Attack When the World Is Not Watching? Conflict ∗

Attack When the World Is Not Watching?
International Media and the Israeli-Palestinian
Conflict∗
Ruben Durante†
Ekaterina Zhuravskaya‡
October 2015
A BSTRACT
Policy makers may strategically time unpopular measures to coincide with other
newsworthy events that distract the media and the public, so as to minimize the
political cost of these measures. We test this hypothesis in the context of the
recurrent Israeli-Palestinian conflict. Combining daily data on attacks on both
sides of the conflict with data on the content of evening news for top U.S. TV
networks, we show that Israeli attacks are more likely to be carried out when
U.S. news are expected to be dominated by important (non-Israel-or-Palestinerelated) events on the following day. Several findings indicate that this association is a result of the strategic behavior of Israeli authorities: i) only attacks
that bear higher risk of civilian casualties are timed to newsworthy events; ii)
attacks are timed to events that are predictable, and iii) the timing of Israeli retaliations against Palestinian attacks is related to U.S. news only in periods of
less intense fighting, when retaliation is less urgent. Based on comprehensive
content analysis of conflict-related news, we document that the strategic timing
of Israeli attacks is aimed at minimizing news coverage on the following day
because next-day news stories are especially charged with negative emotional
content. We find no evidence of strategic timing for Palestinian attacks.
Keywords: Conflict, mass media, accountability, strategic timing
∗ We
thank Morten Bennedsen, Tim Besley, Matteo Cervellati, Antonio Ciccone, Francesco D’Acunto,
Oeindrila Dube, Pascaline Dupas, Irena Grosfeld, Fred Finan, Sophie Hatte, Emeric Henry, Roberto Galbiati,
Guy Laroque, Tommaso Oliviero, Daniele Paserman, Jacob Shapiro, David Stromberg, Uwe Sunde, Nico
Voigtlaender, as well as participants in the 2014 Workshop on Media Economics at CSEF, the 2015 Workshop
on Conflict at the University of Munich, and the 2015 Conference on the “Political economy of Conflict”
(Villars), and seminar participants at Sciences Po, the London School of Economics, INSEAD, Mannheim,
Aix-Marseille, Chicago Harris, UCLA Anderson, UCSB, Queen Mary, and Stockholm University, for helpful
comments. We thank Monica Consorti, Florin Cucu, Etienne Fize, Orcan Sakalli, and especially Iván Torre
for outstanding research assistance. We are grateful to the Israeli Information Center for Human Rights in the
Occupied Territories (B’Tselem), the United Nations Office for Coordination of Humanitarian Affairs in the
Occupied Palestinian Territory (UNOCHA), and the Center for Research on the Epideomiology of Disasters
(CRED) for sharing data.
† Sciences Po and CEPR. E-mail: [email protected].
‡ Paris School of Economics (EHESS) and CEPR. E-mail: [email protected].
1. I NTRODUCTION
Governments are accountable to the extent that the public is informed about their policies.
Mass media ensure accountability by informing citizens about government actions (e.g.,
Besley and Prat, 2006; Snyder and Stromberg, 2010). Yet, how effectively mass media inform the public depends, among other things, on the presence of other newsworthy events
that may crowd out the news coverage of governments’ actions (Eisensee and Stromberg,
2007). To minimize negative publicity policy-makers may strategically manipulate the timing of their unpopular actions so that they coincide with particular moments when the mass
media and the public are distracted by other events.
On July 13 1994, the day Italy qualified to the final of the FIFA World Cup, the government of Silvio Berlusconi passed an emergency decree - the so-called “save-the-thieves”
decree - that allowed hundreds of corrupt politicians to avoid jail sentence. On August 8
2008, day of the opening ceremony of the Beijing Summer Olympics, Russia initiated the
invasion of Georgia. On July 8 2014, day of the FIFA World Cup semifinal between Brazil
and Germany, Israel launched operation “Protective Edge” in Gaza. In this paper we argue that the timing of these seemingly unrelated events is hardly a coincidence.1 Benefiting
from the occurrence of other newsworthy events to minimize the impact of releasing harmful
information is also a well-known strategy of political spin doctors.2
This paper tests whether politicians chose the timing of their policy actions strategically to minimize their news coverage by focusing on the timing of military operations
during an on-going conflict. Military operations often receive negative international publicity, especially when they result in civilian casualties. If the attacker cares about international
public opinion and if the conflict is recurrent, this is a good testing ground for strategic timing considerations as time series data exist on both military operations (i.e., the unpopular
1
2
Online appendix figure A.1 presents an example of the front pages of three national and international newspapers that covered these political events. Noticeably, the space allocated to these events was substantially
smaller than that devoted to the sports events they coincided with.
This is epitomized by the notorious statement of the former UK Labour party’s spin doctor, Jo Moore, who, in a leaked memo sent to her superiors on the afternoon of 9/11,
said that it was “a very good day to get out anything we want to bury”, [i.e., bad
news]. See http://www.telegraph.co.uk/news/uknews/1358985/Sept-11-a-good-day-to-bury-bad-news.html
(accessed on July 7, 2015) and http://www.theguardian.com/politics/2001/oct/10/uk.Whitehall (accessed on
July 7, 2015).
1
action) and the occurrence of other newsworthy events.
Our focus is on the Israeli-Palestinian conflict. Both Israelis and Palestinians care
about international public opinion and recognize the importance of international media in
shaping it. Since the 1970s, the Israeli government places a special emphasis on the efforts to
project a positive image of Israel and of the Israeli army abroad, a policy known as hasbara,
Hebrew for “explanation.” This policy encompasses public diplomacy, the work of Israeli
government with international journalists in Israel, and the presence of Israeli advocates on
social media. Arguably, nothing has a stronger negative impact on the international public
opinion of any side in a military conflict than the presence of civilian casualties as a result of
its military actions. The Israeli-Palestinian conflict is no exception to this rule; and both sides
of the conflict recognize it and acknowledge the role of the media in informing the public
about the conflict. This was exemplified by Israeli Prime Minister Benjamin Netanyahu who
in an interview to CNN on July 20 2014, while discussing the heart-wrenching images of
civilian Palestinian victims in Gaza, stated:“[Hamas] wants to pile up as many civilian dead
as they can... they use telegenically dead Palestinians for their cause.”3
This paper examines whether either side of the conflict chooses the timing of its
attacks to coincide with other newsworthy international events so as to minimize the negative
impact of their actions on international public opinion.
As for Israel, our hypothesis is that Israeli authorities want to avoid international
media coverage of their military operations especially if and when these may yield civilian
casualties. News about civilian casualties triggers severe criticism from international organizations and human rights activists and creates risk of alienation of Israel in the international
arena and of diminished popular support of key partner countries such as the U.S. Importantly, domestic news coverage of military incidents leading to civilian casualties can also be
costly for the domestic popularity of the incumbent Israeli government, particularly among
certain segments of the voting population.
As for Palestinian militant groups, their incentives are a priori less clear cut due to
the countervailing effects of international and domestic publicity. On the one hand, international coverage of Palestinian terrorist attacks sways international public opinion in favor
3
http://cnnpressroom.blogs.cnn.com/2014/07/20/netanyahu-to-cnns-wolf-blitzer-i-support-taking-whateveraction-is-necessary-to-stop-this-insane-situation/(accessed on January 19, 2015).
2
of Israel, which serves against the Palestinian cause. On the other hand, however, domestic
coverage of the attacks against Israel may foster popular support for terrorist organizations
among the Palestinian population, and make it is easier for them to attract new recruits.
Furthermore, terrorists are usually interested in higher domestic publicity because it is associated with higher devastation and fear, which presumably is one of the main goals of the
terrorist attacks. Thus, in the case of Palestinians, the hypothesis is a priori ambiguous. Furthermore, Palestinian terrorist attacks are carried out by a number of independent factions,
which creates a possibility of coordination problems between them.
We test these hypotheses using daily time series data on the occurrence and severity
of both Israeli military operations in the West Bank and Gaza strip, and attacks by Palestinian militant groups on the Israeli territory between 2000 and 2011. These data, which
were compiled by two independent human rights organizations, include information on the
number of attacks (and the resulting number of casualties) carried out by each side on each
day. We combine these data with a measure of the presence of other newsworthy events
on international media that may crowd out news coverage of the Israeli-Palestinian conflict.
In particular, we use the direct analogue of the news pressure variable first proposed by
Eisensee and Stromberg (2007). We compute news pressure as the time devoted to the top
three stories, not related to either Israel or Palestine, featured in the evening news in three
U.S. TV networks, NBC, ABC, and CBS. This measure captures the presence of important
news stories that may crowd-out the coverage of the conflict on each given day.
As a starting point, we confirm that important attacks in the Israeli-Palestinian conflict are often covered by U.S. media and that, as the work of Eisensee and Stromberg (2007)
suggests, the probability that the attacks are covered, conditional on their severity, decreases
with an increase in the news pressure, due to the occurrence of competing newsworthy events
on the same day. Furthermore, using data on Google searches, we document that TV coverage of Israeli and Palestinian attacks, conditional on their severity, increases the public’s
attention to the conflict: the daily volume of Google searches for the topic “Israeli-Palestinian
conflict” increases by 12% with each additional news story on the conflict featured on U.S.
prime-time TV news.
We then examine how the timing of the attacks by both Israelis and Palestinians is
3
related to daily news pressure. We relate the incidence of a deadly attack on a given day
by each side of the conflict and its intensity, measured by the number of fatalities, to the
news pressure recorded on each day around the attack. Our empirical strategy is based on
the assumption that news in the U.S. that are not related to Israel or Palestine - and hence the
news pressure variable - are exogenous to actions by both Israelis and Palestinians. We find
that the likelihood of deadly military attacks by Israeli forces against Palestinians, as well as
the number of victims they cause, is positively and significantly related to the level of news
pressure on the day after the attack is carried out. This suggests that Israeli authorities may
chose the timing of their attacks strategically to minimize negative international publicity. In
contrast, we find no evidence that attacks by Palestinian militant groups are timed to U.S.
news pressure. The relationship between the timing of Israeli attacks and news pressure is
not driven by the presence of outliers, and is robust to controlling for the retaliation motive of
the Israeli attacks (with dummies for the presence of a Palestinian attack in previous days),
for seasonality (with calendar month and day of the week fixed effects), and for the general
dynamics in conflict intensity (with year fixed effects). Finally, it is robust to accounting for
serial correlation by different means, i.e., including lags and leads, adjusting standard errors
to Newey-West estimation, or clustering them by month-year.
If the relationship between Israeli attacks and U.S. news pressure the following day
is, indeed, driven by strategic timing considerations on the part of Israeli authorities, several
testable implications arise. First, one should expect only those military operations that are
likely to generate negative publicity to be timed to news pressure. As the main source of
negative publicity is the presence of civilian casualties, only attacks that carry the risk of
civilians being affected should be subject to strategic timing. We test this prediction using
three alternative measures of whether an attack has an ex ante higher risk of civilian casualties: 1) the presence or absence of fatalities (ex post), 2) the type of weapon used, i.e.,
heavy weapons vs. light ammunition, and 3) whether the attack took place in a more vs. less
densely populated area. While only 34.4% of all Israeli military operations result in deaths,
this is the case for 68% of the attacks involving heavy weapons (e.g., missiles or artillery
shells), and for 49.6% of the attacks carried out in densely populated areas. We find that
the relationship between occurrence and severity of attacks and news pressure is significant
4
only for operations that result in fatalities, involve the use of heavy weapons, and are carried
out in densely populated areas. These findings support the prediction that only operations
that are more likely to generate negative publicity due to higher risk of civilian casualties are
strategically timed to minimize media coverage.
Second, some special military operations, such as targeted killings of terrorists who
are planing attacks against Israel, are considered of primary importance for Israeli defense
forces and and are conducted whenever a proper opportunity to hit the target arises. Consistent with the urgency of these attacks, we find no statistically significant relationship between
the timing of the special targeted-killing operations and news pressure.
Third, in order to chose the timing of military operations so that they occur one
day before news pressure becomes high, Israeli authorities should be able to predict it. As
the news pressure measure is based on all news stories that do not refer to either Israel or
Palestine, high news pressure could in principle be due to the occurrence of both predictable
and unpredictable newsworthy events (e.g., notable political or sports events vs. natural
disasters). To test whether Israeli attacks are timed to the predictable part of news pressure,
we analyze the text of the headlines of the news stories featured on days with very high
news pressure separately for days when an Israeli attack occurred the day before and days
with no Israeli attack on the day before. We find that in days with high news pressure but
no Israeli attacks on the day before 46% of the headlines contained words related to natural
or man-made disasters compared to only 12% for the high-news-pressure days preceded
by an Israeli attack. We also find that on high-news-pressure days preceded by an Israeli
attack news pressure is more likely to be driven by clearly predictable events, e.g., U.S.
holiday shopping, than on high-news-pressure days not preceded by an attack. Furthermore,
using information on all natural and man-made disasters that occurred in the U.S. during
our sample period, we decompose daily news pressure into an unpredictable component,
driven by news on disasters that occurred that day, and a predictable component, driven by
all other news. We show that the timing and severity of Israeli attacks is not related to the
unpredictable (disaster-driven) component of news pressure, but is significantly related to
its predictable (disaster-free) component. Moreover, the relationship of the timing of the
Israeli attacks with the disaster-free news pressure is stronger than that with the overall news
5
pressure. Taken together, these results are consistent with the hypothesis that Israeli attacks
are timed to news pressure associated with predictable rather than unpredictable events.
Fourth, the timing of retaliation may also be affected by strategic considerations, but
only when the urgency of retaliation is relatively low, e.g., when conflict is not very intense.
Our data span across two distinct periods with sharply different conflict intensity: 1) the Second Intifada period (September 28, 2000 - February 8, 2005), characterized by very intense
fighting with deadly attacks on both sides occurring on two out of each three days; and 2) the
post-Intifada period (from February 9, 2005 to the end of our sample period), characterized
by a much lower frequency of attacks, in general, and of Palestinian attacks, in particular.
Given the importance of immediate retaliation during the Second Intifada, we expect little
news-driven adjustment of the timing of retaliation during this period. In contrast, in the
post-Intifada period, when in general retaliation took a few days, we expect concerns about
negative publicity to be more important for the timing of retaliation. This prediction also
finds support in the data: the timing of Israeli retaliation attacks is significantly affected by
news pressure only in the post-Intifada period.4
Finally, we examine the mechanism behind the strategic timing effect. In particular,
we test alternative explanations for why Israel should time its attacks to the news pressure
on the following day rather than on the same day. First, we test and reject the hypothesis that
news coverage of the conflict on U.S. TV is “slow,” i.e., that Israeli attacks are more likely
to appear on the news on the day after the attack than on the same day. We find that, on
average, twice as many stories about the Israeli-Palestinian conflict appear on the news on
the day of an Israeli attack than on the next day. This suggests that it is not the mere fact that
the attacks are covered in the news that the Israeli government is trying to avoid by timing
its attacks strategically, but, rather, a particular type of coverage of the attacks. Using data
on the length of each news story, we examine how much time is devoted by U.S. news media
to stories about Israeli attacks on the same and on the following day. Despite the fact that
Israeli attacks are significantly more likely to appear on the news on the day of the attack
4
Another much shorter episode of a very intense violence broke out during the Gaza War, also known as the
“Operation Cast Lead,” between December 27, 2008 and January 18, 2009. The main results of the paper
are very similar irrespective of whether we exclude the Gaza War period from the sample or not. We report
the results for the full sample in the online appendix.
6
than on the following day, the average length of conflict-related stories is not statistically
different between the same and the next day. This is due to the fact that, provided an attack
is covered, conflict-related news appearing on the day after the attack are, on average, longer
than those appearing on the same day.
To investigate differences in the actual content of news stories about Israeli attacks
appearing on TV on same- vs. the next-day, we coded all videos related to Israeli-Palestinian
conflict on prime-time news on two networks, NBC and CNN, between 2000 and 2011
(available from the Vanderbilt Television News Archive).5 We collected data on the type
of information included in each story and the form in which this information was presented.
Out of total of 582 videos on the Israeli-Palestinian conflict, 499 reported on Israeli or Palestinian attacks occurred on the same or on the previous day, 324 reported on Israeli attacks but
also spoke about Palestinian attacks, and 201 were about a particular Israeli attack and did
not mention any Palestinian attack. Regardless of how we restrict the sample, we find that
the type of coverage of Israeli attacks differs substantially (and statistically significantly) between same-day and next-day reports. In particular, while the same-day and next-day news
stories are equally likely to report information on the number of victims, news stories that
appear on the day after the attack are much more likely to present personal life stories of the
civilian victims and include interviews with their relatives or friends. Furthermore, next-day
coverage is significantly more likely to include emotionally-charged visuals of burial processions and scenes of mourning. In contrast, official reactions by the Israeli authorities about
the incident are less likely to be reported on the next day than on the same day.
Anecdotal evidence suggests that these differences are driven both by technical aspects of news reporting in conflict zones, and by local customs and traditions specific to the
Middle East. First, when Israel strikes against Palestinian militants, there are usually no international reporters in the vicinity of the target area, and it takes time for them to get on site.
Second, even when reporters are nearby, it is often dangerous to reach the actual location
of the attack due to the risk of follow-up attacks; for the same reason (potential) witnesses
of the attack also hide so that even if journalists arrive on site, finding information quickly
can be difficult. Third, it takes time for reporters to uncover the details about what actually
5
We focus on these two networks because they are the only ones for which daily video footage in currently
available in the Vanderbilt Archive.
7
happened. Forth, the local tradition dictates that victims are buried the day after the death.
The burial ceremony presents a relatively easy and safe opportunity to produce emotionallycharged footage, as the ceremony takes place in open air and is generally widely attended by
the local population, which is an insurance against a possibility of an Israeli attack. Furthermore, it provides reporters with a good opportunity to access information about the victims,
as people who personally knew them participate in the ceremony.
Both the actual informational content (statistics vs. personal stories) and the form in
which it is presented (narrative vs. visual), have very different effect on viewers. As is wellestablished in the cognitive and social psychology literatures, people react more strongly to
personal stories than to statistics and dry facts (e.g., Borgida and Nisbett, 1977; Martin and
Powers, 1982; Wilkins, 1983), and information transmitted only by means of words is less
likely to be retained than information accompanied by images (e.g., Mandl and Levin, eds,
1989; Houghton and Willows, eds, 1987a,b; Houts et al., 2006).
Taken together, our findings suggest that Israeli authorities behave strategically in
timing their attacks to predictable international newsworthy events in order to minimize negative publicity abroad, and that such strategy is sophisticated in that it takes into account both
the technology of news reporting in war areas and the cognitive psychology of information
transmission and retention.
Our research relates to several strands of literature. First, our study contributes to
the literature on political accountability and mass media (see, for instance, surveys by Prat
and Stromberg, 2013; Sobbrio, 2014). To the best of our knowledge, our paper is the first
to provide systematic evidence that policy makers act strategically to minimize the negative
impact of media reporting of their unpopular actions by manipulating the timing of their
actions to coincide with other newsworthy events. Our work contributes to this literature also
by documenting that the qualitative aspects of news coverage matter, i.e., the way news are
presented is important for how the message is received and processed by viewers. Second,
our paper contributes to a growing body of work on the role of mass media in conflicts.
While economic contributions on the topic have largely focused on estimating the persuasion
effects of mass media on conflict (DellaVigna et al., 2014; Yanagizawa-Drott, 2014; Adena
et al., 2016), historians and political scientists alike have directly addressed the issue of how
8
media may constrain behavior in a conflict environment without, however, providing any
systematic evidence of the kind presented here (see, for instance, a collection of articles in
Seethaler et al., eds, 2013). Third, our paper contributes to the literature on strategic behavior
in conflicts in general (see Blattman and Miguel, 2010; Jackson and Morelli, 2009) and in
the Israeli-Palestinian conflict, in particular (e.g., Jaeger et al., 2010; Jaeger and Paserman,
2008, 2006, 2009). Finally, our work relates to a long-standing debate in the finance literature
on whether company managers strategically choose the timing of the release of information
by publishing negative earnings reports in periods of low market attention (e.g., Patell and
Wolfson, 1982; Damodaran, 1989; Doyle and Magilke, 2009; DellaVigna and Pollet, 2009;
deHaan et al., 2014).
The remainder of the paper is organized as follows. Section 2 provides background
information on the Israeli-Palestinian conflict. Section 3 describes the data. Section 4
presents the results on the relationship between the timing of attacks and news pressure
and section 5 provides evidence on the mechanism. Section 6 illustrates the magnitudes with
a simple counterfactual exercise. Section 7 concludes.
2. BACKGROUND : I SRAELI -PALESTINIAN CONFLICT
The Israeli-Palestinian conflict is long-lasting. Fighting on both sides continues with varying
intensity since 1948. During our observation period - from September 29, 2000 to November 24, 2010 - the conflict resulted in 7690 fatalities, of which 6401 Palestinians and 1289
Israelis.6 The three weeks of the Gaza War (or operation “Cast Lead”), between December
27, 2008 and January 18, 2009, was the period of most intense fighting accounting for 17.8%
of the fatalities recorded over the entire sample period, of which 1349 Palestinians and 19 Israelis. Figure 1 presents the monthly number of fatalities on both sides of the conflict for the
entire period (plot 1), and excluding the Gaza War (plot 2). Once the Gaza War is excluded,
it becomes evident that our data cover two phases of the conflict that differ substantially in
terms of the intensity of fighting: the Second Intifada (from the beginning of the sample
period to February 8, 2005) and the post-Intifada period (from February 9, 2005 onward).7
6
7
The fatalities tall excludes the suicide bombers.
The Second Intifada ended with an agreement signed by Mahmoud Abbas and Ariel Sharon at the Sharm
el-Sheikh Summit.
9
Horizontal lines represent the average monthly number of fatalities on both sides of the conflict separately for the two periods, and illustrate the much higher intensity of conflict during
the Intifada. We exclude the Gaza war period (as an outlier) from our baseline sample and
verify that the main results are robust to using the full sample with the Gaza war.
Two aspects are especially important for our story. First, the Israeli-Palestinian conflict is characterized by a relatively large number of civilian victims, and particularly children: 47% of all Palestinians casualties and 68% of all Israeli casualties were civilians;
21% of Palestinian casualties and 11% Israeli casualties were children. Second, the IsraeliPalestinian conflict attracts a lot of attention by international media: over our observation
period, 34.6 minutes were devoted to the conflict on an average month by the evening news
on NBC, ABC, and CBS.
3. DATA SOURCES AND MAIN VARIABLES
In our empirical analysis we use daily data on: i) all stories that appeared on U.S. TV primetime news, including information on their order, length, and topic, and, for all conflict-related
stories, also information on various qualitative attributes of the news reports based on detailed video content analysis; ii) all attacks carried out by both sides of the conflict, including
the date, the number of victims, and various characteristics of the attacks; iii) the volume of
Google searches about the conflict; and iv) the occurrence of natural and man-made disasters.
3.1. DATA ON US NEWS
3.1.1. N EWS PRESSURE
Daily data on TV prime-time news broadcasts on the top four U.S. networks–NBC, ABC,
CBS, and CNN–are available from the Vanderbilt Television News Archive for the period of
2000-2011.8 For each day and each network, the following information is available for every
story featured in the evening edition: the order of appearance, the length in seconds, and a
short summary of the topic. We use this information to construct a measure of the presence
of other important newsworthy events that are not related to any Israeli or Palestinian actions
8
Data on FOX News are available for a much shorter period of time.
10
and that may crowd out the coverage of the Israeli-Palestinian conflict. In contrast to CNN,
which features news around the clock, NBC, ABC and CBS have a well-defined 30-minute
time-slot allocated to prime-time evening news. As Eisensee and Stromberg (2007) point
out, the fact that this time-slot is limited to 30 minutes allows to measure the importance
of newsworthy events featured on the news: more important stories both take longer and
appear before less important stories.9 Following Eisensee and Stromberg (2007), we define
a variable, called “news pressure,” to be equal to the time devoted to the top three news
stories that are unrelated to Israel or Palestine in the prime-time evening newscast on NBC,
ABC and CBS, the three networks with 30-minute evening editions.
To construct the news pressure variable, we follow the following procedure. First, for
each day and each network, we identify the news stories that refer to Israel, or Palestine, or
both, i.e., stories whose summaries contain the words Israel, or Palestine, or any words with
related roots. When no story in a given newscast refers to Israel or Palestine, news pressure
is set to be equal to the time devoted to the top three stories. When one or more stories
refer to Israel or Palestine, the time devoted to all other stories is automatically reduced by
the time allocated to the Israel-or-Palestine-related stories. To adjust for this, on these days
we set the news pressure equal to the length of the top three non-Israel-or-Palestine-related
stories, divided by the time allocated to all non-Israel-or-Palestine-related stories, and multiplied by the total length of the newscast. This procedure allows to have comparable units
across days when Israel or Palestine are and are not featured in the news.10 The examples
presented in Table A.1 in the online appendix illustrate how daily news pressure is computed
for a specific network when Israel-or-Palestine-related content is present and absent from the
newscast. Finally, to get a single measure of news pressure for each given day, we compute
the median of the network-specific news pressures on that day. We use 10 minutes as a unit
of measurement for the daily news pressure variable. As reported in Table A.2 in the online
9
The main reason for this is the competition between networks for audience: if the evening news covered
less important news first, they would have lost audience to a competitor network that cover more important
news at the same time.
10 Even though the evening news are limited to 30 minutes, the actual length of time devoted to the news
varies somewhat from one day to the next and the weather forecast takes the rest of the time. Our results are
completely unaffected by whether we adjust the length of the top-thee news stories by the actual length of
the news that day minus the time devoted to Israel/Palestine story or by the median length of time devoted
to news on the particular network, again, minus the time devoted to Israel/Palestine. Our results are also
robust to not adjusting the units on the days when Israel or Palestine are featured in the news.
11
appendix, daily news pressure ranges from 2.3 minutes to 29.3 minutes. As shown in Figure
A.2 in the online appendix, the distribution of news pressure is substantially skewed to the
right. The distance between the median (8.3 minutes) and the 90th percentile (12.3 minutes)
is almost twice as large as between the median and the 10th percentile of the distribution (6.1
minutes).
3.1.2. N EWS COVERAGE OF THE CONFLICT
In addition to the news pressure variable, we compute various measures of daily news coverage of conflict events. We identify all stories devoted to Israeli-Palestinian conflict and
construct the following variables: i) a dummy for whether at least one conflict-related story
appears on any of the three networks newscasts on a given day, ii) the total number of
conflict-related news stories appearing on the three networks on a given day, and iii) the
total length (in seconds) of conflict-related news stories appearing on the three networks on
a given day. Summary statistics for all variables used in the analysis are presented in Table
A.2 in the online appendix.
3.1.3. C ONTENT ANALYSIS OF CONFLICT- RELATED VIDEOS
To test for the mechanism behind the relationship between news pressure and the timing
of attacks, we collected data on the actual content of all conflict-related news stories for
two networks, CNN and NBC, available from the Vanderbilt Television News Archive. We
first identified all the 582 news stories on the Israeli-Palestinian conflict featured during
our observation period. We then asked independent human analysts to code the content of
each of these videos by filling a questionnaire consisting of 23 questions. The aim of the
questionnaire is to evaluate conflict-related videos along several dimensions: i) whether the
video describes a particular attack, and if so, whether the video reports specific information
about the attack (i.e., location, weapon used, number of victims), ii) whether it includes
footage of the aftermath of the attack, personal information about the victims, images of
the victims’ burials and scenes of mourning, iii) whether the video features interviews with
the victims’ relatives or friends, and iv) whether it reports official reactions by Israeli or
Palestinian authorities. The exact list of questions and the summary statistics of the responses
12
are reported in Table A.3 in the online appendix. Out of 582 video extracts, 499 are related
to attacks that occurred on the same or on the previous day. We use these data to analyze
the differences in content of news stories appearing on the day of the attacks and on the
following day.
3.2. DATA ON ATTACKS
Data on the attacks by both Israelis and Palestinians come from two sources: the Israeli
Information Center for Human Rights, B’Tselem (http://www.btselem.org/) and the United
Nations Office for Coordination of Humanitarian Affairs in the Occupied Palestinian Territory, UNOCHA (http://www.ochaopt.org/).
The B’Tselem dataset covers the period between September 29, 2000 and November
24, 2011. It contains information on every attack by Israeli defense forces or Palestinian
militants that resulted in fatalities. For each attack the data include information on the day
of the attack, whether it was conducted by Palestinian militants or Israeli defense forces, the
number of fatal victims, as well as personal information about the victims. In the case of
attacks by Israeli forces, there is information on whether the attack was a special targetedkilling operation against a specific top-level Palestinian terrorist. Additional information
regarding the location of the attack, the type of weapon used in the attack, and whether the
victim participated in the hostilities is available only for a subset of observations.
The UNOCHA dataset covers the period between January 3, 2005 and November 24,
2011, almost entirely during the post-Intifada period. It only cover attacks by Israeli forces,
including those that resulted in non-fatal casualties, and includes comprehensive information
on the location of each attack and the type of the weapon used.
We aggregate the data from each of these two sources by date and attacker. Hence, we
construct daily time series for the following variables: i) the occurrence of attacks by each
side of the conflict, ii) the number of fatal and non-fatal casualties caused by each side’s
attacks, iii) the occurrence of attacks involving the use of heavy vs. non-heavy weapons, and
iv) the location of the attacks in areas with population density above and below the sample
median. The information on occurrence and severity of deadly attacks is available for the
entire period (between 2000 and 2011). Other variables are defined for the shorter period
13
(between 2005 and 2011).To differentiate attacks by the type of weapon used, we define the
following weapons as heavy weapons: all types of missiles (air plane missile, helicopter
missile, surface-to-surface rocket), sound bombs, explosives, tank shells, shock grenades,
and the following weapons as non-heavy: live ammunition, physical assaults, rubber-coated
metal bullets, tear gas, etc. Summary statistics for these variables are presented in the Table
A.2 in the online appendix.
3.3. DATA ON G OOGLE SEARCHES
To construct a daily measure of the U.S. public’s interest in the Israeli-Palestinian conflict,
we collect data on the daily volume of conflict-related searches on Google search engine.
Google Trends provides high-frequency data on the volume of Google searches for specific
queries from 2004 to present. We focus, in particular, on all searches that fall into the search
topic: “Israeli-Palestinian Conflict,” as defined by Google. When measuring the volume
of searches for any particular search topic, Google algorithms count many different search
queries (i.e., search terms or expressions) that relate to the same search topic. Google Trends
reports a measure of the daily volume of searches for each search topic normalized by the
highest search volume recorded over a three-month interval of interest. There are no data
available on the absolute number of searches. Hence, comparing search volumes for the
same topic in different periods requires rescaling of the data using a common scale defined
over the global time frame. We used a single query for the search volume at weekly frequency
for the entire 7-year period to bring the thirty-two separate daily queries to the same scale.
As a result, we were able to construct a daily measure of the search volume for the search
topic “Israeli-Palestinian Conflict” for the period between January 2004 and November 2011.
Summary statistics for this measure are also reported in the Table A.2 in the online appendix.
3.4. DATA ON DISASTERS
To test whether attacks are timed to predictable rather than unpredictable news, we also use
data on the occurrence of disasters. The comprehensive list of disasters, both natural and
man-made, for the period of interest are available from the International Disasters Dataset
(IDD) compiled by the Center for Research on the Epidemiology of Disasters (CRED) at
14
the Catholic University of Louvain. For each disaster event we use the following information: the starting date, the type of disaster, the country where the disaster occurred, and the
resulting number of fatalities. We focus on disasters that occurred in the United States and
resulted in a relatively high number of victims as these events are most likely to be covered
by U.S. news media. In particular, we compile a list of all U.S.-based disasters that fall into
the top 50% of the distribution of the number of fatalities among disasters of the same type.
106 such disasters occurred between 2000 to 2011.
4. A RE ATTACKS TIMED TO NEWS PRESSURE ?
4.1. N EWS ABOUT THE CONFLICT ON THE U.S. TV
Before testing our main hypothesis, we first test the premises upon which it rests. In particular, we verify whether U.S. TV news cover important events in the Israeli-Palestinian
conflict, whether the unrelated-to-Israel-or-Palestine news pressure affects the coverage of
the conflict, and whether conflict-related news on TV draw public attention to the conflict.
The results presented in the first three columns of Table 1 confirm that important
events in the Israeli-Palestinian conflict do get covered by U.S. prime-time TV news. We
estimate time-series regressions in which we relate the dummy for any news on the conflict,
the number of stories and the number of seconds devoted to the conflict daily to whether
an Israeli or Palestinian deadly attack occurred on the same day or on the previous day,
controlling for year, calendar month, and day of the week fixed effects. When the dependent
variable is the dummy, we estimate an OLS model (column 1), when the dependent variable
is the number or the length of conflict-related news, we estimate a maximum likelihood
negative binomial model (columns 2 and 3), as our count data exhibit over-dispersion. We
find that, on average, a fatal attack by Israelis and a fatal attack by Palestinians have a 9
and 8 percent chance to appear in the news, respectively. The occurrence of a fatal Israeli
attack increases the number of conflict-related news stories by 56% and the length of these
stories by 52% compared to days with no Israeli attacks when, on average, 0.225 stories
and 22.8 seconds are devoted to the conflict.11 The occurrence of a fatal Palestinian attack
11
When there are no Israeli attacks, the conflict-related news stories cover other issues, such as Palestinian
attacks, Israeli settlements or negotiation process.
15
is associated with a 44% increase in the number of conflict-related stories (from a baseline
of 0.27 stories), and a 80% increase in the length of these stories (from a baseline of 25.7
seconds).12
In columns 4 to 6, we test whether, conditional on the severity of the attacks, news
pressure has a negative and significant impact on the likelihood that conflict events are covered on the news. To do so, we focus on those days when an attack by either side occurred
on the same or the previous day and regress our measures of conflict-related coverage on
daily news pressure, controlling for the log(1+) of the number of victims, and the three
sets of fixed effects described above. Again, we estimate an OLS model when the outcome
is a dummy, and a maximum likelihood negative binomial model when the outcome is the
number or the length of conflict-related stories. We find that an increase in news pressure
by 4 minutes (equivalent to a shift in the distribution of news pressure from the median to
the 90th percentile or to a shift from the 75th to the 95th percentile) leads to a decrease in
the probability of any news on conflict being reported of 4.4 percentage points (roughly a
50% decrease relative to the baseline probability estimated in column 1). A 4-minute increase in news pressure is also associated with a decrease in the number and in the length of
conflict-related stories by 18% and 26%, respectively.13
In Table 2 we present evidence that the coverage of the conflict on U.S. TV news affects the interest of the U.S. public in the conflict. The daily volume of Google searches in the
U.S. for the search topic “Israeli-Palestinian Conflict” significantly increases with the news
coverage of the conflict, conditional on the severity of the conflict-related events. If at least
one network features a news story on the conflict, the volume of Google searches increases
12
Note that the Israeli attacks are more frequent and more deadly. The number of the Israeli attacks is 3.5
times larger than of the Palestinian attacks for our period of observation. On average, an Israeli attack
causes 4 fatal casualties and a Palestinian attack causes 1.6 casualties. The fact that the US news cover the
Palestinian attacks more in terms of length of coverage is consistent with several alternative explanations.
It could be related to the difference in the frequency of the attacks between the two sides, as the overall
time allocated to all Israeli attacks on prime-time news is substantially larger than that for all Palestinian
attacks: 28.7 vs. 14.9 minutes per month, on average, over our observation period. As we discuss below, it
could also be related to the fact that Israel makes a special effort for inform international journalists about
the Palestinian terrorist attacks and creates favorable conditions, both in terms of access and security, for
the journalists to film the aftermath of these attacks. Finally, it could also be related to a pro-Israeli bias of
the U.S. media.
13 If both Israeli and Palestinian sides strike today or yesterday, the predicted number of conflict-related stories
is 0.47 and their length is 55.9 seconds, holding other observables at their means.
16
by 8%. If all three networks feature one story about the conflict, the volume of Google
searches increases by 35% compared to when none of the networks reports on the conflict.
Five-minutes worth of conflict-related stories increases the volume of Google searchers by
27%.
4.2. T ESTING THE MAIN HYPOTHESIS
Our main hypothesis is that Israeli authorities chose the timing of their operations to coincide
with other newsworthy events that may crowd out news coverage of the attacks in order to
avoid the negative publicity associated with possible collateral damage, and in particular,
with civilian victims. We use news pressure from the U.S. as a proxy for the presence of
competing news-worthy events because: i) Israel cares about U.S. public opinion, ii) the
news pressure measure, the way it is constructed, is completely unrelated to any story about
Israel or Palestine and, therefore, is arguably exogenous from the perspective of both Israelis
and Palestinians. Importantly, as discussed in the introduction, news coverage of the civilian
casualties of Israeli attacks is damaging for Israel both internationally and domestically. This
allows us to formulate an unambiguous prediction. In contrast, the effect of news pressure
on the timing of Palestinian attacks is a priori ambiguous because, from the perspective of
Palestinian terrorists, news coverage of their attacks on international and domestic media
may have contrasting effects. International news coverage of Palestinian attacks, and of the
resulting civilian victims, is likely to shift international public opinion in favor of Israel and
against the Palestinian cause. News coverage of the attacks on domestic media can amplify
the sense of fear and devastation among Israelis (arguably the purpose of terrorists) and
fuel popular support for terrorist organizations among the Palestinians. As U.S. news cycle
depends, in part, on the presence of events that are considered newsworthy globally, domestic
and international news pressures may be correlated. As a consequence, the prediction about
the relationship between the Palestinian attacks and U.S. news pressure is ambiguous.
Our empirical strategy is as follows. We regress daily measures of the occurrence and
severity of the attacks by each side of the conflict on lags and leads of U.S. news pressure,
controlling for seasonality, overall conflict intensity, and the presence of retaliation motive
(studied by Jaeger and Paserman, 2008, 2009). In particular, we estimate the following
17
equation:
7
Ait =
∑
ατ NPt+τ + γ1 A j−1 + γ2 A j−7 + γ3 A j−14 + ηdt + ψmt + ϑyt + εit ,
(1)
τ=−7
where Ait is a measure of the occurrence or the intensity of an attack by side i (either Israelis
or Palestinians) against the opposing side j on day t, and NPt is the news pressure on day
t. A j−1 , A j−7 , and A j−14 are dummies for the occurrence of attacks by side j one day before
day t, between two and seven days prior to day t, and between eight and fourteen days prior
to day t, respectively. These dummies capture the need for retaliation following an attack
by the opposing side. ηdt , ψmt , and ϑyt denote fixed effects for each day of the week, each
calendar month, and each year, respectively. As both attacks and news pressure are serially
correlated, we estimate standard errors with Newey-West estimator or, alternatively, correct
them for clusters by month×year in the error term εit . We estimate all regressions on the
sample of all days in the period under consideration excluding: i) September 11, 2001, for
which news pressure is undefined because the evening news edition on that day far exceeded
30 minutes; and ii) the three weeks of the extraordinarily intense fighting during the operation
“Cast Lead”, i.e., the Gaza War (between 12/27/2008 and 01/18/ 2009).14
Table 3 presents the results for the Israeli attacks that caused at least one fatality. In
the first five columns we estimate a linear probability model with the dummy for occurrence
of fatal Israeli attacks on a given day as dependent variable. In the following five columns,
we focus on the severity of Israeli attacks. Columns 6 to 9 present results of the OLS estimation using the log(1+) of the number of fatalities of Israeli attacks on a given day as
dependent variable. In column 10, we use the number of fatalities as dependent variable
and estimate the maximum likelihood negative binomial regression, more appropriate for
over-dispersed outcome variables. In columns 1 and 6, we look at the contemporaneous relationship between Israeli attacks and news pressure conditional only on day of the week,
calendar month, and year fixed effects: the results indicate that both the timing and the intensity of Israeli attacks are positively correlated with news pressure on the same day. However,
14
In the online appendix we verify that our results do not depend on whether the days of the Gaza War
are included in the sample. We also verify that our results are completely robust to excluding the days
with extraordinarily high news pressure (i.e., top 0.5% of the distribution), which are mainly the days
immediately after 9/11 (these results are available upon request).
18
this relationship is not robust to accounting for autocorrelation in news pressure, which is
necessary as the pairwise correlation between news pressure and its lag is 0.56. To address
this issue, in columns 2, 3, 6 and 7 we include seven lags and seven leads of news pressure as
additional regressors (the number of lags is chosen so that there is no residual autocorrelation
in the main variables). After adding the full set of lags and leads, we find that it is the news
pressure on day t+1 that is significantly and robustly related to the occurrence and severity of
Israeli attacks on day t. Figure 2 illustrates this result graphically by plotting the coefficients
(along with their 95-percent confidence intervals) on the lags and leads of news pressure in
the regression with occurrence of an Israeli attack, presented in column 3. Even though only
the coefficient on the news pressure the day after an attack is statistically significant individually, all 15 coefficients on the 7 lags, the 7 leads, and the contemporaneous news pressure
are jointly statistically significant at 10% level. In columns 4, 8, and 9, we drop all the leads
of news pressure, except for the news pressure on the next day, keeping all 7 lags and the
contemporaneous news pressure in the set of covariates. We do this because the coefficients
on the 2nd to the 7th lead of the news pressure are jointly insignificant, and their inclusion
just adds noise to the estimation of interest. Henceforth, this is our baseline specification.
With regard to the magnitude of the effect: holding everything else constant, a 4minute increase in news pressure increases the probability of an Israeli attack on the previous
day by 3 percentage points (equal to 8 percent of the probability of an attack on an average
day), and increases the death tall of Israeli attacks by 20 percent from a base level of 0.8
fatality per day (according to the estimates in columns 4 and 9). These magnitudes are likely
to be smaller than the true unbiased estimates because they are subject to a severe attenuation
bias due to a measurement error in news pressure variable: Israel could possibly time its
attacks only to a predictable component of the news pressure; whereas the unpredictable
component is the noise added to the predictable news pressure which creates a classical
measurement error and biases the point estimates towards zero. We discuss this issue below.
Importantly, the relationship between the timing of the Israeli attacks and the nextday news pressure is not driven by the choice of the functional form or the list of covariates.
Figure A.3 in the online appendix presents a bivariate non-parametric relationship between
the occurrence of Israeli attacks or their severity and news pressure on the following day. The
19
two upper plots depict the relationship for the entire sample; whereas the two lower plots focus on the days with news pressure between its mean (8.8 minutes) and the 95th percentile
(14 minutes). The unconditional relationship is positive for the larger part of the distribution. To further corroborate this finding, we also plot the frequency of Israeli attacks and
the number of victims (both conditional on the occurrence of an attack and unconditional)
for each of the five quantiles of the distribution of next-day news pressure (see Figure A.4
in the online appendix). Overall, we find that attacks are more likely to occur and are more
deadly when news pressure is high. In Table A.4 in the online appendix, we also verify that
the relationship between Israeli attacks and next-day news pressure is robust to controlling
for the seven lags of the dependent variable and to excluding all lags of news pressure.15 An
auxiliary result of the analysis presented in Table 3 is that the retaliation motive is important
for Israeli attacks: the coefficients on the dummies for the occurrence of recent Palestinian
attacks are positive and statistically significant. A fatal Palestinian attack increases the probability of an immediate (next-day) military response by Israel by 4.5 percentage points and
the probability of a response within the following two weeks by 7 percentage points. We
discuss how retaliation motives interact with news pressure below.
We also investigate the relationship between news pressure and attacks by Palestinians. Table 4 and Figure 3 replicate the analysis presented in Table 3 and Figure 2 for Palestinian attacks resulting in fatalities. We find no evidence of a significant relationship between
the timing of the fatal Palestinian attacks and the U.S. news pressure. The coefficients on
the contemporaneous news pressure or its lags and leads are jointly statistically insignificant.
Some of the coefficients occasionally reach statistical significance, but these effects are not
robust to changes in the set of covariates or assumptions about the variance-covariance matrix, in contrast to the robustness of the effect for Israeli attacks. The only regressor that
significantly affects the severity of Palestinian attacks on a given day is the occurrence of an
Israeli attack on the previous day (see columns 7 and 8). This suggests that the retaliation
motive might be important for Palestinians as well; yet, this result is substantially weaker
than for Israel, as becomes insignificant in the ML estimation.
15
As mentioned above, the baseline results are based on the sample that excludes the three weeks of the Gaza
War from the sample. Online appendix table A.5 replicates the analysis of Table 3 on the full sample,
including the Gaza War. The results are robust and the magnitudes are larger.
20
4.3. T ESTING THE IMPLICATIONS OF STRATEGIC TIMING
In the previous subsection we have presented evidence that the timing of Israeli attacks is
significantly related to news pressure in the U.S. Our hypothesis is that this association is
a result of strategic behavior by Israeli authorities aimed at minimizing the impact of their
operations on international public opinion. This hypothesis has several implications that we
test in this sub-section.
4.3.1. T HE RISK OF CIVILIAN CASUALTIES
Negative publicity for Israeli operations derives mainly from the news coverage of the civilian victims of these operations. Hence, strategic timing should only apply to attacks that
bear a risk of civilian casualties. Comprehensive data on whether the victims of each Israeli
attack were civilians or militants are not available. However, the UNOCHA dataset, which
covers all Israeli attacks, including those that did not result in fatalities, contains detailed information on the exact location of the attack and the weapon used in it. Using these data, we
construct three alternative (imperfect) proxies for the ex ante probability that each particular
Israeli attack affected civilians: 1) whether this attack actually resulted in fatal casualties;
2) whether it involved the use of heavy weapons (such as missiles, rockets, sound bombs,
explosives, tank shells, artillery, or shock grenades) or non-heavy weapons (such as live ammunition, rubber-coated metal bullets, or tear gas); and 3) whether the attack was carried out
in areas of the Palestinian territories (governorates) with higher or lower population density.
Presumably, attacks that result in a higher death tall are executed with heavy weapons, and
attacks that target densely-populated areas are more likely to affect civilians.16
In Table 5, we re-estimate our baseline specification separately for different kinds
of attacks, both for their occurrence (OLS regressions in the upper panel of the table) and
for the number of casualties (ML negative binomial regressions in the lower panel). In
column 1, we focus on all Israeli attacks, including those that resulted only in injuries. The
statistical association of the next-day news pressure with the occurrence of all Israeli attacks
is not statistically significant, and that with the number of all victims (deaths + injuries)
16
As mentioned in the introduction, attacks with heavy weapons and in densely-populated areas are substantially more likely to result in fatalities: 68% and 49.6%, respectively, compared to 34.4% for an average
attack.
21
is significant at the 10% level. In both cases, the point estimates are positive but smaller in
magnitude and less precisely estimated compared to when we only focus on the occurrence of
deadly attacks and the number of fatal casualties (column 2).17 In column 4 and 5, we restrict
the sample to Palestinian governorates with population density above and below the median,
respectively, and find that news pressure significantly affects the timing and the severity
of the Israeli attacks in densely populated areas only. In column 6, we use the dummy for
whether Israel executed an attack using heavy weapons on a particular day and the number of
casualties from these attacks as dependent variables. We find that next-day news pressure is a
strong and significant predictor of the occurrence of attacks involving heavy weapons and of
the number of victims these attacks caused. Again, the magnitude of the coefficient on news
pressure is larger than when all attacks are pooled together, suggesting that more serious
attacks are more likely to be timed to other newsworthy events. Finally, as an illustration,
in columns 3 and 7, we restrict the sample to days with no deadly attacks and days with
no attacks with heavy weapons, respectively, and relate the remaining attacks, i.e., non-fatal
attacks and attacks involving light weapons, to news pressure. As expected, we find no
statistically significant relationship. The results of these regressions, should, however, be
interpreted with caution because the sample selection in these regressions is done on the
basis of the dependent variable (as the most severe attacks are dropped from the sample).
Overall, all the restrictions that we impose on the attacks or the sample so as to raise
the ex ante probability that civilians are affected increase the magnitude and the precision
of the coefficient on the next-day news pressure. This evidence corroborates our prediction
that Israel strategically chooses the timing of only those attacks that bear a risk of civilian
casualties, and hence, are more likely to generate negative publicity.
4.3.2. TARGETED KILLINGS
The B’Tselem data contain information on Palestinian victims deliberately killed by Israeli
security forces in the context of special targeted-killing operations as declared by Israeli authorities. Targeted-killings are usually characterized by higher urgency than other operations
17
Column 2 of Table 5 replicates on the UNOCHA data and sample the results, presented in Table 3 based on
the B’Tselem data, which, as mentioned, include only fatal attacks.
22
because Israeli military use every opportunity to eliminate important terrorist leaders and
avert imminent terrorist attacks. In this case, the potential PR costs, associated with possible
collateral damage, are outweighed by the security benefits. Furthermore, even when such
operations affect innocent civilians, targeted elimination of terrorists is arguably easier to
justify in the eyes of the international community.
Table 6 presents the results of a multinomial logit regression with three potential
mutually exclusive outcomes: i) a day with no Israeli attack; ii) a day with one or more
Israeli attacks not classified as a special targeted-killing; iii) a day with one or more Israeli
attacks of which at least one is classified as a targeted killing. We relate the probability of
each of these outcomes to the level of news pressure on the following day conditional on
the baseline set of covariates, described above. In line with our hypothesis, we find that the
timing of the targeted-killings operations is not significantly affected by the next-day news
pressure: the marginal effect on the probability of the targeted-killing outcome is positive,
but very small and statistically insignificant in contrast to the effect for all other attacks,
which, despite having military aims, were not acknowledged to target specific Palestinian
terrorists.
4.3.3. P REDICTABILITY OF THE NEWSWORTHY EVENTS
Attacks can be timed to important news if these can be predicted. Much of the news are
generally devoted to perfectly predictable and well-timed events; these include elections and
other important political events, big sport events, and holidays. However, other newsworthy
events are unpredictable; examples include natural disasters, terrorist attacks, and industrial
accidents. If the association between the timing of Israeli attacks and the U.S. news pressure
documented above is indeed due to the strategic and forward-looking behavior of the Israeli
authorities, it can only be driven by the predictable component of news pressure.
To test whether this is indeed the case, we analyzed the content of the top three
news stories on days when news pressure was extraordinarily high, i.e., between the 98th
and 99.5th percentile of the distribution. In particular, we identify the most frequent keywords appearing in the summary of these stories provided by the Vanderbilt News Archive.
Figure 4 presents the frequency distribution of these keywords separately for days with and
23
without a deadly Israeli attack occurring one day before, which intuitively can be thought as
the “compliers” and the “non-compliers” of the positive relationship between the next-day
news pressure and Israeli attacks. The six most frequent keywords among the compliers are:
Iraq and war (referring to the then-on-going war in Iraq), campaign, Gore, Bush, Florida,
recount (referring to important events in U.S. politics); these are predictable events. The six
most frequent keywords among the non-compliers are: hurricane, earthquake, coast, Katrina,
Japan, and Tsunami; all of these keywords refer to different natural disasters, with a completely unpredictable onset. We also grouped all keywords from news stories featured during
the extraordinarily-high news pressure days into the following six categories: 1) natural and
man-made disasters, 2) U.S. politics, 3) Iraqi war, 4) holiday shopping, 5) economic news,
and 6) other. Figure 5 presents the distribution of keywords among these categories for compliers and non-compliers. As expected, keywords in the category that is perfectly predictable,
i.e., holiday shopping, are only present among compliers; keywords in the category U.S. politics, which likely refer to predictable political events, are more frequent among compliers;
in contrast, keywords in the disasters category, which refer to events that are likely to be
unpredictable (especially at their onset), are more frequent among non-compliers. Although
these findings are based only on a subset of days with extraordinarily-high news pressure,
they provide suggestive evidence that Israeli attacks are timed only to predictable events.
To investigate the issue of the predictability of news pressure on the entire sample,
we use the data on disasters over the entire period of observation, available from CRED and
described above. The onset of a disaster, both natural and man-made, is arguably unpredictable and, as documented by Eisensee and Stromberg (2007), is likely to appear among
the top news on U.S. TV networks, especially if the disaster is severe and takes place in the
U.S. In column 1 of table 7 we confirm that news pressure is indeed higher when such disasters occur. To do so, we regress the daily news pressure on a dummy variable for whether
a serious disaster breaks out on a given day in the U.S. and the number of fatalities caused
by the disaster, capturing the severity and, therefore, the newsworthiness of the event.18 As
news pressure is significantly higher on the day of a disaster and more so for more deadly
disasters, we can decompose news pressure into a sum of two components: i) the news pres18
As discussed in the data section, we deem disaster to be serious if it falls into the 50th percentile or higher
of the distribution of the number of fatalities among disasters of the same type.
24
sure that is driven entirely by news about the disaster onset, which is equal to the predicted
value from the estimation presented in column 1 and ii) the news pressure that is driven by all
other events, which is the residual from the same estimation. Both components are significantly related to the overall news pressure, but differ in the extent of their predictability. The
disaster-driven component is unpredictable, whereas the component of the news pressure,
free from the effect of the disaster’s onset, is more predictable than the overall news pressure
as it is cleaned from one unpredictable element, i.e., disasters. We use these components
as alternative instruments for the overall news pressure to estimate the relationship between
the occurrence and severity of Israeli attacks and next-day news pressure, controlling for the
same-day news pressure.
Columns 2 and 5 of Table 7 replicate the OLS estimation for the occurrence and
for the log(+1) of number of victims of Israeli attacks, respectively, but using a more parsimonious specification, which omits the lags of news pressure. Columns 3 and 6 present
the second stage of the corresponding IV estimations using the disaster-driven component
of news pressure and its lead to instrument for the contemporaneous news pressure and its
lead. In columns 4 and 7, instead, the news pressure and its lead are instrumented using
the disaster-free component and its lead. The results confirm that there is no statistically
significant relationship between the unpredicted part of news pressure and Israeli attacks. In
contrast, when news pressure is instrumented with the disaster-free component, the effect of
next-day news pressure on both the occurrence and the severity of Israeli attacks is positive,
statistically significant, and slightly larger in magnitude than in the uninstrumented specifications (columns 2 vs. 4 and columns 5 vs. 7). The increase in the magnitude of the effect
in the IV regressions is consistent with the attenuation bias due to the measurement error in
news pressure, as the instrument eliminates one source of the measurement error, namely,
unpredictable disasters. This difference in the magnitude of the coefficients is not large,
however: for example, the difference in the effect of a-4-minute increase in news pressure in
the following day on the probability of Israeli attack between disaster-free news pressure and
25
overall news pressure is 3.2 vs. 3.0 percentage points.19 This is to be expected because: 1)
disasters are rare and disaster-free news pressure is very similar to the overall news pressure
(as evident from the magnitude of the F-statistics reported at the bottom of the table); and
2) a lot of unpredictability remains in the disaster-free news pressure, as many non-disasterrelated news are also unpredictable.20 Taken together, the evidence is supportive of the view
that Israeli attacks are timed to coincide with predictable newsworthy events and are not
related to news that cannot be predicted.
4.3.4. C ONFLICT INTENSITY AND RETALIATION
If the timing of Israeli attacks is subject to strategic considerations, one would expect the
same to apply to attacks executed in retaliation against Palestinian attacks, but only when
there is room for maneuver as far as the timing of the retaliation is concerned, namely,
when the conflict is not too intense and the need for retaliation is relatively less urgent. As
discussed above, our sample period covers two distinct phases of the conflict characterized
by very different intensity of fighting: the second Intifada and the post-Intifada period. Table
A.7 in the online appendix illustrates the difference in the frequency of attacks between the
two periods: whereas during the second Intifada the probability that an attack occurred on a
given day was 59.7% for Israeli attacks and 18.8% for Palestinian attacks, in the post-Intifada
period these probabilities were down to 25.4% and 6.1%, respectively. The extremely high
intensity of fighting during the second Intifada implied urgency for Israel to retaliate against
Palestinian attacks and, arguably, less room for strategic considerations in deciding when to
retaliate.
We investigate this hypothesis in table 8, where we present the estimation of our
usual specifications (OLS for occurrence and ML negative binomial for the number of fatalities), but including a dummy for the post-Intifada period as well as its interaction with
prior Palestinian attacks and a triple interaction term of the the post-Intifada dummy with
19
The magnitude of the OLS coefficient in Table 7 (column 2) is somewhat higher than in the baseline OLS
specification, presented in the column 4 of Table 3, because here we omit lags of news pressure from the list
of covariates and news pressure is serially correlated. We do this in order to avoid the need to instrument
additional seven regressors that are known to be irrelevant. Including the lags of news pressure as controls
does not affect any of the conclusions from this analysis.
20 The reduced form estimation is presented in Table A.6 in the online appendix.
26
prior Palestinian attacks and next-day news pressure. The results in columns 1 and 4 confirm
that Israeli attacks were more frequent and more deadly during the Intifada, as illustrated
by the negative and significant coefficients on the post-Intifada dummy. More importantly,
they validate our conjecture that retaliation was more urgent during the Second Intifada: the
coefficients on the interaction between the post-Intifada dummy and the occurrence of Palestinian attacks one day before are negative and significant. This pattern only holds for the
immediate (next-day) retaliation: we find no significant difference between the Second Intifada and the post-Intifada periods in Israeli response to Palestinian attacks that occurred one
or two weeks ago. Furthermore, we find no significant difference between the two periods
in the effect of news pressure on Israeli attacks: the coefficients on the interaction between
next-day news pressure and the post-Intifada dummy are small, insignificant, and change the
sign across specifications, as presented in columns 2 and 5.
Most importantly, consistent with our prediction, the timing of the retaliation-driven
Israeli attacks is differentially affected by news pressure during and after the Second Intifada.
This is illustrated in columns 3 and 6 of Table 8: the coefficients on the triple interaction
term between a Palestinian attack one day yesterday, next-day news pressure, and the postIntifada dummy are positive, large in magnitude and statistically significant. The magnitudes
of the coefficients imply that a 4-minute increase in news pressure led to an increase in the
probability of an Israeli attack in retaliation to a Palestinian fatal attack by a factor of 2 (from
0.22 to 0.44) during the post-Intifada period compared to a mere 6% increase during the
Second Intifada (i.e., from 0.65 to 0.69). The same-size increase in news pressure led to
an increase in the number of fatalities of Israeli retaliations by a factor of 2.5 post-Intifada
(from 0.45 to 1.12) and by only 15% (from 2.20 to 2.54) during the Second Intifada.21 These
findings corroborate the view that PR considerations are less important for retaliation in the
21
As always, in the baseline analysis we present results for the sample that excludes Gaza War. Table A.8
in the online appendix reports the results for the full sample, including Gaza War. This is the only table
in the paper, for which including Gaza War into the sample makes a difference. This is not surprising,
as Gaza War is the period of the most intense fighting over our observation period and the idea of the
comparison between Intifada and post-Intifada periods is to explore variation in the intensity of fighting.
Including Gaza War into the sample blurs this comparison. The neither the numbers of victims of an average
Israeli attack and of an attack in retaliation to a Palestinian attack nor the effect of news pressure on the
number of victims in retaliation attacks are significantly different between Intifada and post-Intifada, when
post-Intifada period includes Gaza War (as can be seen from columns 4-6 of Table A.8). The results for
occurrence of Israeli attacks remain statistically significant, but the relevant magnitudes are smaller than in
the full sample (columns 1-3 of Table A.8).
27
periods of intense fighting.
5. M ECHANISM :
COVERAGE OF CONFLICT ON THE SAME VS . NEXT DAY
In this section, we explore the mechanism behind the effect. In particular, we shed light on
why Israel times its attacks to the predicted news pressure on the following day rather than on
the same day. The most obvious possible explanation is that news are slow and it takes time
for reporters to prepare a story. If the news about important events in the Israeli-Palestinian
conflict appeared in the media only one day after their actual occurrence, it would not have
been surprising that Israel timed its attacks to the news pressure on the following day. We
test and reject this hypothesis in columns 1 to 4 of Table 9. In column 1, we examine how
the occurrence of an Israeli attack affects news coverage on the same day. To do this, we
regress the number of news stories about the Israeli-Palestinian conflict that appeared on a
given day on a dummy for whether an Israeli attack occurred on the same day, controlling
for contemporaneous news pressure and the day of the week, calendar month and year fixed
effects. To make sure that the news coverage is related to the Israeli attack that took place
on the same day, we restrict the sample to days such that no Israeli attack occurred on the
previous day and no Palestinian attack occurred on the same or the previous day. In column
2, similarly, we examine how the occurrence of an Israeli attack affects news on the following
day. In this case, we restrict the sample to days with no Israeli attack on the same day and
no Palestinian attack either on the same or the previous day. The results indicate that Israeli
attacks get news coverage both on the same and on the following day, but that, on average,
the number of conflict-related stories featured on the same day is 2.4 times as large as on the
next day.
Columns 3 and 4 report results of analogous regressions on the full sample, in which
the dummy for any news on the conflict or the number of conflict-related news is regressed
on the indicators of Israeli and Palestinian attacks separately on the same and on the next
day. Again, holding everything else constant, we find that an Israeli attack is 2.6 times more
likely to be covered on the day of the attack than on the following day (10.5% vs. 4% of the
attacks get news coverage) and that the number of stories devoted to Israeli attacks is 1.64
times larger on the same than on the following day. These results indicate that the reason for
28
why Israel times its attacks to news pressure on the next day is unrelated to the time when
the news stories appear. Therefore, the reason must be related to differences in the nature of
news coverage on the same and on the next day, in particular to the possibility that next-day
news stories are less favorable for the public image of Israel than same-day stories. In what
follows, we test this mechanism.
First, we compare the length of news stories that appear on the same and on the next
day conditional on the story being covered by restricting the sample to days with news on the
conflict. In this case, we estimate a maximum likelihood negative binomial model truncated
at zero, as the length of conflict-related stories is always positive but still over-dispersed. The
results are presented in column 5 of Table 9. The coefficient on the same-day attack is much
smaller in magnitude than the one for previous-day attack, and only the latter is statistically
significant (p-value of the text for equality of these coefficients: 0.1096). The magnitudes of
these coefficients imply that the next-day news stories on Israeli attacks are usually 4 times
longer than the same-day stories. This evidence suggests that the next-day coverage includes
a more in-depth account of the events.
Second, we examine the differences in the actual content between conflict-related
news reports on the same and on the next day. We use data from the coding of all 582
conflict-related videos that appeared on NBC and CNN over our sample period. The coding, performed by independent human analysts, consisted of answering a series of questions
about the content of the videos, with particular regard to: i) whether videos reported factual
information about the attack (i.e., number of victims, weapons used, exact location of the
events), ii) whether the videos contained footage of the victims, footage of the scenes of
burial and mourning, interviews with relatives and friends of the victims, and iii) whether
the videos reported information on the official reactions of Israeli or Palestinian authorities.
The complete list of questions and the mean values for each answer are reported in Table A.3
in the online appendix.
Out of a total of 582 newscasts devoted to the conflict, 326 focused on Israeli attacks
against Palestinians, of which 210 did not mention any Palestinian attack. Of the 326 newscasts on Israeli attacks, 278 were aired on the same day of the attack, 46 on the following
day, and only 2 on other days. 156 videos were fully devoted to Palestinian attacks against
29
Israel and 116 spoke about attacks on both sides. 98 conflict-related videos did not focus on
military incidents but on other related issues such as peace negotiations. Overall, 499 videos
were devoted to attacks on either side that occurred on the same or on the previous day.
In Table 10, we report the main results of the comparison between the content of
same-day and next-day videos. Table A.9 in the online appendix reports analogous results for
all the remaining questions. In Panel A, we focus on the most restrictive sample of videos,
i.e., those that were devoted to a particular Israeli attack, did not mention any Palestinian
attacks, and were aired on the same or the day following the attacks. There are 201 such
videos, 36 of which were aired on the day after the attack. We regress the variables measuring
different aspects of the video content on a dummy for whether the story appeared on the day
after the attack, conditional on network fixed effects. In panel B, we look at the entire
population of videos and regress each characteristic of the content on a dummy for next-day
coverage controlling for the following set of covariates: a dummy for whether videos were
aired neither on the same nor on the next day (leaving same-day videos as the comparison
group), a dummy for whether the video covered a Palestinian attack and the interaction of this
variable with the the dummy for the next-day coverage, a dummy for whether the story was
devoted to a particular attack (rather than, for example, a series of attacks), and network fixed
effects. In both panels, we adjust standard errors to clusters in error terms at month×year
level.
The results of both specifications draw a consistent picture. With regard to factual
informational content, the only difference between same-day and the next-day coverage of
Israeli attacks is that next-day news are 40 percentage points more likely to report information on the exact location of the attack (76% vs. 36% of the newscasts). Other dry facts about
the attack, such as the number of victims (total or civilian victims), or the weapon used, are
as likely to be reported on the same as on the next day. In contrast, next-day newscasts are
significantly more likely to report personal information about the civilian victims, such as
their names and family stories. This difference is substantial: 30.5% of the next-day stories include personal information about civilian victims against only 10% of the same-day
news. The footage of burials and of the scenes of mourning is also significantly more likely
to appear on the next-day newscast. In particular, such highly emotional content is included
30
in 29% of the next-day newscasts compared to 11% of the same-day newscasts. A similar
pattern holds for the footage of the interviews with family members, friends of the victims
and witnesses of the incident, which appear on 14% of the next-day newscasts compared to
only 8% of the same-day newscasts.22 Finally, next-day videos are significantly less likely
to report the official reaction of Israeli authorities about the incident (34% of the next-day
videos vs. 54% of the same-day videos).23
These results provide a clear rationale for why Israel should be more concerned about
the next-day compared to the same-day news coverage of its attacks on international media:
next-day coverage is more damaging for Israel’s image abroad because it is more emotionally charged. This is due to the fact that the next-day newscasts feature personal stories
about civilian victims, rather than simply reporting dry facts, and rely more heavily on visuals rather than just a narrative. As is well-known in cognitive and social psychology (e.g.,
Borgida and Nisbett, 1977; Martin and Powers, 1982; Wilkins, 1983), personal stories are
more powerful at conveying information than dry numbers, as they help listeners, readers,
and viewers to relate to the story. In addition, information transmission is more effective in
the form of visuals than in the form of words (e.g., Mandl and Levin, eds, 1989; Houghton
and Willows, eds, 1987a,b; Houts et al., 2006), as stories appear more real when accompanied by images that evoke strong emotions, e.g., funerals, mourning, relatives’ suffering.
Finally, because of a stronger focus on the personal stories of civilian victims, next-day reports are also less likely to present the official position of Israeli authorities on the events,
which obviously thwarts Israeli public diplomacy efforts. Overall, these findings strongly
support the hypothesis that Israeli authorities time their most severe attacks to international
news pressure so as to avoid the type of news coverage of civilian victims that is most emotionally charged and, therefore, most detrimental for Israel.
From the estimation based on the full sample (Panel B), in addition, we conclude that:
1) same-day stories about the Palestinian attacks are significantly more likely to contain basic
22
We combine 2 questions about the interviews with witnesses and interviews with friends and relatives in one
variable to maximize the variation, as only combined these questions result in sufficient number of videos,
which contain interviews.
23 We report the magnitudes based on the most restrictive sample of videos that are devoted to a particular
Israeli attack, which takes place either on the same day as the newscast or on the previous day and does not
mention any Palestinian attack.
31
information about the attack, personal information about victims, interviews with witnesses,
footage of victims and the reaction of Israeli authorities than same-day stories about Israeli
attacks, as can be seen from the estimated coefficient on the dummy for the story about
Palestinian attack; 2) there is little difference between the content of same-day and nextday videos devoted to Palestinian attacks, as can be seen from the estimated coefficients on
the interaction between the next-day coverage and the dummy for story about Palestinian
attacks (with the exception of the information on location); and 3) stories that appear not on
the same or the next day are much less likely to present any information or visuals regarding
the attacks on either side, which suggests that these stories do not focus on attacks.
An important question is why there are such marked differences in the content between same-day and next-day coverage of Israeli attacks, whereas the same pattern does not
apply to the coverage of Palestinian attacks. The differences in news coverage of Israeli attacks between same and next day are most likely driven by a combination of technological
constraints on news-reporting of an armed conflict, general to most conflict areas, and local
traditions, specific to the Middle East. As for technology of news production is concerned,
in the immediate aftermath of an Israeli attack there are usually no international reporters in
the vicinity, as Israeli military do not share their intentions with the journalists. Furthermore,
even when reporters are close to the site of the attack and can quickly reach it, being on site
is often dangerous both for reporters and (potential) witnesses due to the risk of follow-up
strikes.24 Thus, on the day of the attack, it is hard for reporters to gather footage and collect
any relevant personal details about the victims. Conditions are more favorable to journalists’
work on the day after the attack. Local traditions prescribe burial of victims to take place one
day after death.25 The burial ceremony takes place in open air and is attended by the local
population in large numbers. This is an easy and safe opportunity for reporters to collect
personal information about the victims.26 The funerals also provides a good opportunity for
reporters to produce emotionally-charged visuals of mourning and corpses. As some Israeli
commentators - including PM Nethanyau - have pointed out, Palestinians do use the occa24
See, for instance, the CNN interview with Tyler Hicks, the New York Times photojournalist, about the attack
on July 16, 2016, made the same day, available on the the Vanderbilt Television News Archive.
25 As is the case for most ethnic groups with ancestry originating from places with hot climate.
26 The fact that there are many people on the street is the best insurance against a possible Israeli attack; and
relatives and friends of the victims are present.
32
sion of burials of civilian victims to portray Israelis as violent aggressors and themselves as
innocent victims.27 The reason why the news coverage of Palestinian attacks on Israel is
more similar between the same and the next day may also be due to Israel’s strategic behavior. Israelis make an effort to create favorable conditions allowing international reporters to
access necessary information about Palestinian attacks right after these attacks occur. Israelis
grant the international reporters prompt access to the attack area, allow them to film the site,
interview witnesses, and produce footage of the damage and the victims. Evidently, this policy is motivated by the Israeli understanding of the potential effect of the news coverage on
international public opinion.
6. A COUNTERFACTUAL EXERCISE
The predicted number of conflict-related stories featured on the three U.S. TV networks following an average Israeli attack is 0.8 and the predicted length is 2 minutes. A simple exercise of looking at the bivariate relationship between conflict-related news and news pressure,
conditional on Israeli attack taking place one day before, yields that when news pressure is
at its 99th percentile, the predicted number of conflict news stories is reduced to 0.57 and the
predicted length to 50 seconds. In contrast, when news pressure is at its 1st percentile, an
Israeli attack gets, on average, almost one news story lasting a total of 193 seconds. We do
not have enough statistical power to estimate to what extent news pressure affects the number
of Israeli attacks in addition to their timing. Theoretically, the effect of news pressure on the
number of attacks could be anywhere between two alternative extreme scenarios. The first
scenario is that news pressure only affects the timing of the attacks and not their number, i.e.,
each attack planned for a particular day and not realized because of the expectation of low
news pressure on the following day is just postponed to another day when the next-day news
pressure is expected to be high. The second scenario is that there is a constant probability of
an attack every day; and whether an attack occurs on a given day depends on the costs and
benefits of the attack on that day. Thus, if the PR cost of an attack on a given day is too high
because of low expected next-day news pressure, the attacks is canceled and the cost-benefit
27
Note that our sample period ends before the twitter revolution and therefore we cannot analyze how social
media, and twitter, in particular, affected the strategic timing effects that we uncover.
33
calculation for the following day starts anew. Under this scenario, the decision on whether
to attack on a given day is a Markov process, and if the attack is not carried out on a given
day, it gets canceled forever.
Table A.10 in the online appendix presents the calculation of the predicted number
of attacks and their respective news coverage under these different scenarios. In the first
scenario (the “time-displacement”), the number of predicted attacks over our sample period
is 1,560 and the number of predicted Palestinian victims is 3,384 regardless of the level of
news pressure. In the second scenario (the “attack-cancellation”), the number of attacks
depends on the level of news pressure: at the 99th percentile of news pressure, the predicted
number of attacks increases by 361 attacks (to a total of 1,921 attacks), and the number of
victims by 1840 (to a total of 5,224); at the 1st percentile of news pressure, instead, the
predicted number of attacks and fatalities decrease respectively by 185 and 676 (to a total of
1,375 attacks and 2,708 victims). If only the timing of the attacks, but not their number, were
responding to news pressure, the U.S. public would have been exposed to 3.9 times more
conflict-related content with news pressure at its 1st percentile than at its 99th percentile. If
news pressure also affected the number of attacks, however, the corresponding ratio would
be 2.8, as low news pressure would imply not only a larger probability of conflict coverage
given the attack, but also fewer attacks.28
7. C ONCLUSIONS
We present systematic evidence of strategic behavior of policy makers, who time their unpopular actions to coincide with other newsworthy events that distract the public’s attention
so as to minimize the resulting negative publicity by focusing on the relationship between the
timing of attacks in the context of the recurrent Israeli-Palestinian conflict and the presence
of important events on U.S. TV news.
Israeli authorities choose the timing of their attacks to minimize next-day news coverage of their attacks, which is more likely to feature personal stories of civilian victims and
emotionally charged footage of burials and scenes of mourning. This strategy is aimed at
28
See Table A.10 in the online appendix for more details on these calculations and additional predictions
made for news pressure at the 25th and 75th percentiles.
34
reducing the negative impact of Israeli attacks on the perception of Israel’s image by international public opinion. It is developed with full understanding of both the technological
aspects of news reporting in conflict zones and the cognitive psychology of the effect of the
media coverage of conflict on international public opinion.
The strategic timing of Israeli attacks is applied only to predictable newsworthy
events and only to Israeli military actions that are likely to generate negative publicity (i.e.,
when the risk of having civilians affected is particularly high). We also show that strategic
timing is applied to retaliation only when retaliation is not urgent (i.e., when the conflict is
not very intense).
There is no effect of U.S. news on Palestinian terrorist attacks, which could be explained by the lack of coordination among different Palestinian factions involved in violence
against Israel, lower responsiveness of Palestinians to U.S. public opinion, a possibly lower
level of sophistication of some Palestinian terrorists relative to IDF, or the potentially contrasting impact of news coverage of Palestinian attacks on international and domestic news.
Our results suggest that strategic behavior on the part of policy-makers may undermine the effectiveness of mass media as a watchdog and, thus, reduce citizens’ ability to
keep public officials accountable.
35
R EFERENCES
Adena, Maja, Ruben Enikolopov, Maria Petrova, Veronica Santarosa, and Ekaterina Zhuravskaya, “Radio and the Rise of the Nazis in Prewar Germany,” Quarterly Journal of Economics,
2016, Forthcoming.
Besley, Timothy and Andrea Prat, “Handcuffs for the Grabbing Hand? Media Capture and Political
Accountability,” American Economic Review, 2006, 96 (3), 720–736.
Blattman, Christopher and Edward Miguel, “Civil War,” Journal of Economic Literature, March
2010, 48 (1), 3–57.
Borgida, Eugene and Richard E. Nisbett, “The Differential Impact of Abstract vs. Concrete Information on Decisions,” Journal of Applied Psychology, 1977, 7 (3), 258–271.
Damodaran, Aswath, “The Weekend Effect in Information Releases: A Study of Earnings and Dividend Announcements,” Review of Financial Studies, 1989, 2 (4), 607–623.
deHaan, Ed, Terry Shevlin, and Jacob Thornock, “Market (In)Attention and Earnings Announcement Timing,” 2014. mimeo.
DellaVigna, Stefano and Joshua M. Pollet, “Investor Inattention and Friday Earnings Announcements,” Journal of Finance, 2009, 64 (2), 709–749.
, Ruben Enikolopov, Vera Mironova, Maria Petrova, and Ekaterina Zhuravskaya, “CrossBorder Media and Nationalism: Evidence from Serbian Radio in Croatia,” American Economic
Journal: Applied Economics, July 2014, 6 (3), 103–32.
Doyle, Jeffrey T. and Matthew Magilke, “The Timing of Earnings Announcements: An Examination of the Strategic Disclosure Hypothesis,” Accounting Review, 2009, 84 (1), 157–182.
Eisensee, Thomas and David Stromberg, “News Droughts, News Floods, and U.S. Disaster Relief,”
The Quarterly Journal of Economics, 05 2007, 122 (2), 693–728.
Houghton, Harvey A. and Dale M. Willows, eds, The psychology of illustration: Volume 1. Basic
research. Springer-Verlag New York 1987.
and , eds, The psychology of illustration: Volume 2. Instructional issues. Springer-Verlag New
York 1987.
Houts, Peter S., Cecilia C. Doak, Leonard G. Doak, and Matthew J. Loscalzo, “The role of
pictures in improving health communication: A review of research on attention, comprehension,
recall, and adherence,” Patient Education and Counseling, 2006, (61), 173–190.
Jackson, Matthew O. and Massimo Morelli, “The Reasons for Wars Ð an Updated Survey,” in
Christopher J. Coyne and Rachel L. Mathers, eds., The Handbook on the Political Economy of
War, Cheltenham, UK: Elgar Publishing, 2009.
Jaeger, David A. and Daniele M. Paserman, “The Shape of Things to Come? Assessing the Effectiveness of Suicide Attacks and Targeted Killings,” American Economic Review, 2009, 4, 315–342.
and M. Daniele Paserman, “Israel, the Palestinian Factions, and the Cycle of Violence,” American Economic Review, May 2006, 96 (2), 45–49.
and , “The Cycle of Violence? An Empirical Analysis of Fatalities in the Palestinian-Israeli
Conflict,” American Economic Review, September 2008, 98 (4), 1591–1604.
, Esteban F. Klor, Sami H. Miaari, and M. Daniele Paserman, “Can Militants Use Violence to
Win Public Support? Evidence from the Second Intifada,” NBER Working Papers 16475, National
Bureau of Economic Research, Inc October 2010.
Mandl, Heinz and Joel R. Levin, eds, Knowledge acquisition from text and pictures. North-Holland
Amsterdam 1989.
Martin, Joanne and Melanie E. Powers, “Organizational stories: More vivid and persuasive than
quantitative data,” in Barry Staw, ed., Psychological Foundations of Organizational Behavior,
Glenview, IL: Scott Foresman, 1982.
36
Patell, James M. and Mark A. Wolfson, “Good News, Bad News, and the Intraday Timing of
Corporate Disclosures,” Accounting Review, 1982, 57 (3), 509–527.
Prat, Andrea and David Stromberg, “The political economy of mass media,” in Daron Acemoglu,
Manuel Arellano, and Eddie Dekel, eds., Advances in Economics and Econometrics, Cambridge:
Cambridge University Press, 2013.
Seethaler, Josef, Matthias Karmasin, Gabriele Melischek, and Romy Wohlert, eds, Selling War:
The Role of the Mass Media in Hostile Conflicts from World War I to the “War to Terror.” Intellect
Ltd., University of Chicago Press Chicago 2013.
Snyder, James M. and David Stromberg, “Press Coverage and Political Accountability,” Journal
of Political Economy, 04 2010, 118 (2), 355–408.
Sobbrio, Francesco, “The political economy of news media: theory, evidence and open issues,”
in Francesco Forte, Ram Mudambi, and Pietro Maria Navarra, eds., A Handbook of Alternative
Theories of Public Economics, Chelthenam: Edward Elgar Press, 2014.
Wilkins, Alan L., “Organizational Stories as Symbols Which Control the Organization,” in Samuel B.
Bacharach, ed., Monographs in Organization Behavior and Industrial Relations, Greenwich: JAI
Press, Inc., 1983.
Yanagizawa-Drott, David Hans, “Propaganda and Conflict: Evidence from the Rwandan Genocide,”
Quarterly Journal of Economics, 2014, 129 (4), 1947–1994.
37
F IGURE 1: N UMBER OF I SRAELI AND PALESTINIAN FATALITIES BY MONTH
WITH AND WITHOUT G AZA WAR , 2000-2011
800
600
400
200
0
Victims of Israeli and Palestinian attacks
1000
E NTIRE SAMPLE PERIOD
2000m7
2001m12
2003m5
2004m10
2006m3
Victims of Israeli attacks
2007m8
2009m1
2010m6
2011m11
Victims of Palestinian attacks
200
150
100
50
0
Victims of Israeli and Palestinian attacks
250
E XCLUDING G AZA WAR (D ECEMBER 27, 2008 - JANUARY 18, 2009)
2000m7
2001m12
2003m5
2004m10
2006m3
Victims of Israeli attacks
2007m8
2009m1
2010m6
2011m11
Victims of Palestinian attacks
The figure reports the monthly number of fatalities caused by Israeli and Palestinian attacks.
The shaded area indicates the period of the Gaza War. The vertical line marks the end of
the Second Intifada. The horizontal lines indicate the average monthly number of fatalities
separately for the Second Intifada and the post-Intifada periods.
38
F IGURE 2: I SRAELI ATTACKS AND U.S. N EWS P RESSURE (2000-2011)
The figure reports the estimated coefficients (and respective 95% confidence intervals for NeweyWest standard errors) from the regression of occurrence of Israeli attacks on news pressure between
three days before and five days after the event (table 3, column 3). The regression includes year,
calendar month, and day of the week fixed effects, seven lags and seven leads of news pressure,
and controls for the occurrence of Palestinian attacks in the 1, 7 and 14 days before.
39
F IGURE 3: PALESTINIAN ATTACKS AND U.S. N EWS P RESSURE (2000-2011)
The figure reports the estimated coefficients (and respective 95% confidence intervals for NeweyWest standard errors) from the regression of occurrence of Palestinian attacks on news pressure
between three days before and five days after the event (table 4, column 3). The regression includes year, calendar month, and day of the week fixed effects, seven lags and seven leads of news
pressure, and controls for the occurrence of Israeli attacks in the 1, 7 and 14 days before.
40
F IGURE 4: M OST FREQUENT WORDS AMONG KEYWORDS FOR DAYS
WITH EXTREMELY HIGH NEWS PRESSURE
DAYS WITH A DEADLY I SRAELI ATTACK ON THE PREVIOUS DAY
DAYS WITH NO DEADLY I SRAELI ATTACK ON THE PREVIOUS DAY
41
F IGURE 5: C LASSIFICATION OF FREQUENT WORDS IN HEADLINES FOR DAYS
WITH EXTREMELY HIGH NEWS PRESSURE
42
TABLE 1: C OVERAGE OF I SRAELI -PALESTINIAN C ONFLICT
AND U.S. N EWS P RESSURE (2000-2011)
(1)
(2)
Model:
Any news
on conflict
OLS
Number of
conflict news
ML Neg. Bin.
Sample:
All days
All days
All days
Israeli attack
(t or t-1)
0.086***
(0.018)
0.444***
(0.105)
0.425***
(0.154)
Palestinian attack
(t or t-1)
0.079***
(0.027)
0.367***
(0.091)
0.617***
(0.167)
Dependent variable:
(3)
(4)
Length of
Any news
conflict news on conflict
ML Neg. Bin.
OLS
(5)
(6)
Number of
Length of
conflict news conflict news
ML Neg. Bin. ML Neg. Bin.
Attack at
t or t-1
Attack at
t or t-1
Attack at
t or t-1
News Pressure (t)
-0.111**
(0.047)
-0.487***
(0.167)
-0.754**
(0.330)
Ln victims
(Isr. attacks t or t-1)
0.081***
(0.011)
0.266***
(0.036)
0.315***
(0.056)
Ln victims
(Pal. attacks t or t-1)
0.084***
(0.013)
0.300***
(0.052)
0.437***
(0.086)
Year FEs
Yes
Yes
Yes
Yes
Yes
Yes
Calendar month FEs
Yes
Yes
Yes
Yes
Yes
Yes
Day of the week FEs
Yes
Yes
Yes
Yes
Yes
Yes
Observations
(Pseudo) R-squared
4,051
0.210
4,003
0.1301
4,003
0.0198
2,443
0.222
2,409
0.125
2,409
0.019
The table explores the relationship between occurrence of conflict events, conflict-related news coverage, and news pressure
on U.S. media. In columns 1 to 3 we regress various measures of news coverage of conflict-related stories on U.S. media
on dummies for the occurrence of Israeli and Palestinian attacks in the same or previous day; the dependent variables are
respectively: a dummy for any conflict-related story appearing on the news (col. 1), the average number of conflict-related
stories appeared on the news (col. 2), the average length (in seconds) of conflict-related stories appeared on the news (col.
3). In columns 4 to 6 we restrict the sample to those observations for which an Israeli or Palestinian attack occurred on the
same or on the previous day, and regress each of the three measure of conflict-related news coverage on same-day news
pressure, controlling for the intensity of the attacks measured by the log of the number of victims (+1). OLS regressions
presented in columns 1 and 4; maximum likelihood negative binomial regressions in all other columns. All regressions
include year, calendar month, and day of the week fixed effects. R-squared reported in columns 1 and 4, pseudo R-squared
in all others. Standard errors clustered by month-year reported in parentheses. *** p<0.01, ** p<0.05, * p<0.1.
43
TABLE 2: C ONFLICT- RELATED N EWS AND G OOGLE S EARCHES (2004-2011)
Dependent variable:
Model:
Log Daily Volume of Google Searches for “Israeli-Palestinian Conflict”
OLS
OLS
OLS
OLS
Ln victims Isr. attacks (t, t − 1)
0.037**
(0.016)
0.033**
(0.016)
0.027*
(0.016)
0.029*
(0.015)
Ln victims Pal. attacks (t, t − 1)
0.023
(0.017)
0.015
(0.017)
0.010
(0.017)
0.014
(0.016)
0.080**
(0.039)
Any news on conflict (t, t − 1)
0.115***
(0.035)
Number of conflict news (t, t − 1)
0.0008***
(0.0002)
Length of conflict news (t, t − 1)
Year FEs
Calendar month FEs
Day of the week FEs
Linear time trend
Observations
R-squared
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
2,392
0.504
2,392
0.508
2,333
0.515
2,333
0.520
The table reports the results of a series of regressions of the daily volume of Google searches for the topic “Israeli-Palestinian
Conflict” on the intensity and the news coverage of Israeli and Palestinian attacks conducted on the same and on the previous
day. To control for the intensity of attacks, all columns include the log of the victims of Israeli and Palestinian attacks on
the same or on the previous day. We measure news coverage of the conflict as a dummy for whether any conflict-related
story appeared on the news on the same or on the previous day (col. 2), ii) the average number of conflict-related stories
appeared on the news on the same or on the previous day (col. 3), iii) the average length (in seconds) of conflict-related
stories appeared on the news on the same or on the previous day (col. 4). Data on the volume of Google searches are from
Google Trends. Since these data are only available from 2004, we restrict our attention to the post-Intifada period. All
regressions include day of the week, calendar month, year fixed effects as well as a linear time trend. Corrected Newey-West
standard errors are reported in parentheses. *** p<0.01, ** p<0.05, * p<0.1.
44
45
0.029
(0.036)
0.081**
(0.035)
-0.012
(0.035)
0.029
(0.035)
-0.026
(0.035)
-0.022
(0.033)
-0.033
(0.035)
0.073**
(0.032)
4,006
0.184
(2)
OLS
(1)
OLS
4,048
0.181
dummy
dummy
3,999
0.190
0.027
(0.036)
0.076**
(0.035)
-0.011
(0.035)
0.034
(0.034)
-0.027
(0.035)
-0.022
(0.033)
-0.035
(0.035)
0.047*
(0.025)
0.048**
(0.020)
0.072***
(0.021)
(3)
OLS
dummy
4,017
0.189
-0.021
(0.035)
-0.023
(0.033)
-0.034
(0.034)
0.045*
(0.025)
0.047**
(0.020)
0.072***
(0.021)
0.025
(0.036)
0.077**
(0.035)
(4)
OLS
dummy
Occurrence of Israeli Attacks
4,048
0.175
0.128**
(0.052)
(5)
OLS
log(1+)
4,006
0.176
0.024
(0.047)
0.113**
(0.047)
-0.036
(0.047)
0.069
(0.050)
-0.037
(0.046)
0.026
(0.048)
-0.023
(0.046)
(6)
OLS
log(1+)
(8)
OLS
log(1+)
3,999
0.187
4,017
0.184
0.021
0.027
(0.047)
(0.047)
0.102**
0.119**
(0.048)
(0.049)
-0.035
(0.047)
0.077
(0.050)
-0.037
-0.027
(0.046)
(0.046)
0.025
0.030
(0.047)
(0.047)
-0.023
-0.018
(0.046)
(0.045)
0.129*** 0.127***
(0.045)
(0.045)
0.101*** 0.102***
(0.032)
(0.032)
0.097*** 0.096***
(0.033)
(0.033)
(7)
OLS
log(1+)
Fatalities of Israeli Attacks
4,017
0.067
-0.159
(0.159)
0.262
(0.161)
-0.140
(0.159)
0.258***
(0.088)
0.266***
(0.087)
0.210**
(0.091)
0.016
(0.145)
0.463***
(0.159)
(9)
ML Neg. Bin.
number
The table examines the relationship between timing and intensity of Israeli attacks and news pressure on U.S. media. The dependent variables are: occurrence
of Israeli attacks (columns 1-4), log of the number of fatalities of Israeli attacks (columns 5-8), number of fatalities of Israeli attacks (column 9). OLS
regressions presented in columns 1-8; maximum likelihood negative binomial regression in column 9. Regressions in columns 2-3 and 6-7 include seven lags
and seven leads of news pressure; regressions in columns 4, 8 and 9 only include seven lags of news pressure. All regressions include year, calendar month,
and day of the week fixed effects. Pseudo R-squared reported in column 9, R-squared in all other columns. Standard errors clustered by month-year reported
in parentheses in columns 1 and 5; corrected Newey-West standard errors in all other columns. *** p<0.01, ** p<0.05, * p<0.1.
Observations
(Pseudo) R-squared
Palestinian attacks
(previous day)
Palestinian attacks
(previous week)
Palestinian attacks
(week before previous)
News Pressure t-3
News Pressure t-2
News Pressure t-1
News Pressure t+3
News Pressure t+2
News Pressure t+1
News Pressure t
Model:
Dependent variable:
TABLE 3: I SRAELI ATTACKS AND U.S. N EWS P RESSURE (2000-2011)
46
4,048
0.083
(3)
OLS
dummy
4,006
0.086
3,999
0.087
0.005
0.002
(0.024) (0.024)
0.016
0.016
(0.026) (0.026)
0.040
0.041
(0.026) (0.026)
-0.036
-0.036
(0.023) (0.023)
-0.042* -0.042*
(0.025) (0.025)
-0.007
-0.006
(0.024) (0.024)
0.005
0.004
(0.022) (0.022)
0.015
(0.011)
0.009
(0.011)
0.017
(0.011)
(2)
OLS
(1)
OLS
0.004
(0.018)
dummy
4,017
0.086
-0.040
(0.025)
-0.005
(0.023)
0.002
(0.022)
0.017
(0.011)
0.009
(0.011)
0.017
(0.011)
0.002
(0.024)
0.028
(0.024)
(4)
OLS
dummy
Occurrence of Palestinian Attacks
dummy
4,048
0.076
0.003
(0.023)
(5)
OLS
4,006
0.079
0.002
(0.029)
0.028
(0.030)
0.040
(0.029)
-0.045
(0.028)
-0.040
(0.031)
-0.008
(0.026)
-0.005
(0.025)
(6)
OLS
3,999
0.080
-0.002
(0.029)
0.028
(0.030)
0.040
(0.029)
-0.045
(0.028)
-0.041
(0.031)
-0.007
(0.027)
-0.005
(0.025)
0.027**
(0.013)
-0.003
(0.019)
0.013
(0.013)
(7)
OLS
log(1+)
4,017
0.079
-0.040
(0.031)
-0.007
(0.026)
-0.010
(0.025)
0.029**
(0.013)
-0.003
(0.019)
0.012
(0.013)
-0.005
(0.029)
0.033
(0.026)
(8)
OLS
log(1+)
4,017
0.075
-0.458
(0.347)
-0.081
(0.365)
0.427
(0.371)
0.130
(0.134)
0.037
(0.278)
-0.121
(0.355)
-0.032
(0.314)
0.197
(0.325)
(9)
ML Neg. Bin.
number
Fatalities of Palestinian Attacks
log(1+) log(1+)
The table examines the relationship between timing and intensity of Palestinian attacks and news pressure on U.S. media. The dependent variables are:
occurrence of Palestinian attacks (columns 1-4), log of the number of fatalities of Palestinian attacks (columns 5-8), number of fatalities of Palestinian
attacks (column 9). OLS regressions presented in columns 1-8; maximum likelihood negative binomial regression in column 9. Regressions in columns 2-3
and 6-7 include seven lags and seven leads of news pressure; regressions in columns 4, 8 and 9 only include seven lags of news pressure. All regressions
include year, calendar month, and day of the week fixed effects. Pseudo R-squared reported in column 9, R-squared in all other columns. Standard errors
clustered by month-year reported in parentheses in columns 1 and 5; corrected Newey-West standard errors in all other columns. *** p<0.01, ** p<0.05,
* p<0.1.
Observations
R-squared
Israeli attacks
(previous day)
Israeli attacks
(previous week)
Israeli attacks
(week before previous)
News Pressure t-3
News Pressure t-2
News Pressure t-1
News Pressure t+3
News Pressure t+2
News Pressure t+1
News Pressure t
Model:
Dependent variable:
TABLE 4: PALESTINIAN ATTACKS AND U.S. N EWS P RESSURE (2000-2011)
47
All days
All days
(2)
0.073*
(0.043)
0.040
(0.044)
News pressure (next day)
0.029
(0.056)
2,447
36.0%
Observations
Prob. of fatalities
for a type of attack
100.0%
2,447
Yes
Yes
Yes
0.0%
1,800
Yes
Yes
Yes
-0.033
(0.184)
Injuries
49.6%
2,447
Yes
Yes
Yes
Casualties in
DP areas
0.670***
(0.247)
34.0%
2,447
Yes
Yes
Yes
Casualties in
NDP areas
0.018
(0.161)
-0.026
(0.049)
Attacks in
NDP areas
All days
NDP areas
(5)
69.5%
2,447
Yes
Yes
Yes
Casualties w.
heavy weapons
0.686*
(0.366
0.076**
(0.037)
Attacks w.
heavy weapons
All days
(6)
35.1%
1,971
Yes
Yes
Yes
Casualties w.
light weapons
-0.003
(0.179)
0.038
(0.051)
Attacks w.
light weapons
Days w/o attack
w. heavy weapons
(7)
THE LIKELIHOOD OF CIVILIAN CASUALTIES
The table examines the relationship between timing and severity of Israeli attacks and news pressure on U.S. media separately for attacks that are more
and less likely to produce civilian victims. The sample includes all days (columns 1, 2, 4, 5, and 6), only days on which no fatal Israeli attack was
carried out (column 3), and only days on which no Israeli attack involving the use of heavy weapons was carried out (column 7). All regressions
control for same-day news pressure, seven lags of news pressure, prior Palestinian attacks, as well as day of the week, calendar month, and year fixed
effects. OLS regressions presented in Panel A, maximum likelihood negative binomial regressions in Panel B. In Panel A the dependent variables are
dummies for whether: i) any Israeli attack was carried out on a given day (column 1), ii) any Israeli attack resulting in fatalities was carried out on
a given day (column 2), iii) any Israeli attack not resulting in fatalities was carried out on a given day (column 3), iv) any Israeli attack was carried
out in high-population-density areas on a given day (column 4), v) any Israeli attack was carried out in low-population-density areas on a given day
(column 5), vi) any Israeli attack involving the use of heavy weapons was carried out on a given day (column 6), vii) any Israeli attack not involving
the use of heavy weapons was carried out on a given day (column 7). In Panel B the dependent variables are the number of: i) casualties caused by
Israeli attacks carried out on a given day (column 1), ii) fatalities caused by Israeli attacks carried out in a given day (column 2), iii) injuries caused by
Israeli attacks carried out on a given day (column 3), iv) casualties caused by Israeli attacks carried out in high-population-density areas on a given day
(column 4), v) casualties caused by Israeli attacks carried out in low-population-density areas on a given day (column 5), vi) casualties caused by Israeli
attacks involving the use of heavy weapons carried out on a given day (column 6), vii) casualties caused by Israeli attacks not involving the use of heavy
weapons carried out on a given day (column 7). Newey-West adjusted standard errors reported in parentheses.*** p<0.01, ** p<0.05, * p<0.1.
Yes
Yes
Yes
All
Fatalities
casualties
0.317*
0.731***
(0.162)
(0.250)
News pressure (same day)
News pressure (lags)
Prior Palestinian attacks
News pressure (next day)
Dep. variable:
0.091*
(0.050)
Attacks in
DP areas
All days
DP areas
Days w/o fatal
attack
Non-fatal
attacks
(4)
(3)
Panel B: Number of victims (ML Negative Binomial regressions)
Fatal
attacks
All
attacks
Dep. variable:
Panel A: Occurrence of an attack (OLS regressions)
Sample:
(1)
TABLE 5: I SRAELI ATTACKS AND U.S. NEWS PRESSURE (2005-2011):
TABLE 6: I SRAELI ATTACKS AND NEWS PRESSURE :
TARGETED KILLINGS VS . NON - TARGETED ATTACKS
MULTINOMIAL LOGIT REGRESSION ( MARGINAL EFFECTS REPORTED )
Outcome:
News Pressure (next day)
(1)
No attacks
(2)
Non-targeted Attacks
(3)
Targeted Killings
-0.078**
(0.038)
0.066*
(0.038)
0.012
(0.014)
News pressure (same day)
News pressure (lags)
Prior Palestinian attacks
Yes
Yes
Yes
Observations
Pseudo R-squared
4,017
0.141
The table examines the relationship between news pressure on U.S. media and the timing
of Israeli attacks, separately for attacks aimed at eliminating specific top-priority targets
(targeted-killing special operations) and for all others (non-targeted). The table reports the
marginal effects of a multinomial logit regression in which the dependent variable is a categorical variable for the type of Israeli attacks coded as follows: 1 for days with no attacks,
2 for days with non-targeted attacks only, and 3 for days with targeted killings. All regressions control for same-day news pressure, seven lags of news pressure, previous Palestinian
attacks, as well as day of the week, calendar month, and year fixed effects. Robust standard
errors clustered by month-year reported in parentheses. *** p<0.01, ** p<0.05, * p<0.1.
48
49
0.316***
(0.066)
Num. victims
(1000’s)
4,031
0.107
16.64
—
—
Observations
R-squared
F disaster victims
F inst. NP (t)
F inst. NP(t+1)
—
—
—
4,031
0.188
Yes
Yes
0.029
(0.034)
0.081**
(0.034)
OLS
—
(2)
Occurrence
(4)
Occurrence
—
9.7
9.2
4,031
0.081
Yes
Yes
0.691
(0.816)
-0.585
(0.707)
—
547,387
289,646
4,031
0.188
Yes
Yes
0.028
(0.034)
0.083**
(0.034)
2SLS
2SLS
Disaster-driven NP Disaster-free NP
(3)
Occurrence
—
—
—
4,031
0.185
Yes
Yes
0.059
(0.049)
0.133***
(0.047)
OLS
—
(5)
Log Fatalities (+1)
—
9.7
9.2
4,031
0.127
Yes
Yes
0.730
(1.123)
-0.585
(0.963)
2SLS
Disaster-driven NP
(6)
Log Fatalities (+1)
—
547,387
289,646
4,031
0.185
Yes
Yes
0.057
(0.049)
0.136***
(0.048)
2SLS
Disaster-free NP
(7)
Log Fatalities (+1)
VS . UNPREDICTABLE ( DISASTER - DRIVEN ) NEWS
The table examines the relationship between the timing and severity of Israeli attacks, news pressure on U.S. media, and the occurrence of disasters. In column
1 we regress news pressure on a dummy variable for whether a disaster started on that day, and the number of fatalities caused by the disaster. In column 2 we
report our baseline result on the relationship between occurrence of Israeli attacks and news pressure. In column 3 and 4 we regress occurrence of Israeli attacks on
news pressure instrumented respectively with the predicted value and the residual obtained from the regression in column 1. The predicted value should capture the,
arguably unpredictable, variation in news pressure due to the occurrence of disasters; the residual should instead capture the variation in news pressure not driven
by those events. In columns 5 to 7 we estimate analogous OLS regressions using the log of the number of victims of Israeli attacks (+1) as dependent variable. All
regressions include day of the week, calendar month, and year fixed effects; regressions in columns 2 to 7 also control for attacks by Palestinians on previous days.
“F disaster victims” indicates the F-statistic for the test for joint significance of the dummy for the onset of the disaster and the number of victims;“F inst. NP (t)” the
F-statistic from the first stage for the excluded instrument predicting same-day news pressure. “F inst. NP(t+1)” the F-statistic from the first stage for the excluded
instrument predicting next-day news pressure. Robust standard errors clustered by month-year reported in parentheses in column 1; corrected Newey-West standard
errors in all other columns. *** p<0.01, ** p<0.05, * p<0.1.
Yes
Yes
FEs
Prior Pal. attacks
News pressure
(same day)
0.064**
(0.025)
OLS
—
Day disaster
started
News pressure
(next day)
Model:
Instrument:
Dep. variable:
(1)
News Pressure
TABLE 7: I SRAELI ATTACKS AND PREDICTABLE ( DISASTER - FREE )
TABLE 8: I SRAELI RETALIATION AND U.S. NEWS PRESSURE
DURING THE S ECOND I NTIFADA (2000-2004) AND AFTER IT (2005-2011)
Dependent variable:
Model:
News Pressure (t+1)
Occurrence of Israeli Attacks
(1)
(2)
(3)
OLS
OLS
OLS
0.075**
(0.035)
0.089
(0.056)
0.088
(0.056)
-0.022
(0.068)
-0.044
(0.068)
News Pressure (t+1)
× Post-Intifada
Palestinian attack (t-1)
× News Pressure (t+1)
× Post-Intifada
Fatalities of Israeli Attacks
(4)
(5)
(6)
ML Neg. Bin. ML Neg. Bin. ML Neg. Bin.
0.472***
(0.160)
0.363*
(0.194)
0.357*
(0.194)
0.222
(0.327)
0.130
(0.333)
0.509***
(0.164)
1.788*
(0.934)
Post-Intifada
-0.291*** -0.281***
(0.088)
(0.105)
-0.259**
(0.105)
-0.937***
(0.331)
-1.240***
(0.435)
-1.150***
(0.436)
Palestinian attack (t-1)
0.093***
(0.030)
0.093***
(0.030)
0.093***
(0.030)
0.388***
(0.096)
0.385***
(0.096)
0.386***
(0.096)
Palestinian attack (t-1)
× Post-Intifada
-0.130**
(0.052)
-0.133**
(0.052)
-0.564***
(0.149)
-0.454**
(0.230)
-0.456**
(0.232)
-2.032**
(0.878)
Palestinian attack
(previous week)
0.064**
(0.029)
0.045**
(0.020)
0.047**
(0.020)
0.262***
(0.100)
0.252***
(0.087)
0.257***
(0.087)
Palestinian attack
(previous week)
× Post-Intifada
-0.036
(0.041)
Palestinian attack
(week before previous)
0.060*
(0.033)
0.186**
(0.091)
0.188**
(0.091)
Palestinian attack
(week before previous)
×Post-Intifada
0.010
(0.043)
News Pressure (same day)
News Pressure (lags)
FEs
Observations
(Pseudo) R-squared
-0.017
(0.176)
0.065***
(0.021)
0.066***
(0.021)
0.265**
(0.115)
-0.148
(0.180)
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
4,017
0.194
4,017
0.193
4,017
0.195
4,017
0.068
4,017
0.068
4,017
0.069
The table examines the relationship between retaliation motive in Israeli attacks and news pressure on U.S. media
during and after the second Intifada. In the first three columns we use occurrence of Israeli attacks as dependent
variable. In column 1 we augment our baseline specification by including a dummy for the post-Intifada period and
its interactions with the three measures of previous Palestinian attacks. In column 2 we also include the interaction
between next-day news pressure and the post-Intifada dummy. In column 3 we include triple interactions between
next-day news pressure, the post-Intifada dummy, and a dummy for Palestinian attacks occurring the day before.
In columns 3 to 6 we estimate analogous maximum likelihoood negative binomial regressions using as dependent
variable the number of victims of Israeli attacks. All regressions control for same-day news pressure, seven lags of
news pressure, and for day of the week, calendar month, and year fixed effects. R-squared reported in columns 1 to
3, Pseudo R-squared in all others. Corrected Newey-West standard errors reported in parentheses; *** p<0.01, **
p<0.05, * p<0.1.
50
51
Days with no Palestinian attack
or other Israeli attack at t-1 or t
Sample:
2,117
0.134
Observations
(Pseudo) R-squared
0.138
2,124
0.221
4,047
0.137
3,999
Yes
Yes
0.026
1,106
Yes
Yes
0.222**
(0.096)
0.237***
(0.079)
0.209**
(0.084)
0.056
(0.074)
Days with
news on conflict
ML Neg. Bin.
zero-truncated
Length of
conflict news
(5)
The table examines the relationship between occurrence of Israeli and Palestinian attacks and conflict-related coverage on U.S. news on the
same and the following day. In column 1, we focus on days for which no Palestinian attacks occurred on the same or the previous day and
no Israeli attack on the previous day and regress a dummy for any conflict-related story appearing on the news on a dummy for at least one
Israeli attack occurring the same day. In column 2, we regress the same variable on a dummy for at least one Israeli attack occurring the
day before, focusing on days with no Palestinian attacks on the same or previous day and no Israeli attack on the same day. In column 3
and 4, using the whole sample, we regress occurrence and the number of conflict-related coverage on dummies for Israeli attacks occurring
on the same day, Israeli attacks occurring on the previous day, Palestinian attacks occurring on the same day, Palestinian attacks occurring
on the previous day. In columns 5 we estimate we focus on the sample of days in which conflict-related stories appeared on the news and
use length of coverage as dependent variable. OLS regression reported in column 3; maximum likelihood negative binomial regressions in
all other columns (zero-truncated in column 5). R-squared reported in column 3, pseudo R-squared in all other columns. Standard errors
clustered by month-year reported in parentheses. *** p<0.01, ** p<0.05, * p<0.1.
Yes
Yes
News Pressure (same day)
FEs
Yes
Yes
0.207**
(0.081)
0.057**
(0.025)
Palestinian attack at t-1
Yes
Yes
0.452***
(0.080)
0.104***
(0.026)
Israeli attack at t-1
Palestinian attack at t
0.412***
(0.068)
All days
ML Neg. Bin.
Number of
conflict news
(4)
0.257***
(0.073)
0.201*
(0.113)
0.105***
(0.017)
All days
OLS
Any news
on conflict
(3)
0.040**
(0.017)
0.414***
(0.108)
ML Neg. Bin.
Model:
Israeli attack at t
Number of
conflict news
Number of
conflict news
Dependent variable:
ML. Neg. Bin.
(2)
(1)
TABLE 9: S AME - DAY VS . N EXT- DAY N EWS C OVERAGE OF C ONFLICT E VENTS
52
Info on
exact
location
Info on
weapons
Info on
# of civilian
victims
(3)
Personal
info on
victims
(4)
Video of
burials or
mourning
(5)
0.0935
(0.0904)
201
0.007
0.71
0.403***
(0.0759)
201
0.141
0.36
0.0593
0.326***
(0.0839)
(0.0810)
0.0143
-0.388***
(0.113)
(0.116)
-0.631*** -0.295***
(0.0553)
(0.0651)
0.0270
0.0831*
(0.0466)
(0.0492)
0.147*** 0.252***
(0.0458)
(0.0444)
582
582
0.321
0.235
-0.0862
(0.0992)
0.165
(0.126)
-0.466***
(0.0529)
-0.0545
(0.0443)
0.103**
(0.0438)
582
0.189
-0.0502
(0.119)
201
0.033
0.55
0.191***
(0.0604)
-0.0310
(0.0990)
-0.0808**
(0.0348)
0.00930
(0.0360)
0.0909**
(0.0345)
582
0.064
0.205***
(0.0673)
201
0.071
0.10
0.128*
(0.0686)
0.00291
(0.0958)
-0.0306
(0.0407)
0.0518
(0.0366)
0.0405
(0.0318)
582
0.044
0.180**
(0.0884)
201
0.050
0.11
0.0504
(0.0360)
0.0180
(0.0984)
-0.0839
(0.0565)
-0.0610
(0.0480)
0.0762**
(0.0354)
582
0.037
0.0636*
(0.0353)
201
0.009
0.8
-0.138*
(0.0771)
0.0587
(0.120)
-0.0810
(0.0778)
0.193***
(0.0601)
0.150**
(0.0688)
582
0.087
-0.197***
(0.0729)
201
0.106
0.54
The table examines qualitative differences between same-day and next-day coverage of Israeli attacks on U.S. news. Panel A considers all news reports
about an Israeli attack occurred on the same or previous day that do not mention contemporary or previous Palestinian attacks. Panel B considers all news
reports about Israelo-Palestinian conflict. The dependent variables are dummies indicating whether the news report includes: i) information on the weapon
used in the attack (col. 1), ii) information on the exact location of the attack (col. 2), iii) information on the number of civilian victims of the attack (col. 3),
iv) personal information of the victims of the attack(s) (col. 4), v) footage of the victims’ burials or scenes of mourning by the victims’ relatives or friends
(col. 5), vi) interviews with the victims’ relatives or friends or with witnesses (col. 6), vii) information about the reactions of Israeli authorities (col. 7).
All regressions include network and analyst fixed effects. Standard errors clustered by month-year are reported in parentheses. *** p<0.01, ** p<0.05, *
p<0.1. The mean responses to the questions for sample of stories about a particular Israeli attack that took place on the same day or the day before and that
do not mention Palestinian attacks are presented in the last row of Panel A. The mean responses to the questions for the entire sample of videos is presented
in Table A.3 in the online appendix.
Observations
R-squared
Story about Palestinian attack
Story about a particular attack
Next-day coverage ×
Palestinian attack
Story appears not on same
day and not the day after
Next-day coverage
Panel B: all news stories about Israelo-Palestinian conflict
Observations
R-squared
Mean same-day coverage
Next-day coverage
(7)
Interview of
Reactions
friends/ relatives of Israeli
or witnesses
authorities
(6)
Panel A: news stories about an Israeli attack occurred on the same or previous day not mentioning Palestinian attacks
Content:
(2)
(1)
TABLE 10: D IFFERENCE IN CONTENT BETWEEN STORIES THAT APPEAR ON THE SAME AND ON THE NEXT DAY
A. O NLINE A PPENDIX
TABLE A.1: C ONSTRUCTION OF THE N EWS P RESSURE VARIABLE
Panel A. Israel or Palestine are not on the news:
Date
Network
15Jan2004
15Jan2004
15Jan2004
15Jan2004
15Jan2004
15Jan2004
15Jan2004
15Jan2004
15Jan2004
15Jan2004
15Jan2004
CBS
CBS
CBS
CBS
CBS
CBS
CBS
CBS
CBS
CBS
CBS
Length Time to top 3 non-­‐conflict News pressure News pressure (secs)
news stories (secs)
(secs)
(10 mins)
Weather Watch (Extreme Weather)
290
440
440
0.73
Iraq / New Government
30
440
440
0.73
Atlanta, Georgia / Bush Protests
120
440
440
0.73
Campaign 04 / Iowa
160
440
440
0.73
Market Watch: Consumer Prices, Inflation, Stocks
20
440
440
0.73
Inside Story (Internet Child Pornography)
120
440
440
0.73
Space: Mars Exploration
20
440
440
0.73
Flu Season
20
440
440
0.73
Eye on America (Mad Cow Disease)
200
440
440
0.73
Iraq / Homecoming
140
440
440
0.73
Good Night
10
440
440
0.73
Headline
Panel B. Israel or Palestine are covered in top three stories:
Length Time to top 3 non-­‐conflict News pressure News pressure (secs)
news stories (secs)
(secs)
(10 mins)
Date
Network
Headline
11Jun2003
11Jun2003
11Jun2003
11Jun2003
11Jun2003
11Jun2003
11Jun2003
11Jun2003
11Jun2003
CBS
CBS
CBS
CBS
CBS
CBS
CBS
CBS
CBS
Middle East / Israelis vs. Palestinians / Violence
Iraq: After Saddam / Weapons of Mass Destruction
Economy / Tax Cut Plan
Medicine: Monkeypox
Shreveport, Louisiana / Hudspeth Shooting
International News
California / Coma Birth
Eye on America (Bon Appetit!)
Good Night
200
120
150
160
130
70
110
80
10
430
430
430
430
430
430
430
430
430
533.6
533.6
533.6
533.6
533.6
533.6
533.6
533.6
533.6
0.89
0.89
0.89
0.89
0.89
0.89
0.89
0.89
0.89
Panel C. Israel or Palestine are covered, but not in top three stories:
Date
Network
Headline
17-­‐Apr-­‐01
17-­‐Apr-­‐01
17-­‐Apr-­‐01
17-­‐Apr-­‐01
17-­‐Apr-­‐01
17-­‐Apr-­‐01
17-­‐Apr-­‐01
17-­‐Apr-­‐01
17-­‐Apr-­‐01
17-­‐Apr-­‐01
17-­‐Apr-­‐01
17-­‐Apr-­‐01
17-­‐Apr-­‐01
17-­‐Apr-­‐01
17-­‐Apr-­‐01
CBS
CBS
CBS
CBS
CBS
CBS
CBS
CBS
CBS
CBS
CBS
CBS
CBS
CBS
CBS
Economy / Signs of Recovery
Economy / Intel Profit / Cisco Sales
Bush / Environmental Policy / Lead Laws
Middle East / Israelis vs. Palestinians / Violence
US-­‐China Relations / Negotiations
Vietnam / Mia Mission
Foot-­‐and-­‐Mouth Disease / Prevention
Weather Watch (Upper Midwest Floods)
Africa / Slave Ship / Child Slavery
Concorde Test Flight
Health Watch (St. John's Wort and Depression)
Health / Dietary Supplements / Poor
Health / Alcohol and Heart Disease Study
Ellis Island / Immigrant Data Base
Good Night
53
Length Time to top 3 non-­‐conflict News pressure News pressure (secs)
news stories (secs)
(secs)
(10 mins)
160
430
441.7
0.74
140
430
441.7
0.74
130
430
441.7
0.74
30
430
441.7
0.74
100
430
441.7
0.74
20
430
441.7
0.74
140
430
441.7
0.74
30
430
441.7
0.74
130
430
441.7
0.74
30
430
441.7
0.74
110
430
441.7
0.74
20
430
441.7
0.74
30
430
441.7
0.74
50
430
441.7
0.74
10
430
441.7
0.74
54
News Pressure (10 mins)
Presence of conflict-related news stories
Number of conflict-related news stories
Length of conflict related stories (secs.)
Log daily volume of Google searches for “Israeli-Palestinian conflict”
Occurrence of fatal Israeli attacks
Occurrence of fatal Palestinian attacks
Number of fatalities caused by Israeli attacks
Number of fatalities caused by Palestinian attacks
Log number of fatalities caused by Israeli attack
Log number of fatalities caused by Palestinian attacks
Occurrence of Israeli targeted attacks
Occurrence of Israeli non-targeted attacks
Occurrence of any Israeli attacks
Occurrence of fatal Israeli attacks
Occurrence of non-fatal Israeli attacks
Occurrence of Israeli attacks with heavy weapons
Occurrence of Israeli attacks with light weapons
Occurrence of Israeli attacks in high-population-density areas
Occurrence of Israeli attacks in low-population-density areas
Fatal victims of Israeli attacks
Non-fatal victims of Israeli attacks
Victims of Israeli attacks with heavy weapons
Victims of Israeli attacks with light weapons
Victims of Israeli attacks in high-population-density areas
Victims of Israeli attacks in low-population-density areas
Variable
Vanderbilt News Archive
Vanderbilt News Archive
Vanderbilt News Archive
Vanderbilt News Archive
Google Trends
B’Tselem
B’Tselem
B’Tselem
B’Tselem
B’Tselem
B’Tselem
B’Tselem
B’Tselem
UNOCHA
UNOCHA
UNOCHA
UNOCHA
UNOCHA
UNOCHA
UNOCHA
UNOCHA
UNOCHA
UNOCHA
UNOCHA
UNOCHA
UNOCHA
Source
TABLE A.2: D ESCRIPTIVE S TATISTICS
4,071
4,026
4,026
4,026
2,750
4,074
4,074
4,074
4,074
4,074
4,074
4,074
4,074
2,517
2,517
2,517
2,517
2,517
2,517
2,517
2,517
2,517
2,517
2,517
2,517
2,517
Observations
0.886
0.182
0.544
78.359
2.196
0.392
0.112
1.571
0.316
0.493
0.122
0.036
0.357
0.755
0.272
0.483
0.191
0.697
0.453
0.573
0.755
0.483
0.191
0.191
3.551
2.868
Mean
0.258
0.386
1.117
257.612
0.616
0.488
0.316
7.423
1.799
0.738
0.395
0.185
0.479
0.430
0.445
0.500
0.394
0.460
0.498
0.495
0.430
0.500
0.394
0.394
13.096
5.567
Std. Dev.
Max
0.233 2.933
0
1
0
9
0
3370
-0.082 4.763
0
1
0
1
0
356
0
59
0
5.878
0
4.094
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
324
0
91
Min
55
First name, gender and age of the person watching the newscast
Date and time of the newscast
Network of the newscast
Does the newscast focus on a particular military attack? (Yes/No)
Does the newscast focus on an Israeli attack against Palestinians? (Yes/No)
Does the newscast focus on a Palestinian attack against Israelis? (Yes/No)
Did this attack occurr the same day of the newscast? (Yes/No)
Did this attack occurr on the day before the newscast? (Yes/No)
Is the newscast based on an on-site report? (Yes/No)
Is the news correspondent interviewed by the host of the news program? (Yes/No)
Does the newscast report information about the weapon or weapons used in this attack? (Yes/No)
Does the newscast report information on the exact location of this attack? (Yes/No)
Does the newscast report information on the number of victims (if any) caused by the attack? (Yes/No)
Does the newscast report information on the number of civilian victims (if any) caused by the attack? (Yes/No)
Does the newscast show images of the actual site of the attack? (Yes/No)
Does the newscast show footage of the immediate aftermath of the incident? (Yes/No)
Do photos of the victims of this attack appear in the newscast? (Yes/No)
Does the newscast show footage of the victims of this attack? (Yes/No)
Does the newscast report personal information of the civilian victims (e.g., first or last name, age, family situation, etc.)? (Yes/No)
Does the newscast include footage of burials and/or scenes of mourning by family members? (Yes/No)
Does the newscast include interviews with witnesses of the accident? (Yes/No)
Does the newscast include interviews with friends and/or relatives of the civilian victims? (Yes/No)
Does the newscast report information about the reaction of Israeli authorities to the incident? (Yes/No)
Does the newscast report information about the reaction of Palestinian authorities to the incident? (Yes/No)
Overall, how emotional is the newscast on a hypothetical scale from 1 to 4: 1 being not emotional at all, 2 just a little bit emotional,
3 emotional, 4 very emotional
Overall, how would you assess the tone of the newscast on a hypothetical scale from -3 to 3: -3 being very pro-Palestine,
-2 pro-Palestine, -1 somewhat pro-Palestine, 0 neutral, 1 somewhat pro-Israel, 2 pro-Israel, and 3 very pro-Israel
Question
-
60
56
47
74
13
26
10
68
43
69
52
46
31
3
48
15
12
10
3
53
34
% Yes
Note: Total number of responses to each question is 582. The distribution of answers to question # 25 is as follows: “-3” 1; “-2” 18; “-1” 20; “0” 509; “1” 14; “2” 19; “3”
1. The distribution of answers to question # 26 is as follows: “0” 17; “1” 407; “2” 100; “3” 58.
26
25
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
Num.
TABLE A.3: Q UESTIONNAIRE FOR THE CONTENT ANALYSIS OF CONFLICT- RELATED VIDEOS
TABLE A.4: I SRAELI ATTACKS AND U.S. NEWS PRESSURE
( ADDITIONAL SPECIFICATIONS )
(1)
Dependent variable:
Model:
(2)
(3)
(4)
Occurrence of Israeli Attacks
Fatalities of Israeli Attacks
OLS
OLS
ML Neg. Bin. ML Neg. Bin.
News pressure (t+1)
0.080**
(0.034)
0.081**
(0.034)
0.370**
(0.147)
0.473***
(0.155)
Palestinian attack (t-1)
0.028
(0.024)
0.049*
(0.025)
0.080
(0.086)
0.252***
(0.088)
Palestinian attack
(previous week)
0.020
(0.018)
0.047**
(0.020)
0.102
(0.071)
0.267***
(0.088)
0.048***
(0.018)
0.072***
(0.021)
0.164**
(0.073)
0.214**
(0.092)
Yes
Yes
Yes
Yes
No
No
Yes
Yes
Yes
Yes
Yes
Yes
No
No
Yes
Yes
4,017
0.216
4,031
0.188
4,017
0.086
4,031
0.066
Palestinian attack
(week before previous)
Lagged news pressure (7 lags)
Lagged dep. variable (7 lags)
News pressure (same day)
FEs
Observations
(Pseudo) R-squared
The table reports the results of a set of regressions analogous to those presented in columns 4 and 9 of Table 3 (our
baseline) but with alternative specifications. In columns 1 and 3 we augment the baseline specification including also
seven lags of the dependent variable. In columns 2 and 4 we show instead the results without lags of news pressure.
All regressions include day of the week, calendar month, and year fixed effects. R-squared reported in columns 1 and
2, pseudo R-squared in 3 and 4. Newey-West robust standard errors reported in parentheses; *** p<0.01, ** p<0.05,
* p<0.1.
56
57
-0.028
(0.035)
-0.024
(0.033)
-0.037
(0.034)
0.058**
(0.026)
-0.028
(0.034)
-0.023
(0.033)
-0.035
(0.034)
News pressure t-1
News pressure t-2
News pressure t-3
4,040
0.182
0.081***
(0.022)
0.059***
(0.021)
-0.036
(0.034)
0.057**
(0.026)
-0.024
(0.033)
-0.022
(0.035)
0.080**
(0.035)
0.026
(0.035)
OLS
(4)
dummy
4,048
0.139
0.112**
(0.054)
OLS
(5)
log(1+)
4,071
0.140
-0.038
(0.047)
0.013
(0.049)
-0.037
(0.047)
0.117**
(0.048)
0.016
(0.049)
OLS
(6)
log(1+)
0.019
(0.048)
-0.024
(0.047)
0.137***
(0.052)
0.024
(0.050)
OLS
(8)
log(1+)
4,022
0.161
4,040
0.158
0.140*** 0.139***
(0.045)
(0.045)
0.152*** 0.153***
(0.045)
(0.045)
-0.037
-0.031
(0.047)
(0.046)
0.172*** 0.171***
(0.055)
(0.055)
0.011
(0.049)
-0.038
(0.048)
0.104**
(0.049)
0.013
(0.049)
OLS
(7)
log(1+)
Fatalities of Israeli Attacks
WITH FULL SAMPLE
4,040
0.056
0.372***
(0.114)
0.455***
(0.103)
-0.210
(0.164)
0.341***
(0.101)
-0.045
(0.189)
0.141
(0.192)
0.572***
(0.170)
-0.187
(0.194)
ML Neg. Bin.
(9)
number
The table examines the relationship between timing and intensity of Israeli attacks and news pressure on U.S. media. The dependent variables are: the occurrence
of Israeli attacks (columns 1-4), the log of the number of fatalities of Israeli attacks (columns 5-8), the number of fatalities of Israeli attacks (column 9). OLS
regressions presented in columns 1-8, maximum likelihood negative binomial regression in column 9. Regressions in columns 2-3 and 6-7 include seven lags and
seven leads of news pressure; regressions in columns 4, 8 and 9 only include seven lags of news pressure. All regressions include year, calendar month, and day of
the week fixed effects. Pseudo R-squared reported in column 9, R-squared in all other columns. Standard errors clustered by month-year are reported in parentheses
in columns 1 and 5; corrected Newey-West standard errors are reported in parentheses in columns 2-4 and 6-9. *** p<0.01, ** p<0.05, * p<0.1.
4,022
0.184
0.082***
(0.022)
Palestinian attacks
(week before previous)
Observations
(Pseudo) R-squared
0.059***
(0.021)
Palestinian attacks
(previous week)
Palestinian attacks
(previous day)
4,029
0.175
0.075**
(0.035)
0.080**
(0.035)
News pressure t+1
4,071
0.172
0.026
(0.036)
0.028
(0.036)
News pressure t
OLS
(3)
0.070**
(0.032)
(2)
(1)
dummy
OLS
dummy
dummy
Occurrence of Israeli Attacks
OLS
Model:
Dependent variable:
TABLE A.5: I SRAELI ATTACKS AND U.S. N EWS P RESSURE (2000-2011),
TABLE A.6: I SRAELI ATTACKS AND PREDICTABLE VS . UNPREDICTABLE NEWS
REDUCED - FORM REGRESSIONS
(3)
Occurrence
OLS
(4)
Ln fatalities
ML Neg. Bin.
Disaster-free news pressure
(next day, residual)
0.083**
(0.034)
0.486***
(0.156)
Disaster-free news pressure
(same day, residual)
0.028
(0.034)
0.054
(0.141)
Dependent variable:
Model:
Disaster-driven news pressure
(next day, predicted)
Disaster-driven news pressure
(same day, predicted value)
FEs
Past Palestinian attacks
Observations
(Pseudo) R-squared
(1)
Occurrence
OLS
(2)
Num. fatalities
ML Neg. Bin.
-0.197
(0.283)
0.154
(0.383)
-1.078
(1.398)
0.923
(2.611)
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
4,031
0.185
4,031
0.065
4,031
0.188
4,031
0.066
The table examines the relationship between the timing and severity of Israeli attacks, news pressure,
and the occurrence of disasters. In column 1 and 2 we regress the occurrence and severity of the Israeli
attacks respectively on the predicted value of the news pressure obtained from the regression presented
in column 1 of Table 7, i.e., of daily news pressure on the occurrence and severity of the most deadly US
disasters. The predicted value captures the unpredictable disaster-driven component of news pressure. In
columns 3 and 4, we report analogous regressions, but with the residual of the news pressure obtained
from the regression presented in column 1 of Table 7. The residual captures the variation in news
pressure not driven by the onset of disasters. All regressions include day of the week, calendar month,
and year fixed effects and control for attacks by Palestinians on previous days. R-squared reported in
columns 1 and 3, pseudo R-squared in 2 and 4. Corrected Newey-West standard errors reported in
parentheses. *** p<0.01, ** p<0.05, * p<0.1.
58
TABLE A.7: F REQUENCY OF ATTACKS DURING AND AFTER THE S ECOND I NTIFADA
(2000-04 VS . 2005-11)
Intifada
Post-Intifada
Share days with Israeli attacks
59.7
25.4
Share days with Palestinian attacks
18.8
6.1
Share days with any attacks
65.6
28.8
The table reports the share of days in which an attack was carried out by either Israeli forces or Palestinian combatants, separately for the Intifada period
(September 29th 2000, beginning of our sample period, to February 8th 2005)
and the post-Intifada period (February 9th 2005 to November 24th of 2011,
end of our sample period, excluding the Gaza War period of a very intense
fighting.
59
TABLE A.8: I SRAELI RETALIATION AND U.S. NEWS PRESSURE
DURING THE S ECOND I NTIFADA (2000-2004) AND AFTER IT (2005-2011)
( FULL SAMPLE )
Dependent variable:
Model:
News Pressure (t+1)
Occurrence of Israeli Attacks
Fatalities of Israeli Attacks
(1)
(2)
(3)
(4)
(5)
(6)
OLS
OLS
OLS
ML Neg. Bin.
ML Neg. Bin.
ML Neg. Bin.
0.079**
(0.036)
0.095*
(0.056)
0.094*
(0.056)
0.569***
(0.169)
0.692***
(0.246)
0.690***
(0.246)
-0.027
(0.069)
-0.046
(0.069)
-0.225
(0.378)
0.277
(0.381)
News Pressure (t+1)
× Post-Intifada
Palestinian attack (t-1)
× News Pressure (t+1)
× Post-Intifada
0.420**
(0.173)
1.056
((0.818)
Post-Intifada
-0.270***
(0.084)
-0.234**
(0.102)
-0.216**
(0.102)
-0.282
(0.392)
0.390
(0.546)
0.443
(0.550)
Palestinian attack (t-1)
0.093***
(0.030)
0.093***
(0.030)
0.093***
(0.030)
0.405***
(0.104)
0.402***
(0.104)
0.403***
(0.105)
Palestinian attack (t-1)
× Post-Intifada
-0.097*
(0.056)
-0.097*
(0.057)
-0.451***
(0.164)
-0.204
(0.231)
-0.170
(0.235)
-1.071
(0.737)
Palestinian attack
(previous week)
0.064**
(0.029)
0.057***
(0.021)
0.059***
(0.021)
0.223**
(0.118)
0.459***
(0.103)
0.464***
(0.104)
Palestinian attack
(previous week)
× Post-Intifada
-0.012
(0.042)
Palestinian attack
(week before previous)
0.060*
(0.034)
0.373***
(0.114)
0.374***
(0.115)
Palestinian attack
(week before previous)
×Post-Intifada
0.027
(0.045)
News Pressure (same day)
News Pressure (lags)
FEs
Observations
(Pseudo) R-squared
0.462**
(0.208)
0.076***
(0.022)
0.076***
(0.022)
0.222*
(0.134)
0.260
(0.220)
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
4,040
0.186
4,040
0.186
4,040
0.178
4,040
0.058
4,040
0.057
4,040
0.057
The table replicates the analysis performed in Table but including in the sample also the days of operation “Cast-Lead”
conducted by the Israeli army between the end of December 2008 and the beginning of January 2009. All regressions
control for same-day news pressure, seven lags of news pressure, and for day of the week, calendar month, and year
fixed effects. R-squared reported in columns 1-3, pseudo R-squared in columns 4-6. Corrected Newey-West standard
errors reported in parentheses; *** p<0.01, ** p<0.05, * p<0.1.
60
61
On-site
report
(3)
Interview of
Info on
correspondent numumber of
victims
(2)
Images of
site of
attack
(4)
Footage
of the
aftermath
(5)
Photos of
victims
(6)
0.0162
(0.107)
201
0.004
0.26
-0.0175
(0.0632)
201
0.099
0.10
-0.0202
(0.0911)
Next-day coverage ×
-0.102
Palestinian attack
(0.0956)
Story appears not on the same -0.0729
day and not the day after
(0.0620)
0.0885
Story about a particular attack
(0.0642)
0.121**
Story about Palestinian attack
(0.0508)
Observations
582
R-squared
0.039
-0.0509
(0.0534)
0.0430
(0.0844)
-0.0101
(0.0351)
0.102**
(0.0463)
0.0704**
(0.0334)
582
0.129
-0.0863
(0.0852)
0.0626
(0.0979)
-0.524***
(0.0547)
0.0729*
(0.0398)
0.211***
(0.0355)
582
0.349
-0.0701
(0.0977)
201
0.048
0.74
0.0523
(0.108)
-0.0469
(0.149)
-0.420***
(0.0782)
0.0867
(0.0714)
0.110
(0.0813)
582
0.177
-0.0852
(0.122)
201
0.037
0.49
0.0287
(0.0763)
-0.0326
(0.106)
-0.292***
(0.0617)
-0.0158
(0.0684)
0.0598
(0.0498)
582
0.068
0.0452
(0.0907)
201
0.009
0.34
0.0564
(0.0272)
-0.00405
(0.0408)
-0.0174
(0.0263)
-0.0168
(0.0237)
0.0169
(0.0129)
582
0.015
0.0433
(0.0304)
201
0.014
0.01
0.0516
(0.0733)
0.0399
(0.0966)
-0.284***
(0.0587)
0.0629
(0.0536)
0.252***
(0.0463)
582
0.182
0.0454
(0.0730)
201
0.022
0.41
(9)
0.205
(0.0780)
0.0716
(0.0947)
0.0704
(0.0570)
0.185***
(0.0506)
0.220***
(0.0630)
582
0.075
-0.0195
(0.0806)
201
0.024
0.28
0.0361
(0.130)
-0.318*
(0.169)
-0.00661
(0.117)
0.389***
(0.119)
0.357***
(0.0938)
582
0.126
0.203
(0.162)
201
0.013
1.30
Reaction of Emotional
Palestinian
scale
authorities
(8)
(0.132)
0.0699
(0.172)
-0.0626
(0.0954)
-0.0911
(0.0789)
0.0763
(0.0636)
582
0.015
0.115
(0.124)
201
0.011
-0.08
Tone of
the story
(10)
The table examines qualitative differences between same-day and next-day coverage of attacks along additional dimensions than those presented in Table 10. All regressions
use the same specification as in Table 10. In columns 1 to 8 the dependent variables are dummies for whether the newscast includes: i) on site report, ii) interviews of the
correspondent, iii) information on the number of total victims, iv) images of the site of the attack, v) footage of the aftermath of the attack, vi) photos of the victims, vii)
videos of the victims, viii) information on reactions by Palestinian authorities. In columns 9 and 10 the dependent variables are i) a measure of the emotional content of
the newscast [on a 4-point scale from “not emotional at all” to “very emotional’], and of the slant of the newscasts (on a 7-point scale from “very pro-Palestine” to “very
pro-Israel”. *** p<0.01, ** p<0.05, * p<0.1. The last row of Panel A reports the mean responses to the questions for the relevant sample. The exact wording of each
question and the mean responses for the entire population of videos are reported in Appendix Table A.3.
Next-day coverage
0.0204
Panel B: all news stories about Israelo-Palestinian conflict
Observations
R-squared
Mean same-day coverage
Next-day coverage
(7)
Video of
victims
Panel A: news stories about an Israeli attack occurred on the same or previous day not mentioning Palestinian attacks
Content:
(1)
TABLE A.9: D IFFERENCE IN CONTENT BETWEEN SAME - DAY AND NEXT- DAY COVERAGE
( OTHER DIMENSIONS )
62
120.3
Predicted length of conflict-related news per attack
0.87
147.0
0.97
193.2
3384
23.4
Predicted number of victims
Predicted length of conflict-related news per month
1560
3384
28.6
1560
3384
37.6
3384
23.4
—
—
Predicted number of victims
Predicted length of conflict-related news per month
Difference in predicted number of attacks
(counter-factual vs. actual)
Difference in predicted number of victims
(counter-factual vs. actual)
27.2
33.1
-676
-291
-75
3093
2708
-185
1485
1375
213
51
21.6
3597
1611
21.0
3384
1840
361
12.0
5224
1921
9.7
3384
1560
50.0
0.57
In the first two rows we report the predicted values of the number and length of conflict-related news for different levels of news pressure, based on simple
OLS bivariate regressions of these variables on daily news pressure conditional of Israeli attack occurring on the previous day. The number of attacks
and of victims are the predicted values from OLS regressions of these variables on same-day and next-day news pressure, conditional on prior Palestinian
attacks and the whole set of fixed effects. In the time-displacement scenario the predicted values are computed at the mean value of both same-day and
next-day news pressure as well as of all covariates. In the second scenario, the predicted values are computed at different level of both same-day and nextday news pressure (mean, 1st, 25th, 75th, and 99th percentile), keeping all the rest of the covariates at their means. The predicted length of conflict-related
news per month is simply the product of the predicted length of conflict-related news per attack (in minutes) and the (scenario-specific) predicted number
of attacks divided by 133.77, i.e., the total number of months over the entire sample period.
1560
Predicted number of attacks
Cancellation scenario: attacks are canceled due to low news pressure
1560
Predicted number of attacks
1560
107.8
0.77
Percentile of the distribution of news pressure:
1
25
75
99
Time-displacement scenario: news pressure only affects the timing of attacks
0.80
Predicted number of conflict-related news per attack
Mean news pressure
TABLE A.10: C OUNTERFACTUAL EXERCISE
F IGURE A.1: F RONT- PAGE P RESS C OVERAGE OF
UNPOPULAR GOVERNMENT ACTIONS AT THE TIME OF IMPORTANT SPORTS EVENTS
Example 1: Beijing Olympics and Russia-Georgia war
Example 2: FIFA World cup and Israeli attack on Gaza
Example 3: FIFA World cup and “Save the Thief” Decree
63
F IGURE A.2: T HE DISTRIBUTION OF NEWS PRESSURE , THE U.S. TV NETWORKS
0
.5
Density
1
1.5
2
Distribution of daily news pressure (2000-2011)
0
10%
50%
1
90%
2
3
News Pressure
Note: The figure reports the distribution of news pressure on U.S. broadcast TV networks for the entire period
of interest (2000-2011). The unit of measurement is 10 minutes. The blue line represents the corresponding
Epanechnikov Kernel density estimate. The red lines represent the 10th, 50th and 90th percentiles of the
distribution.
64
F IGURE A.3: N ON - PARAMETRIC LOCAL LINEAR LEAST SQUARES BIVARIATE RELATIONSHIP
BETWEEN OCCURRENCE OR LOG NUMBER OF VICTIMS OF I SRAELI ATTACKS
AND NEXT- DAY NEWS PRESSURE
.35
lowess: Israeli attack occurred today
.4
.45
.5
.55
.6
lowess: Nat. log of nm. of victims due to Israeli attacks
.4
.6
.8
1
F ULL SAMPLE
0
1
2
3
0
1
2
NP T+1
3
NP T+1
.38
lowess: Israeli attack occurred today
.4
.42
.44
.46
lowess: Nat. log of nm. of victims due to Israeli attacks
.45
.5
.55
.6
.65
DAYS WITH NEXT- DAY NEWS PRESSURE BETWEEN MEAN AND THE 95 TH PERCENTILE
.9
1
1.1
1.2
1.3
1.4
.9
NP T+1
1
1.1
1.2
NP T+1
65
1.3
1.4
F IGURE A.4: F REQUENCY AND THE NUMBER OF VICTIMS OF I SRAELI ATTACKS
BY QUINTILES OF NEXT- DAY NEWS PRESSURE
Occurrence(and(news(pressure(
Frac%on(of(days(with(a1ack(
0.44%
0.42%
0.4%
0.38%
0.36%
0.34%
0.32%
1%
2%
3%
4%
5%
Quin%les(of(next8day(news(pressure(
Fatali-es'and'news'pressure'
Fatali.es"
4"
Fatali.es,"given"occurrence"
3.5"
Number'dead'
3"
2.5"
2"
1.5"
1"
0.5"
0"
1"
2"
3"
4"
Quin-les'of'next4day'news'pressure'
66
5"