins, Daniel, and Gary King. 2010. “A Method of

Automated Nonparametric Content Analysis for Social

Science.” American Journal of Political Science 54(1):

229–47.

Howard, Philip. 2011. The Digital Origins of Dictatorship

and Democracy: Information Technology and Political

Islam. London: Oxford University Press.

Huntington, Samuel P. 1996. The Clash of Civilizations

and the Remaking of World Order. New York: Simon &

Schuster.

Jamal, Amaney. 2012a. “The Youth and the Arab Spring:

Cohort Differences and Similarities.” Middle East Law

and Governance 4(1): 168 –88.

. 2012b. Of Empires and Citizens: Pro-American

Democracy or No Democracy at All? Princeton, NJ:

Princeton University Press.

Joffe, Josef. 2006. Überpower: The Imperial Temptation of

America. New York: W.W. Norton & Co.

Katzenstein, Peter J., and Robert O. Keohane 2007.

“Varieties of Anti-Americanism: A Framework for

Analysis.” In Anti- Americanisms in World Politics, ed.

Peter J. Katzenstein and Robert O. Keohane. Ithaca,

NY: Cornell University Press.

Keohane, Robert O. 2001. “Governance in a Partially

Globalized World.” American Political Science Review

95(1): 1–13.

Keohane, Robert O., and Joseph S. Nye Jr. 2000.

“Govern ance in a Globalizing World.” In Governance in

a Globalizing World, ed. Joseph S. Nye Jr., and John D.

Donahue. Washington, DC: Brookings Institution.

Khalidi, Rashid. 2004. Resurrecting Empire: Western

Footprints and America’s Perilous Path in the Middle East.

Boston: Beacon Press.

King, Gary, Jennifer Pan, and Margaret E. Roberts. 2013.

“How Censorship in China Allows Government

Criticism but Silences Collective Expression.” American

Political Science Review 107(2): 326–43.

Krasner, Stephen D. 1999. Sovereignty: Organized

Hypocrisy. Princeton, NJ: Princeton University Press.

Laver, Michael, Kenneth Benoit, and John Garry. 2003.

“Extracting Policy Positions from Political Texts Using

Words as Data.” American Political Science Review 97

(2): 311–31.

Lee, Taeku 2002. Mobilizing Public Opinion: Black In-

surgency and Racial Attitudes in the Civil Rights Era.

Chicago, IL: University of Chicago.

Lewis, Bernard. 1990. “The Roots of Muslim Rage.”

Atlantic Monthly 266(3): 47–60.

Lucas, Christopher, Richard Nielsen, Margaret Roberts,

Brandon Stewart, Alex Storer, and Dustin Tingley.

forthcoming. “Computer Assisted Text Analysis for

Comparative Politics.”Political Analysis .

Lynch, Marc. 2007. “Anti-Americanism in the Arab

World.” In Anti-Americanisms in World Politics, ed.

Peter J. Katzenstein and Robert O. Keohane. Ithaca,

NY: Cornell University Press.

. 2011. “After Egypt: The Limits and Promise of

Online Challenges to the Authoritarian Arab State.”

Perspectives on Politics 9(2): 301–310.

Makdisi, Ussama. 2002. “‘Anti-Americanism’ in the

Arab World: An Interp retation of a Brief History.”

Journal of American History 89(Septemb er):

538–57.

Mossberger, Karen, Caroline J. Tolbert, and Ramona S.

McNeal. 2008. Digital Citizenship : The Internet,

Society, and Participation. Cambridge, Mass: MIT

Press.

72 Perspectives on Politics

Articles | Anti-Americanism and Anti-Interventionism in Arabic Twitter Discourses

Norris, Pippa. 2001. Digital Divide: Civic Engagement,

Information Poverty and the Internet Worldwide. New

York: Cambridge University Press.

O’Connor et al. 2010 – see n.11

Robert, Margaret, Brandon Stewart, Dust in Tingley,

Christopher Lucas, Jetson L eder-Luis, Shana

Gadarian, Bethany Albertson, and David Rand.

2014. “Structural Topic Models for Open-Ended

Survey Responses.” American Journal of Political

Science 58(4): 1064–82.

Tarrow, Sydney 2013. The Language of Contention:

Revolutions in Words, 1688–2012. New York:

Cambridge University Press.

Telhami, Shibley. 2002. The Stakes:

America and the Middle East.Boulder,CO:

Westview Press.

Tessler, Mark. 2003. “Arab and Muslim Political

Attitudes: Stereotypes and Evidence from Survey

Research.” International Studies Perspectives 4(May):

175–81.

March 2015

Vol. 13/No. 1 73