Content uploaded by Sabine Carey
Author content
All content in this area was uploaded by Sabine Carey on Dec 19, 2019
Content may be subject to copyright.
Content uploaded by Sabine Carey
Author content
All content in this area was uploaded by Sabine Carey on Mar 09, 2017
Content may be subject to copyright.
Canaries in a coal-mine? What the killings
of journalists tell us about future repression
Anita R Gohdes
Department of Political Science, University of Zurich
Sabine C Carey
School of Social Sciences, University of Mannheim
Abstract
An independent press that is free from government censorship is regarded as instrumental to ensuring human
rights protection. Yet governments across the globe often target journalists when their reports seem to offend them
or contradict their policies. Can the government’s infringements of the rights of journalists tell us anything about
its wider human rights agenda? The killing of a journalist is a sign of deteriorating respect for human rights. If a
government orders the killing of a journalist, it is willing to use extreme measures to eliminate the threat posed by
the uncontrolled flow of information. If non-state actors murder journalists, it reflects insecurity, which can lead to
a backlash by the government, again triggering state-sponsored repression. To test the argument whether the
killing of journalists is a precursor to increasing repression, we introduce a new global dataset on killings of
journalists between 2002 and 2013 that uses three different sources that track such events across the world. The
new data show that mostly local journalists are targeted and that in most cases the perpetrators remain uncon-
firmed. Particularly in countries with limited repression, human rights conditions are likely to deteriorate in the
two years following the killing of a journalist. When journalists are killed, human rights conditions are unlikely to
improve where standard models of human rights would expect an improvement. Our research underlines the
importance of taking the treatment of journalists seriously, not only because failure to do so endangers their lives
and limits our understanding of events on the ground, but also because their physical safety is an important
precursor of more repression in the future.
Keywords
human rights, journalists, political violence, press freedom, repression
Introduction
Governments across the globe often target journalists
when reports and stories seem to offend them or contra-
dict their policies. In 2015 alone, over 70 journalists
were killed; most of them wrote about political issues.
1
Figure 1 maps the killing of journalists between 2002
and 2013. The darker the shading, the more journalists
were killed during that time period. Syria and Iraq are
among the most dangerous places for journalists, with
162 and 287 journalists, respectively, reportedly killed
between 2002 and 2013. But journalists are not only
targeted in countries that experience a civil war, as in
those two examples. Between 2002 and 2003, members
of the press corps were killed in over 80 countries. To
show that journalists are not only targeted in the most
repressive countries, Figure 2 plots the number of jour-
nalists killed under varying overall human rights condi-
tions. The x-axis represents the Political Terror Scale
Corresponding author:
gohdes@ipz.uzh.ch
1
See Committee to Protect Journalists: https://www.cpj.org/killed/
2015/ (accessed 23 February 2016).
Journal of Peace Research
2017, Vol. 54(2) 157–174
ªThe Author(s) 2017
Reprints and permission:
sagepub.co.uk/journalsPermissions.nav
DOI: 10.1177/0022343316680859
journals.sagepub.com/home/jpr
(PTS; Wood & Gibney, 2010), which captures the
extent of physical integrity rights violations – the higher
the value, the more repressive the regimes. In yellow,
we highlight the killings of journalists that occurred in
years of armed conflict, in red those that took place
outside of armed conflict. Surprisingly, outside of
armed conflict, journalists are mostly killed in countries
where governments show at least some respect for
human rights.
Journalists are frequently targeted for reporting
uncomfortable news, not just during armed conflicts and
not only in already repressive regimes. What does this tell
us about the trajectory of the overall human rights situ-
ation? Can the killing of a journalist act as an indicator
for subsequently increasing repression?
We suggest that the killing of journalists acts as a
precursor for worsening state-sponsored repression –
just as canaries in a coal-mine would be the first to
signal distress if toxic gases had leaked into the mine
and were polluting the air quality for miners. Indepen-
dent journalists will be a thorn in the side of govern-
ments who are attempting to cover up violence, or who
are trying to dominate the public narrative of why these
measures are justified. The killing of a journalist pro-
vides us with information about the government’s will-
ingness to use extreme measures to eliminate potential
dissent and to remain in control. Journalists writing
about organized crime and violence are also often tar-
geted by those whose illegal activities are brought to
light. In those instances, the killing of a journalists can
signal spiraling violence, which often results in more
repressive government behavior. In short, information
about the killings of journalists should be a valuable
indicator of human rights trajectories, highlighting
which countries are at risk of deteriorating human
rights. It can act as an early-warning signal for worsen-
ing repression, enabling policymakers to intervene or
build resilience before violence has escalated.
The killing of journalists should be particularly useful
as an early-warning signal in countries that show some,
but not great, respect for physical integrity rights. For
such countries it is difficult to tell whether they have
‘settled’ in this middle position, whether they are about
to improve their respect for human rights, or whether
they are on the verge of increasing repression. Stable and
secure countries like Canada and Australia are unlikely to
experience a sudden turn for the worse. Similarly,
nobody expects immediate serious improvements of the
bleak human rights conditions in countries embroiled in
conflict and state-sponsored violence, such as in Syria or
Sudan. While structural factors, such as economic devel-
opment or democracy, can identify countries that are
Figure 2. Number of journalists killed, by repression level, and
whether there was an armed conflictin the same year (2002–13)
Figure 1. Number of journalists reported killed between 2002 and 2013
158 journal of PEACE RESEARCH 54(2)
more likely to be at either end of the human rights scale,
they are less suited for forecasting their short-term tra-
jectory because structural characteristics change
extremely slowly (see Chiba & Gleditsch, 2017). The
killing of journalists reflects more dynamic changes that
provide us with valuable insights on likely changes in
human rights conditions in the following year, particu-
larly for moderately repressive countries, for which the
assessment of future human rights developments is
inherently difficult.
To show that the killing of journalists can provide us
with useful information about the trajectory of the over-
all human rights conditions in that country, we present a
new global dataset on killings of journalists between
2002 and 2013.The dataset provides new insights into
how many national or international journalists are tar-
geted by different perpetrators across the years. It shows
that international journalists are targeted only in rare
circumstances, and that across all killings, the perpetra-
tors usually remain unconfirmed. Our analysis shows
that if a journalist is killed, state repression is significantly
more likely to deteriorate in the following two years –
particularly in countries that previously displayed low
levels of repression. The killing of a journalist also high-
lights countries with continuing repression, where struc-
tural characteristics would suggest an improvement of
human rights conditions. In the following, we summar-
ize explanations for state-sponsored repression and ela-
borate why the killing of a journalist should help us assess
the risk of worsening of human rights conditions.
Assessing changes in government repression
Since the early global studies that analyzed why physical
integrity rights are violated by government-related actors
(see Mitchell & McCormick, 1988; Poe & Tate, 1994),
research has made great progress in identifying charac-
teristics that are associated with government repression.
Scholars often utilize decisionmaking models to explain
human rights violations. According to these models, gov-
ernments weigh up the costs and benefits of repression,
often comparing their own strengths with the perceived
threat to their leadership (Poe, 2004). For example, the
findings that democracies have better human rights
records than other forms of political regimes is explained
with democratic institutions increasing the cost of
repressive behavior (Bueno de Mesquita et al., 2005).
Empirical studies on state repression often concen-
trate their theoretical arguments and empirical models
on structural factors (for a summary, see Hill & Jones,
2014). Structural characteristics, such as regime type,
development, and population size, are useful for separat-
ing countries into more or less likely human rights offen-
ders, but are less suited for explaining developments
within countries. They can usually tell us little about
when and why governments suddenly change their strat-
egy of torturing, disappearing or killing their own peo-
ple, because these indicators change extremely slowly.
Another strand of research asks how the behavior of
actors shapes government responses. The most consis-
tent finding is that challenges to political authority in the
form of civil war and violent dissent make governments
respond with repression, labeled the ‘Law of Coercive
Responsiveness’ (Davenport, 2007: 7). Yet exactly how
dissent and repression affect each other is difficult to
assess given their endogenous relationship. Ritter &
Conrad (2016) suggest that if governments repress as a
preventative measure, the state’s response to dissent is
unclear. Although dissent should be particularly useful to
assess the risk of repression because it varies across time,
not only are data on dissent highly endogenous to repres-
sion, but reliable and valid data on dissent are hard to
come by and are usually based on coding news sources
(see Landman & Gohdes, 2013; Weidmann, 2016).
Before we outline why data on journalist killings is likely
to be reliable, we explain why the killing of journalists
can act as a precursor for worsening repression.
Government repression and the treatment
of journalists
Press freedom is widely recognized as an important
element of a well-functioning democracy. A free press
ensures that political competition can take place and
that the population is informed about the leaders’
decisions and behaviors, an essential element for
holding rulers to account (Whitten-Woodring, 2009).
Accountability is crucial for making democracies
respect human rights; only democracies that have insti-
tutionalized effective accountability measures have bet-
ter human rights records than other regimes (Bueno de
Mesquita et al., 2005).
Leaders go to great lengths to avoid accountability, for
example by outsourcing particularly heinous violence to
irregular forces (Mitchell, Carey & Butler, 2014; Carey,
Colaresi & Mitchell, 2015). Leaders who want to dom-
inate the public narrative of their performance domesti-
cally, and who want to distort the representation of their
rule internationally, are likely to have a strong incentives
to interfere with press freedom. Whenever government
actors arrange for a journalist to be killed, it is likely to
feel under pressure and is concerned about losing control
Gohdes & Carey 159
over the framing of certain issues. The killing of a jour-
nalist might be triggered by a story being uncovered that
is unfavorable for the government. The killing would
then result from unobserved earlier behavior that puts
the government in a bad light. The killing of a journalist
then picks up deteriorating security that leaves govern-
ments willing to order the murder, and can be the begin-
ning of an overall deteriorating human rights situation.
The following case summarizes this mechanism.
Ando Ratovonirina was a reporter and cameraman for
a private broadcasting company in Madagascar. He was
shot dead by presidential guards while covering antigo-
vernment demonstrations in Antananarivo in February
2009. He was the first journalist killed on duty in Mada-
gascar since the first recording of journalist deaths in
1992.
2
In the following month, the mayor of Madagas-
car’s capital city proclaimed himself president (Bearak,
2009). The next year Amnesty International expressed
serious concern about excessive violence used by the
security forces and the extent of arbitrary imprisonment.
In this example, the overall respect for human rights
declined after the reporter was murdered.
Governments might also order the killings of journal-
ists anticipating that they will carry out actions that are
worth keeping hidden from public view. In this scenario
the killing foreshadows government transgressions.
Whether it is the urge to drastically intervene in how
public debates are framed, or the desire to keep certain
facts from coming to the surface, or whether it is the
anticipation of having something to hide in the near
future, killing a journalist reflects the government’s will-
ingness and capability to use extreme measures to influ-
ence the flow of information to avoid accountability for
illegal or very unpopular actions. Once governments
have taken the step of ordering the killing of a journalist,
over time they might grow less hesitant to apply repres-
sion more widely.
Maintaining plausible deniability can involve govern-
ments ordering the killing of a journalist in order to
avoid being linked to the crime. For almost half of the
journalists killed between 2002 and 2013, the perpetra-
tor remains unconfirmed. Since governments have a
motivation to hide their connection to the killing, and
are likely to have the resources to do so, we assume that
the majority of unconfirmed cases are linked to the gov-
ernment. The murder of a journalist, irrespective of
whether the perpetrator can be linked to state forces or
whether their identity remains unconfirmed, indicates
that the government is feeling increasingly under pres-
sure, forced to act, and willing and able to act violently,
suggesting that they will become more repressive in sub-
sequent years.
For example, in August 2013, Luis de Jesu
´s Lima, a
radio journalist working in Zacapo, Guatemala was
gunned down outside his office.
3
Lima was one of five
members of the media killed in Guatemala in 2013. In
the following year, respect for human rights declined, as
police brutality and organized crime became increasingly
common. In March 2015, another journalist, Danilo
Lo
´pez, known for covering issues of corruption and the
misuse of public funds, was also shot dead while on duty.
His death goes hand in hand with increasing instability
in Guatemala, painting a bleak picture for 2016 (see the
Economist, 2015). We argue that this example reflects a
wider trend, where the killing of a journalist signals the
deterioration of overall human rights conditions.
Empirical expectation 1: The killing of a journalist by a
government-sponsored or an unconfirmed actor is
likely to signal subsequently deteriorating human
rights conditions.
There are other causes of death for journalists. Some
die in crossfire in ongoing conflicts, while others are
killed by rebel groups or criminal gangs. During armed
conflicts, rebels might have an incentive to murder
journalists as they can interfere with their own preferred
narrative of the nature and outcome of the conflict. The
murder of a journalist by a criminal gang also reflects
increasing insecurity. If journalists are killed by political
opposition groups or criminal gangs, the state has lost
the upper hand in controlling these violent groups,
which increases the risk of more government-
sponsored repression.
The death of Ahmed Rajib Haider presents an egre-
gious example for such instances. On 15 February 2013
the Bangladeshi blogger was found dead in his own
home; he had been hacked to death by religious fanatics
with machetes. Haider had written a series of critical blog
posts about extremist Islamist groups and Islamist fun-
damentalism in Bangladesh. Five perpetrators were
arrested who confessed having received the order from
Islami Chhatra Shibir, the militant student wing of
Bangladesh’s Islamist organization Jamaat-e-Islami.
4
2
See report by Committee to Protect Journalists at https://cpj.org/
killed/2009/ando-ratovonirina.php.
3
See the Committee to Protect Journalists at https://www.cpj.org/
killed/2013/luis-de-jesus-lima.php, accessed 5 October 2015.
4
See https://cpj.org/killed/2013/ahmed-rajib-haider-1.php.
160 journal of PEACE RESEARCH 54(2)
One year later, Human Rights Watch warned that Ban-
gladesh was ‘tumbling backwards’ on human rights, with
increasing powers for the government, increased civil
society restrictions, and abductions, killings, and arbi-
trary arrests by security forces becoming part of daily life
(Human Rights Watch, 2014).
Crisis hot spots are also likely to attract journalists
who aim to report about the events, putting them at
greater risk of getting in the line of fire and indicating
a deterioration of the safety in the country. During
wars,governmentsoftengotoextremestoreinin
violent groups. Valentino, Huth & Balch-Lindsay
(2004) show that governments kill large numbers of
civilians to drain the support base of the guerrillas.
Violent political groups, such as guerrilla groups, are
more threatening than non-political groups like crim-
inal gangs. The killing of journalists by political non-
state actors, such as rebel forces, can signal the
increasing threat of such forces, triggering a harsh and
quick response from the government. When criminal
gangs kill journalists, governments might not imme-
diately respond with increasing repression, but the
increase in government violence might take place
more slowly as their violence is less threatening than
those of insurgencies. But in both cases we expect
that the killing of a journalist by nongovernmental
actors leads to increasing state repression.
Empirical expectation 2: The killing of a journalist by a
nongovernmental actor is likely to signal subsequently
deteriorating human rights conditions.
We expect our measure for killed journalists to be
most useful as an early-warning indicator for increased
repression in countries with currently limited
government-sponsored violence. For regimes that
already widely use torture, political imprisonment, extra-
judicial killings, and disappearances, the killing of one
(or more) journalist is unlikely to provide us with addi-
tional insights on subsequent levels of repression.
Whitten-Woodring (2009) finds that in the most
authoritarian regimes, media freedom is related to worse
respect for human rights. Media freedom is linked to
better human rights only in the most democratic
regimes. We expect that worsening repression in the
aftermath of a journalist being killed is only visible in
countries where repression is not yet directed at the
entire population. While we expect to see a general
effect, the killing of a journalist should act as a precursor
of increased repression primarily in countries with cur-
rently limited repression.
Empirical expectation 3: The killing of a journalist is
likely to signal subsequently deteriorating human
rights conditions, particularly in countries where
repression is limited.
Under certain conditions, killings of journalists
should not be linked to increasing repression. If a gov-
ernment kills journalists to cover up future repression,
and if this strategy is successful, then we should not
observe an increase in reported repression after the kill-
ing. While not all information flows about state violence
depend on journalists, targeting journalists should
restrict the reporting of future human rights violations,
if successful – which would counteract any predictive
value the killings of journalist might have. If killing
journalists is a successful strategy in limiting the knowl-
edge of or need for repression, then the recorded level of
repression should not increase in the years following the
killing of a journalist.
A new global database on killed journalists
We present a new database that documents journalists
who were reported to have been killed between 2002 and
2013. We focus on killings as the most reliable and valid
indicator of violence committed against journalists. By
definition, lethal violence can only take place once,
whereas all other forms of violence such as imprison-
ment, torture, kidnapping, or intimidation can take
place multiple times, and vary substantially in length and
circumstances, making it harder to establish a definitive
and comparable number of reported cases. Our data use
three different sources that track such events across the
world: the Committee to Protect Journalists (CPJ), the
International Press Institute (IPI), and Reporters with-
out Borders (RWB).
Our operational definition of journalists, based on the
Committee to Protect Journalists, defines journalists as
‘people who cover news or comment on public affairs
through any media – including in print, in photographs,
on radio, on television, and online’.
5
Our database
includes associated personnel of media professionals,
such as translators and administrative staff. The inclusive
nature of this definition is particularly well suited given
the time period we study: from 2002 communication via
the Internet became an increasingly important part of
news distribution, making blogging and other forms of
online journalism increasingly influential and thus
potentially threatening for governments.
5
See https://cpj.org/killed/methodology.php.
Gohdes & Carey 161
The Committee to Protect Journalists (CPJ) is an
independent non-profit organization that promotes press
freedom, especially the right of journalists to work safely
across the world. CPJ has full-time staff working in
Africa, the Americas, Asia, Europe, Central Asia, the
Middle East, and North Africa monitoring attacks on
the press.
6
Since 1992, CPJ has maintained a list of killed
journalists across the world.
7
The International Press Institute (IPI) is a global net-
work of editors, media executives, and journalists commit-
ted to promoting press freedom and the safety of
journalists. IPI was founded in 1950 in the aftermath of
World War II and has members in over 120 countries.
Since 1997, IPI has maintained a ‘Death Watch’ list, which
records the names of journalists and media staff who were
deliberately targeted in their role as members of the press.
8
Reporters without Borders (RWB) is a non-profit orga-
nization, founded in 1985, that aims to promote the
safety of journalists, particularly by monitoring attacks
on press freedom, fighting against censorship, and provid-
ing support to journalists working in dangerous environ-
ments. RWB counts correspondents in 150 countries and
has been reporting the names of journalists killed across
the world since 2002 as part of their Press Freedom Bar-
ometer.
9
RWB only includes cases in their database where
the motive of the killing was clearly established.
We consider the information collected by these
groups to be superior to information collected through
news sources alone. All three groups act as global interest
groups for journalists and count members of the media
as part of their teams. Since all groups provide informa-
tion on each victim’s name, date, and country of death,
we match every record of a journalist killed across the
three lists. Matching was done by hand. All records were
compared to each other and determined to be a match or
not. We limit our dataset to the years from 2002 to
2013, the period for which we have data from all three
sources. The matched records provide us with the de-
duplicated number of identified journalists killed
between 2002 and 2013.
Coding the perpetrators
We use auxiliary information collected by the three
groups to establish perpetrator identities. This is no easy
undertaking, as killing members of the press corps vio-
lates international law. Governments will generally have
few incentives to broadcast the silencing of media per-
sonnel to the outside world. Furthermore, governments
will usually be in a superior position to cover their tracks
and to deny responsibility or involvement in the murder
of journalists. Therefore, for a large number of killed
journalists the perpetrator was not clearly identifiable.
In contrast to governments, nongovernmental groups
might try to gain international attention by claiming
responsibility for the killing of journalists. For example,
the radical extremist Islamic State claimed to have
beheaded US journalist James Foley in August 2014,
and al-Qaeda in Yemen claimed responsibility for the
Charlie Hebdo attack in France in January 2015.
If the killing of a journalist can be linked to officials of
the government, military, the police, pro-government
militias, paramilitary groups or troops, national guards
or death squads, intelligence or security agents, or inter-
national forces working for the government, the perpetra-
tor is coded as belonging to the State. If a journalist was
killed by a political group that was not part of the gov-
ernment, by a rebel, religious, or extremist group, by
antigovernment militants or tribal groups, the perpetrator
is coded as being a Non-state political group. Killings com-
mitted by criminals, mobs, drug gangs or influential fam-
ilies are coded as Non-political actors. Some journalists in
the database died of other causes, such as natural disasters,
diseases or in accidents. These are coded as Accidents.
Accidental deaths are excluded from all analyses. All other
recorded killings are coded as Perpetrator unconfirmed.
Journalist killings and changes in state
repression
To measure state repression we turn to the widely used
Political Terror Scale (PTS; Wood & Gibney, 2010).
PTS includes two categorical measures of physical integ-
rity rights that range from 1 to 5. The scale is based on
information from the US State Department’s yearly
Human Rights reports and Amnesty International’s
yearly country reports (Wood & Gibney, 2010).
10
The
PTS distinguishes between countries (1) under secure
rule of law, (2) with limited amount of imprisonment
6
See https://cpj.org/about/faq.php.
7
See https://cpj.org/killed/methodology.php, for their methodology
and distinction between confirmed and unconfirmed killings. From
2003 onwards, CPJ includes the deaths of media support workers (i.e.
translators and administrative staff) and we include this information
in our database as well. We include confirmed and unconfirmed
killings in our database, but we differentiate between death by
accident and deaths perpetrated by other individuals.
8
See http://ipi.freemedia.at/death-watch.html.
9
See https://rsf.org/en/barometer.
10
We rely on the measure based on the US State Department, but
where this information is missing (for example for the USA), we
impute it with the measure by Amnesty International.
162 journal of PEACE RESEARCH 54(2)
for nonviolent political activity, (3) where extensive
political imprisonment and political murders are com-
mon, (4) with large numbers of murder, imprison-
ment, and disappearances, and (5) with terror
expanded to the whole population (see Poe, Carey
& Vazquez, 2001: 658).
Changes in human rights are generally difficult to
explain because the level of observable repression changes
only very slowly; the extent of repression at time tis thus
highly dependent on the level of repression at time t–1
(Carey, 2010). This is not only because governments
change their behavior very slowly, but also because com-
mon human rights measures capture broad and therefore
relatively stable categories. For a country to move from
one category to another of the Political Terror Scale
(Wood & Gibney, 2010), substantial changes in human
rights conditions are necessary.
While using a broad human rights measure makes
detecting and explaining changes more difficult, it avoids
potential problems of endogeneity, where changes in our
main indicator, the killing of journalists, could automat-
ically trigger changes in the recorded level of repression.
Our measure of killed journalists is less dependent on
changes in media reports than measures that attempt to
count instances of repression, such as data on one-sided
killings, for example (see Weidmann, 2016).
11
If the
killing of a journalist automatically led coders to put the
country in a worse human rights category, the results
would be driven by the coding procedures. Given the
broad nature of the PTS categories, it is highly unlikely
that the killing of a journalist leads to a change in the
coding of the human rights condition. As shown below,
the overwhelming majority of journalists killed work for
locally owned news outlets and the perpetrators remain
unconfirmed. These cases receive little attention, and
due to lack of evidence, governments are not held
accountable. To further account for potential problems
of endogeneity, we model the relationship between jour-
nalist killings in the previous year and repression in the
following year, and look at two, three, four, and five-year
lags of journalists killed.
The right panel in Figure 3 reveals that the over-
whelming majority of all journalists killed between
2002 and 2013 were working in their home country.
Excluding accidents, 93% of all journalists killed were
working locally; only 7% of all journalists killed were
working in a foreign country. Since we do not identify
the country of residence of the foreign journalist, the
count also includes instances where the killed journalist
was based in a neighboring country. Hence, not all inter-
national journalists represent large international media
outlets. The distinction between national and interna-
tional journalists helps us understand what the potential
effects on the information environment are in the after-
math of a member of the press being killed. While the
killing of prominent international journalists is usually
met by international outrage, the killing of local journal-
ists generally draws far less attention, particularly where
perpetrators remain unconfirmed.
The left and middle panel of Figure 3 distinguish
between journalists who were killed during an armed
conflict and those outside of conflict years (as defined
by Melander, Pettersson & Themne
´r, 2016). As Figure 2
already showed, more journalists are killed during armed
conflict years, and the level of repression during these
years is very high.
Figure 3. Journalists killed in and outside of conflict years, 2002–13
11
As our study only covers recent years (after 11 September 2001),
we expect the effect of changes in the level of latent repression
associated with different PTS categories to be negligible (Fariss,
2014).
Gohdes & Carey 163
However, a substantial number of journalists are also
killed by state and unconfirmed perpetrators outside of
conflict. Unsurprisingly, more journalists are killed by
non-state political actors during armed conflicts than
during peace years. In contrast, a larger number of
journalists are killed by non-political actors during
peace times.
The graph in Figure 3 demonstrates that the patterns
of government targeting of journalists (including those
with unconfirmed perpetrators) look almost the same –
albeit at a different scale – during and outside of armed
conflict. A difference can be found in the number of
journalists killed by non-state political groups. Armed
groups such as rebels are more likely to attack journalists
in the midst of armed conflict than otherwise. Absent
armed conflict, very few international journalists are tar-
geted by non-state political groups.
Governments rarely target foreign journalists in
times of relative peace. Since only few killed journalists
are foreign, the risk is very small that the killing of
journalists would increase international attention and
therefore lead to more detailed human rights reports,
which might result in coding the country as more
repressive. In the majority of cases, it is local journalists
who are target and killed.
Multivariate analysis
To investigate whether the killing of a journalist can tell
us something about the overall subsequent human
rights situation, we conduct a series of multivariate
analyses controlling for the most common predictors
ofstaterepressionidentifiedintheliterature(seeHill
& Jones, 2014). Our sample includes 166 countries
between 2003 and 2014 for which information on state
repression is available and which have a population
above 500,000.
Our principal variable of interest is the number
(Count) of journalists killed in the previous year.
12
We
also test whether killed journalists signal the deteriora-
tion of human rights in the following two, three, four, or
five years. Additionally, we construct a categorical vari-
able with specific Categories that measures whether zero,
one to four, five to nine, or ten or more journalists were
killed in the previous year.
Internal dissent consistently predicts increasing
state repression (Carey, 2010). We use the UCDP/PRIO
armed conflict data (Melander, Pettersson & Themne
´r,
2016) and include a binary indicator that takes on the
value 1 if the government was involved in any organized
armed confrontation that resulted in at least 25 battle-
related deaths (UCDP/PRIO, 2014: 9) and 0 otherwise.
To account for possible changes in repression due to
institutional configurations, we include the revised Polity
scores as a measure for how democratic a country was in
any given year (Marshall & Jaggers, 2001). The Polity
project recommends recoding countries classified as
going through ‘interregnum’ periods as anocracies (with
a value of 0), while Gleditsch & Ruggeri (2010) suggest
that they are more accurately represented as highly auto-
cratic (with a value of –10). We test our models using
both operationalizations. To account for size and wealth
of a country, we include the natural log of population
size, as well as the natural log of the real gross domestic
product per capita (World Bank, 2015). With the excep-
tion of the armed conflict indicator, we lag all control
variables by one year. We include k–1 binary variables
that indicate the previous year’s repression level in each
country, where kis the number of possible categories.
Lastly, we test our third hypothesis by interacting our
journalist count variable with the binary variables that
indicate whether previous year’s repression was either at
PTS level 2 or 3.
13
How does repression change after the killing
of journalists?
We use ordered probit estimation to analyze how likely it
is that the level of repression changes in the year after a
journalist was killed. We commence by including one-
and two-year lags of the number of journalists killed.
Table I presents the results. Comparing the first and the
second models, we see that including the number of
journalists killed improves the model fit, reducing both
AIC and BIC values. The model in the second column
provides a general confirmation of our argument: the
number of journalists killed in the previous two years
is positively and significantly associated with higher lev-
els of repression. We thus find overall support that kill-
ings of journalists are associated with higher levels of
repression in the future. The remaining control variables
show the expected effects: higher levels of democracy and
economic development are associated with lower levels of
repression, while armed conflict and population size are
significantly correlated with increased repression. The
12
Our analysis commences in 2003, as the first indicator for killed
journalists is available for 2002, and includes 2014, where we look at
the effect of journalists killed in 2013.
13
Including more than two interaction terms at a time leads to an
overidentified model.
164 journal of PEACE RESEARCH 54(2)
Table I. Number of journalists killed, by perpetrator
Baseline All killed State
State þ
unconfirmed Unconfirmed Pol. group
Non-pol.
group Interaction
All killed (t–1) 0.04y0.01
(0.02) (0.02)
All killed (t–2) 0.06** 0.06**
(0.02) (0.02)
0.01
(0.03)
State perp (t–2) 0.06*
(0.03)
State þ
unconfirmed (t–1)
0.04y
(0.02)
State þunconfirmed (t–2) 0.07**
(0.02)
Unconfirmed (t–1) 0.09*
(0.05)
Unconfirmed (t–2) 0.09y
(0.05)
Pol. group (t–1) 0.30**
(0.11)
Pol. group (t–2) 0.11
(0.11)
Non-pol. group (t–1) 0.11
(0.11)
Non-pol. group (t–2) 0.21y
(0.11)
All killed (t–1)*LDV¼20.45*
(0.17)
All killed (t–1)*LDV¼30.13**
(0.05)
LDV ¼2 1.88*** 1.87*** 1.88*** 1.87*** 1.87*** 1.88*** 1.87*** 1.85***
(0.11) (0.11) (0.11) (0.11) (0.11) (0.11) (0.11) (0.11)
LDV ¼3 3.39*** 3.35*** 3.38*** 3.36*** 3.35*** 3.39*** 3.37*** 3.35***
(0.14) (0.14) (0.14) (0.14) (0.14) (0.14) (0.14) (0.14)
LDV ¼4 5 14*** 5.03*** 5.11*** 5.05*** 5.05*** 5.10*** 5.09*** 5.11***
(0.17) (0.17) (0.17) (0.17) (0.17) (0.17) (0.17) (0.18)
LDV ¼5 6.99*** 6.87*** 6.95*** 6.88*** 6.90*** 6.93*** 6.97*** 6.94***
(0.26) (0.27) (0.26) (0.27) (0.27) (0.27) (0.26) (0.27)
(continued)
165
Table I. (continued)
Baseline All killed State
State þ
unconfirmed Unconfirmed Pol. group
Non-pol.
group Interaction
Polity2 0.03*** 0.04*** 0.03*** 0.04*** 0.04*** 0.03*** 0.03*** 0.04***
(0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01)
Log pop. 0.15*** 0.14*** 0.15*** 0.14*** 0.14*** 0.15*** 0.14*** 0.14***
(0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02)
Log GDP p.c. 0.21*** 0.22*** 0.21*** 0.22*** 0.22*** 0.21*** 0.21*** 0.22***
(0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02)
Armed conflict 0 74*** 0.70*** 0.72*** 0 70*** 0.72*** 0.69*** 0.76*** 0.71***
(0.10) (0.10) (0.10) (0.10) (0.10) (0.11) (0.10) (0.10)
Threshold 1/2 0.75y0.44 0.71y0.51 0.41 0.65y0.52 0.37
(0.38) (0.39) (0.38) (0.39) (0.39) (0.38) (0.39) (0.39)
Threshold 2/3 2.96*** 2.66*** 2.93*** 2 74*** 2.64*** 2.86*** 2.74*** 2.62***
(0.39) (0.39) (0.39) (0.39) (0.39) (0.39) (0.39) (0.39)
Threshold 3/4 5.39*** 5.10*** 5.36*** 5.17*** 5.08*** 5.29*** 5.17*** 5.07***
(0.40) (0.41) (0.40) (0.40) (0.41) (0.40) (0.41) (0.41)
Threshold 4/5 7.88*** 7.64*** 7.86*** 7.70*** 7.61*** 7.84*** 7.68*** 7.60***
(0.42) (0.43) (0.42) (0.43) (0.43) (0.42) (0.43) (0.43)
AIC 2,531.89 2,514.82 2,531.11 2,519.08 2,515.72 2,520.83 2,526.80 2,505.53
BIC 2,598.30 2,592.29 2,608.58 2,596.56 2,593.19 2,598.30 2,604.27 2,594.07
Log Likelihood –1,253.94 –1,243.41 –1,251.56 –1,245.54 –1,243.86 –1,246.41 –1,249.40 –1,236.77
No. obs. 1,871 1,870 1,870 1,870 1,870 1,870 1,870 1,870
***p < 0.001, **p < 0.01, *p < 0.05, yp < 0.1. Ordered probit regression.
166
thresholds refer to the estimated cutpoints of the probit
model for the different levels of the dependent variable.
Distinguishing between different perpetrators reveals
that state-perpetrated killings of journalists today are a
significant indicator for deteriorating human rights
respect two years later. The two-year lag confirms the
assumption that levels of repression generally change
slowly, and that violence committed against the press
corps will be an early precursor of worsening human
rights conditions.
Since we expect the majority of unconfirmed killings to
be perpetrated by forces loyal to the government, we
group these cases together in the next model, and indeed
find that state and unconfirmed killings are not only sig-
nificantly associated with higher levels of repression two
years into the future, but also already in the following year.
When only looking at unconfirmed cases, the results look
quite similar. The combined evidence of these models
supports our first hypothesis that killings committed by
the state or unconfirmed perpetrators are likely to signal
subsequently deteriorating human rights conditions.
Moving to the killings committed by political groups
not affiliated with the government, we find a more
immediate relationship than with the state-related kill-
ings. The model shows that where political groups kill
journalists, the government is significantly more likely to
increase its level of repression already in the following
year. This result supports our second hypothesis and
suggests that governments increase repression fairly
quickly if political actors create insecurity that manifests
itself in the killing of journalists. The last perpetrator
group in our database is the non-political actors, such
as criminal groups or gangs. Here we see a more delayed
relationship again, where killings perpetrated by such
groups are likely to be associated with increased govern-
ment violence two years later. Killings by non-political
groups do not trigger an equally quick response from the
government as killings perpetrated by political groups,
such as insurgents. Instability due to crime is likely to be
seen as less threatening than instability resulting from
armed political groups, which could explain the different
time lags with which governments respond with
increased repression.
The final model aggregates all killings into one indi-
cator and interacts this count variable with the binary
variables that measure whether the previous levels of
repression were either 2 or 3.
14
Both interaction
coefficients are positive and significant, offering support
for the third hypothesis, stating that the killing of jour-
nalists is a particularly prominent signal of worsening
human rights in countries where repression is limited.
Comparing goodness of fit over all models in Table II,
the last model, accounting for different levels repression
in the previous year, displays the best fit. Overall, the
Table II. Different thresholds of number of journalists killed
and repression in the following year
All killed
(thresholds, t–1)
All killed
(thresholds, t–2)
1–4 killed (t–1) 0.34***
(0.09)
5–9 killed (t–1) 0.71**
(0.23)
>¼10 killed (t–1) 1.05**
(0.35)
1–4 killed (t–2) 0.38***
(0.10)
5–9 killed (t–2) 0.74**
(0.24)
>¼10 killed (t–2) 1.07**
(0.38)
LDV ¼2 1.86*** 1.85***
(0.11) (0.11)
LDV ¼3 3.33*** 3.32***
(0.14) (0.14)
LDV ¼4 4.99*** 5.01***
(0.18) (0.18)
LDV ¼5 6.83*** 6.83***
(0.27) (0.27)
Polity2 0.04*** 0.04***
(0.01) (0.01)
Log pop. 0.14*** 0.14***
(0.02) (0.02)
Log GDP p.c. 0.22*** 0.22***
(0.02) (0.02)
Armed conflict 0.71*** 0.71***
(0.10) (0.10)
Threshold 1/2 0.33 0.32
(0.39) (0.39)
Threshold 2/3 2.57*** 2.56***
(0.39) (0.39)
Threshold 3/4 5.01*** 5.01***
(0.41) (0.41)
Threshold 4/5 7.55*** 7.54***
(0.43) (0.43)
AIC 2,511.93 2,509.13
BIC 2,594.94 2,592.13
Log likelihood –1,240.96 –1,239.56
No. obs. 1,871 1,870
***p<0.001,**p< 0.01, *p<0.05,yp< 0.1. Ordered probit
regression.
14
Further model specifications including interactions with previous
levels of repression can be found in the Online appendix.
Gohdes & Carey 167
results show that models including an aggregated mea-
sure of journalists killed (regardless of perpetrator) dis-
play a better fit.
Building on these results, we rely on the aggregated
measure to construct a categorical variable with specific
Categories that measures whether zero, one to four, five
to nine, or ten or more journalists were killed in the
previous year. Table II reports the results when including
the categories as factorial variables, where the reference
category captures when no journalist was previously
killed. The first model includes the category variable as
a one-year lag, and the second includes the two-year lag.
The results show that compared to observations where
no journalists were killed, all else equal, all categories are
statistically significantly associated with higher levels of
repression. Unsurprisingly, the higher the category (i.e.
number of journalists killed) the larger the coefficient
and thus the more substantial the effect is. The second
model further confirms that two years after a journalist is
killed, and regardless of perpetrator, levels of repression
are very likely to increase.
Next, we simulate predicted probabilities using the
regression parameters of the category model with the
one-year lag, and apply them to two hypothetical obser-
vations where all control variables are held constant at
their mean or modal value, but in one case no journalists
were killed in the previous year, while in the other case a
certain number of killings occurred.
Figure 4 plots the changes in the simulated predicted
probability of each repression level, given different levels
of repression in the previous year. The lines denote the
2.5% and 97.5% quantiles. The dark red line plots the
change in predicted probability when going from zero to
one to four journalists killed, the light red line plots the
effects going from zero to five to nine journalist killed,
and the yellow line represents the model going from zero
to ten or more journalists killed in the previous year.
The top left panel shows the changes in predicted
probabilities of different repression levels in a country
without physical integrity rights violations in the previ-
ousyear(PTSat(t–1)¼1). If between one and four
journalists are killed, the probability of this country
maintaining a clean human rights record in the following
year drops by about 15 percentage points. Conversely,
the predicted probability of such a country becoming
more repressive (moving from PTS¼1toPTS¼2)
increases by about 15 percentage points. The changes
become more pronounced when between five and nine
journalists are killed, and even more extreme when ten or
more journalists are killed. The top right panel shows a
similar picture, this time for a country with limited
repression (PTS at (t–1)¼2). In the event of a low num-
ber of journalists being killed (between one and four),
the probability of moving into a higher category of
repression (PTS¼3) increases by about 10 percentage
points. Again, the changes in predicted probabilities are
substantially larger the more journalists are killed, irre-
spective of perpetrator.
The upper panels offer important evidence sup-
porting our argument: the killings of journalists seem
to be a pertinent precursor of a deteriorating human
rights situation in countries where citizens previously
enjoyed relatively high levels of protection. Regardless
of who the perpetrator is, killings of journalist are
associated with higher levels of repression in the fol-
lowing year.
The lower two panels of Figure 4 show the changes in
predicted probabilities in countries that already exercised
considerable levels of repression. To recall, a score of 3
on the Political Terror Scale means that governments
commonly torture, execute, and imprison people for
political reasons and that dissidents are frequently held
in detention for unlimited periods of time. One might
assume that the killing of journalists should not be asso-
ciated with equally visible changes in repression in coun-
tries that already practice frequent repression. The lower
left panel shows that while the killing of a journalist in
thepreviousyeardoesnotsignificantlydecreasethe
probability of a country maintaining its current level of
repression, it substantially and significantly decreases its
chances of improving its human rights score, and signif-
icantly increases the chances of this country becoming
more repressive. Similarly, countries that already saw
high levels of repression against dissidents, opposition
groups, and other politically dangerous elites in the pre-
viousyear(PTSat(t–1)¼4) are highly unlikely to
improve their human rights conditions if a journalist was
killed at (t–1).
The results suggest that where human rights are gen-
erally respected and opposition leaders can speak their
mind without running the risk of being imprisoned,
disappeared, or executed, the killing of journalists can
signal a dangerous shift in the government’s willingness
to use force against those who challenge their political
authority. Where repression is already frequently used
against perceived opponents, the killing of a journalist
is a viable indicator that there will be no improvement in
this situation in the near future.
Predictions
We turn to testing the predictive power of our model in
two steps. First, we perform one detailed out-of-sample
168 journal of PEACE RESEARCH 54(2)
prediction, where we look at individual cases that can be
correctly predicted when including information on jour-
nalists killed.
15
In a second step we compare the overall
accuracy of the different empirical models presented here
through the rank probability score, similar to Daxecker
& Prins (2017).
We divide the full sample of observations into a train-
ing set and a test set. The training set includes all obser-
vation through to the year 2011, and the test set consists
of the last three years of the dataset (2012–14). We
estimate each model using the training set, and then use
the parameters to predict the levels of repression for the
years 2012–14. We compare the baseline model and the
last model in Table I. The number of correctly predicted
outcomes as well as the rank probability score are pre-
sented as a measure of accuracy of the predictions.
Table III compares the number of correctly predicted
outcomes made by the baseline model and the model
including information on killed journalists. Including
information on journalist killings adds five correctly pre-
dicted outcomes. Table IV lists the observations where
the baseline model predicted incorrect outcomes, but
Figure 4. Changes in predicted probabilities when moving from no journalists to 1–4, 5–9, and 10 or more journalists killed in
the year before
15
See Ward & Beger (2017) on the benefits of out-of-sample
prediction.
Gohdes & Carey 169
that were correctly predicted by the journalist model. In
both Iraq and Sierra Leone repression increased in 2012.
In both cases the baseline model predicts no change in
repression, but drawing on information about journalists
killed in previous years, the journalist model is able to
predict these correctly. Interestingly, the baseline pre-
dicts a worsening human rights situation in Malaysia in
2013, but the journalist model correctly predicts a con-
sistently high human rights record. The remaining cases
confirm the pattern visible in Figure 4: where repression is
already comparably high, the killing of journalists will not
necessarily signal an increase in repression, but it will
certainly not predict an improvement in human rights
respect, even when other important variables (such as
democratic institutions, economic development, or the
absence of armed conflict) might predict improvements.
In all five cases, the baseline model predicts human rights
improvements, whereas the journalist model correctly pre-
dicts consistently high levels of repression.
Lastly, we calculate the rank probability scores as a
measure of accuracy for different model specifications
including the killing of journalists, and compare them
to the baseline model. For this we make out-of-sample
predictions for each model following the same procedure
as above, and calculate the rank probability score for each
case. The rank probability score is a measure of accuracy
of probability forecasts that is useful when forecasting
more than two categories, as with different levels of repres-
sion (measured with the PTS). Instead of only comparing
the number of correctly predicted outcomes, the rank
probability scores evaluate the difference between
probabilities produced by the forecast with the observa-
tions (see e.g. Brandt, Freeman & Schrodt, 2014: 948).
Both higher probabilities assigned to the correct out-
come category and a sharper distribution of probabil-
ities over the different categories will lead to a better
rank probability score, thus indicating a higher accuracy
of the forecast.
16
Due to the big differences in the number of journalists
killed within and outside of armed conflicts, we run our
out-of-sample predictions for different subsets of our data.
Figure 5 plots the different rank probability scores for each
model and for different subsets. The top left panel shows
the scores for the full sample using models including all
journalist killings, whereas the top right panel plots the
scores for the models that only include journalists killed by
the state and unconfirmed perpetrators. The left middle
panel shows the accuracy of the predictions of models for
the subsample of country-years where no armed conflict
took place. The right middle panel’s subsample only
excludes intrastate conflicts, and the bottom panel shows
the models trained and predicted on a sample excluding
major conflicts with over 1,000 battle-deaths.
Across all panels, the baseline model (in black) has the
highest rank probability scores, indicating that it has the
lowest level of accuracy. Adding information on journal-
ists killed thus improves the accuracy of the predictions.
For the full sample models, the interaction model (in
yellow) performs best, which is not surprising, given that
it is also the model with the best goodness of fit statistics.
The two middle panels excluding armed and intrastate
conflict do not have scores for the interaction models as
their inclusion overfits the model. The scores here indi-
cate that in ‘peaceful’ years, the models including mul-
tiple lags (all in red) of the journalist variable provide the
best accuracy – evidently killings that occurred further in
the past still function as important signals for future
repression. Keeping in mind the small scale of the x-
axis, we can conclude that while the improvements of
accuracy are substantially not very large, they are consis-
tent across different model specifications and subsamples
of the data. Including information on journalists killed
leads to previously unexpected insights on otherwise
unforeseeable levels of repression in the future. Further-
more, our models also show that when journalists are
killed, human rights conditions are unlikely to improve
where standard models of human rights would have pre-
dicted an improvement.
Table IV. Improved predictions through inclusion of informa-
tion on journalists killed in previous years
Year Country Lag_PTS PTS Baseline
2012 Iraq 3 4 3
2012 Sierra Leone 2 3 2
2013 Libya 4 4 3
2013 Malaysia 2 2 3
2013 Mexico 4 4 3
2014 Mexico 4 4 3
2014 Pakistan 5 5 4
Table III. Number of correct/false predictions made with and
without inclusion of information on journalists killed in pre-
vious years
Prediction Journalists Baseline
Wrong 110 115
Correct 361 356
16
The RPS was calculated using the verification package in R. The
scores rank from 0 to 1 where lower values indicate higher accuracy.
170 journal of PEACE RESEARCH 54(2)
Conclusion
We sought to understand whether the killing of journal-
ists is a precursor for deteriorating human rights
conditions. Independent journalists will be a thorn in
the side of governments who face internal unrest and fail
to subdue it without resorting to further repressive
means, such as political imprisonments and murder.
Figure 5. Rank probability scores for different models
Gohdes & Carey 171
Silencing critical voices in the media also plays into gov-
ernments’ attempts to dominate the public narrative of
why repressive measures against political opponents are
justified. Using new data on the killing of journalists, our
analysis shows that where a journalist was killed, repres-
sion was significantly more likely to increase in the fol-
lowing two years. Despite the difficulties in using a very
specific event, the killing of a (single) journalist, to assess
changes in very static characteristics, the violation of
physical integrity rights, our new measure can help us
to pinpoint countries that might otherwise have not been
on our radar for deteriorating human rights respect. We
find that regardless of the perpetrator, the number of
journalists killed in a country is a useful indicator of
future repression, and the relationship is most pro-
nounced in countries that currently experience only lim-
ited forms of state-sponsored repression, for which
subsequent levels of repression are usually hard to pre-
dict. Our findings emphasize the policy relevance of
carefully observing the treatment of journalists in coun-
tries with relatively good human rights records. Our
results are instructive for circumstances where targeted
and measured responses from the international commu-
nity are most likely to successfully prevent further escala-
tion of violence.
Our new dataset on journalists killed between 2002
and 2013 shows that journalists are often targeted out-
side of armed conflict. It also highlights that foreign
journalists are killed comparatively rarely and that for
most killings of a (local) journalist the perpetrator cannot
be confirmed. The killing of a journalist is usually carried
out in a way that attracts either minimal attention or
cannot be traced back to whoever ordered the killing.
It suggests that governments resort to arranging the kill-
ing of local journalists to control or limit public debate
and the flow of information, while minimizing the risk of
being linked to these crimes.
The findings also suggest that killing a journalist will
not prevent a government from being subject to some
international scrutiny. If murdering a member of the
press was a successful tool in avoiding information about
human rights violations becoming known, then we
would not be able to observe any link between the killing
of a journalist and the human rights classification of that
country. It also highlights the important work of orga-
nizations like Amnesty International and the US State
Department in uncovering instances of human rights
violations, providing alternative sources of information
that are not directly linked to the media.
Our results point to important questions for which we
currently have few systematic answers. For example,
while our new data tell us that few foreign journalists
are killed because of their profession, we do not know
how many are deported or how many are prohibited
from entering a country in the first place. Both of these
aspects are likely to provide us with insights into the
intentions of a government to avoid visibility and (inter-
national) accountability – and to use more violence
against its own population.
What might our study tell us about human rights
conditions one year from now? According to the
Committee to Protect Journalists, over 70 journalists
were murdered in 2015. For example, four journalists
were killed in Bangladesh between January and Octo-
ber 2015. Three of those were local bloggers, known
for their critical stance against radical Islamists, the
fourth a naturalized US citizen of Bangladeshi origin.
While the investigation of this latter journalist
attracted the help of the FBI, the others received little
international attention. One of the bloggers, Ananta
Bijoy Das, was murdered by unidentified perpetrators
the day after he criticized the police on Facebook for
their investigation of the murder of two other blog-
gers. While the killers of the blogger Washiqur Rah-
man Babu were captured, they had apparently no
knowledge of the activities of the blogger and
reported to have been ordered to kill this person. The
killing of bloggers by unidentified perpetrators does
not bode well for the overall human rights conditions
in Bangladesh in 2016. Other countries with mur-
dered journalists in 2015 include Brazil, Mexico, Tur-
key, Ukraine, and Kenya.
17
Killing members of the media is in and of itself an
egregious crime that demonstrates an utter failure to
respect the importance of an independent and free press.
Every mistreatment of journalists is a serious violation of
the basic right to freedom of speech. Our analysis high-
lights that it should also send out warning signals for
possible future and more widespread repression.
Replication data
The dataset and replication files for the empirical analysis
in this article, along with the Online appendix, can be
found at http://www.prio.org/jpr/datasets and https://
dataverse.harvard.edu/. All analyses were conducted
using R.
17
For details on these cases, see the Committee to Protect Journalists
at https://www.cpj.org/killed/2015/, accessed 23 February 2016.
172 journal of PEACE RESEARCH 54(2)
Acknowledgements
This project received funding from the European Research
Council under the European Union’s Seventh Framework
Programme (FP7/2007-2013)/ERC Grant Agreement no
336019. Christoph Steinert, Emmy Lindstam, and Milena
Dietrich provided excellent research assistance for this
project, and we would like to thank Lauren Young, Chris
Fariss, Håvard Hegre, and Julian Wucherpfenningfor their
helpful comments. Gohdes thanks the Belfer Center and
the Women and Public Policy Program at the Harvard
Kennedy School for institutional support.
References
Bearak, Barry (2009) Mayor declares a coup in Madagascar.
New York Times, 31January.
Brandt, Patrick T; John R Freeman & Philip A Schrodt
(2014) Evaluating forecasts of political conflict dynamics.
International Journal of Forecasting 30(4): 944–962.
Bueno de Mesquita, Bruce; George W Downs, Alastair Smith
& Feryal M Cherif (2005) Thinking inside the box: A
closer look at democracy and human rights. International
Studies Quarterly 49(3): 439–457.
Carey, Sabine C (2010) The use of repression as a response to
domestic dissent. Political Studies 58(1): 167–186.
Carey, Sabine C; Michael P Colaresi & Neil J Mitchell (2015)
Governments, informal links to militias, and accountabil-
ity. Journal of Conflict Resolution 59(5): 850–876.
Chiba, Daina & Kristian Skrede Gleditsch (2017) The shape
of things to come? Expanding the inequality and grievance
model for civil war forecasts with event data. Journal of
Peace Research 54(2): 275–297.
Davenport, Christian (2007) State repression and political
order. Annual Review of Political Science 10(1): 1–23.
Daxecker, Ursula & Brandon C Prins (2017) Financing rebel-
lion: Using piracy to explain and predict conflict intensity
in Africa and Southeast Asia. Journal of Peace Research
54(2): 215–230.
Economist (2015) A Central American spring? Fury at corrup-
tion sparks mass demonstrations. Economist, 15 August.
Fariss, Christopher J (2014) Respect for human rights has
improved over time: Modeling the changing standard of
accountability. American Political Science Review 108(2):
297–318.
Gleditsch, Kristian Skrede & Andrea Ruggeri (2010) Political
opportunity structures, democracy, and civil war. Journal of
Peace Research 47(3): 299–310.
Hill, Daniel J & Zachary M Jones (2014) An empirical eva-
luation of explanations for state repression. American Polit-
ical Science Review 108(3): 661–687.
Human Rights Watch (2014) Human Rights Watch World
Report 2014. New York: Seven Stories.
Landman, Todd & Anita R Gohdes (2013) A matter of con-
venience: Challenges of nonrandom data in analyzing
human rights violations during conflicts in Peru and Sierra
Leone. In: Jay D Aronson, Baruch Fischhoff & Taylor B
Seybolt (eds) Counting Civilian Casualties: An Introduction
to Recording and Estimating Nonmilitary Deaths in Conflict.
Oxford: Oxford University Press, 77–92.
Marshall, Monty G & Keith Jaggers (2001) Polity IV Project:
Political regime characteristics and transitions, 1800–1999.
Polity IV dataset (http://www.systemicpeace.org/polity/).
Melander, Erik; There
´se Pettersson & Lotta Themne
´r (2016)
Organized violence, 1989–2015. Journal of Peace Research
34(5): 727–742.
Mitchell, Neil J & James M McCormick (1988) Economic
and political explanations of human rights violations.
World Politics 40(4): 476–498.
Mitchell, Neil J; Sabine C Carey & Christopher K Butler
(2014) The impact of pro-government militias on human
rights violations. International Interactions 40(5): 812–836.
Poe, Steven C (2004) The decision to repress: An integrative
theoretical approach to the research on human rights and
repression. In: Sabine C Carey & Steven C Poe (eds)
Understanding Human Rights Violations: New Systematic
Studies. Aldershot: Ashgate, 16–42.
Poe, Steven C & C Neil Tate (1994) Repression of human
rights to personal integrity in the 1980s: A global analysis.
American Political Science Review 88(4): 853–872.
Poe, Steven C; Sabine C Carey & Tanya M Vazquez (2001)
How are these pictures different? A quantitative compari-
son of the US State Department and Amnesty International
human rights reports, 1976–1995. Human Rights Quarterly
23(3): 650–677.
Ritter, Emily H & Courtenay R Conrad (2016) Preventing
and responding to dissent: The observational challenges of
explaining strategic repression. American Political Science
Review 110(1): 85–99.
UCDP/PRIO (2014) UCDP/PRIO Armed Conflict Dataset
codebook. Version 4-2014 (https://www.prio.org/Data/
Armed-Conflict/UCDP-PRIO/).
Valentino, Benjamin A; Paul K Huth & Dylan Balch-Lindsay
(2004) Draining the sea: Mass killing and guerrilla warfare.
International Organization 58(2): 375–407.
Ward, Michael D & Andreas Beger (2017) Lessons from near
real-time forecasting of irregular leadership changes. Journal
of Peace Research 54(2): 141–156.
Weidmann, Nils B (2016) A closer look at reporting bias in
conflict event data. American Journal of Political Science
60(1): 206–218.
Whitten-Woodring, Jenifer (2009) Watchdog or lapdog? Media
freedom, regime type, and government respect for human
rights. International Studies Quarterly 53(3): 595–625.
Wood, Reed M & Mark Gibney (2010) The Political Terror
Scale (PTS): A re-introduction and a comparison to CIRI.
Human Rights Quarterly 32(2): 367–400.
World Bank (2015) World development indicators data (http://
data.worldbank.org/data-catalog/world-development-
indicators).
Gohdes & Carey 173
ANITA R GOHDES, b. 1986, PhD in Political Science
(University of Mannheim, 2015), Assistant Professor in
International Relations, University of Zurich (2016– ); main
research interests: communication technology and political
violence, repression, human rights measurement; recent
articles in Journal of Peace Research,Significance,Journal of
Conflict Resolution.
SABINE C CAREY, b. 1974, PhD in Government
(University of Essex, 2003); Professor, University of
Mannheim (2010– ); Research Professor, Centre for the
Study of Civil War (CSCW) at PRIO (2008–12); main
research interests: violent conflict, repression, human rights;
recent articles in International Studies Quarterly,Journal of
Conflict Resolution,International Interactions.
174 journal of PEACE RESEARCH 54(2)
