Data Set Feature

Journal of Conflict Resolution

2019, Vol. 63(9) 2207-2232
ª The Author(s) 2019
Strategies and Tactics Article reuse guidelines:
sagepub.com/journals-permissions
in Armed Conflict: How DOI: 10.1177/0022002719828103
journals.sagepub.com/home/jcr

Governments and Foreign

Interveners Respond
to Insurgent Threats
Patricia Lynne Sullivan1 , and Johannes Karreth2

Abstract
We introduce a new data set on the strategies and tactics employed by
belligerents in 197 internal armed conflicts that occurred between 1945 and
2013. The Strategies and Tactics in Armed Conflict (STAC) data set provides
scholars with a rich new source of information to facilitate investigations of how
regimes and their foreign supporters have responded to insurgent threats and
the effects of actors’ force employment choices on a wide variety of intra- and
postconflict outcomes. In addition to seventeen novel variables that measure the
strategies and tactics employed by governments and intervening states, the
STAC data set contains independently coded measures of many variables that
overlap with existing data sets—a feature that facilitates the replication of existing
studies and robustness checks on the results of new studies. We demonstrate the
utility of the STAC data with an analysis of the impact of rebel mobilization on the
basis of ethnicity on the propensity of governments to employ forced resettlement,
civilian protection, civilian welfare projects, and civilian targeting to counter the
insurgent threat.

1
Department of Public Policy, University of North Carolina at Chapel Hill, Chapel Hill, NC, USA
2
Department of Politics and International Relations, Ursinus College, Collegeville, PA, USA

Corresponding Author:
Patricia Lynne Sullivan, Department of Public Policy, University of North Carolina at Chapel Hill, 117
Abernethy Hall, Chapel Hill, NC 27599, USA.
Email: [email protected]
2208 Journal of Conflict Resolution 63(9)

Keywords
civil wars, internal armed conflict, military intervention, war outcomes, rebellion,
civilian casualties, insurgency, military strategy

As armed conflicts in Afghanistan, Yemen, and Syria grind on, policy makers,
military leaders, and scholars have all struggled to draw lessons from both current
and historical conflicts about the most effective strategies and tactics in internal
conflicts. Can insurgents be defeated through the use of brute force alone? Under
what conditions can governments undermine support for rebel forces by providing
security and public goods to civilians? Does leadership decapitation weaken rebel
groups? At the same time, recent conflicts have motivated academic research on the
causes of civilian victimization by government and rebel forces. Are combatants
who receive external support more likely to target civilians? Are democracies less
likely to engage in mass killing during counterinsurgency (COIN) campaigns?
To help address open questions in this research area and facilitate the systematic
investigation of new questions, this project provides the first comprehensive data
collection on the use of a variety of strategies and tactics by governments and foreign
interveners in nearly 200 internal armed conflicts from 1945 to 2013. The Strategies
and Tactics in Armed Conflict (STAC) data set builds upon existing data sets and
introduces novel variables coded from a wide variety of sources. While STAC codes
many variables contained in other widely used data sets, it introduces a series of
unique measures of specific COIN tactics—from public welfare projects and civilian
protection to strategic bombing and forced resettlement—employed by incumbent
governments and external interveners. Other variables, like our measures of troop
numbers, fatalities, and conflict outcomes, are similar to those coded by other
projects. For these variables, STAC often covers a longer time frame and/or provides
more detailed information about sources and coding decisions.
The data set is accompanied by case coding notes for each conflict, a detailed
codebook, and a bibliography of the 400þ sources used to code the cases. The data
set itself identifies which sources were used to code a particular case. Each case has a
unique STAC conflict identifier as well as conflict, actor, and dyad identifiers from
the Uppsala Conflict Data Program (UCDP) Armed Conflict data set (Allansson,
Melander, and Themnér 2017; Gleditsch et al. 2002) and intrastate war name and
number from the Correlates of War project (Sarkees and Wayman 2010) whenever a
corresponding observation exists in these data sets. This will allow researchers to
easily merge in additional data and to evaluate the robustness of their findings by
comparing analyses using similar variables coded in multiple data sets. A researcher
may, for example, want to compare results of models estimated with our measure of
civilian targeting by government forces to results from a model estimated with the
one-sided violence data from UCDP.
Sullivan and Karreth 2209

As we demonstrate below, these data can provide new insight into questions
about the causes and consequences of the strategic approaches adopted by govern-
ments and intervening states. In addition, the data set can address more basic ques-
tions about the historical record. How have the strategies and tactics governments
employ to counter armed opposition movements changed over the past six decades?
How common is it for regimes to provide civilian protection or other public goods in
an attempt to gain civilian support? Finally, this source of independently coded
alternative measures of many important characteristics of civil wars facilitates inves-
tigations into the robustness of results from previous and future studies.

Existing Studies and Data Sources

A substantial body of existing work has addressed questions about the strategies and
tactics employed in civil wars. First, scholars have explored the use and effective-
ness of various military strategies using in-depth case studies. Nagl’s (2002) seminal
study analyzes COIN lessons from Malaya and Vietnam, emphasizing the impor-
tance of adaptability during COIN campaigns. Kilcullen (2009) highlights, among
other things, the centrality of “information warfare” in asymmetric wars. Downes
(2007) shows that in the case of the Second Anglo-Boer War at the end of the
nineteenth century, indiscriminate violence was a partly effective COIN tool. Kauf-
mann (1996) suggests that foreign interveners’ most likely strategy for success in
halting ethnic civil wars is to separate ethnic groups. Each of these studies, and many
others, focuses on different aspects of COIN strategies, with diverging lessons
learned about the efficacy of various approaches.
Second, a considerable amount of recent empirical work uses data at the level of
local communities and subnational geographic regions to evaluate the impact of
different types of COIN strategies. These studies rely mostly on recent conflicts
in Iraq (Berman, Shapiro, and Felter 2011; Biddle, Friedman, and Long 2012; Con-
dra and Shapiro 2012), Afghanistan (Condra et al. 2010; Hultman 2012; Sexton
2016), and the Caucasus (Lyall 2009; Toft and Zhukov 2012), although some also
make use of microlevel data from the Vietnam War (Kalyvas and Kocher 2009;
Kocher, Pepinsky, and Kalyvas 2011). While these studies offer superior approaches
to estimating local effects with proper identification, their external validity is limited
by design and does not allow researchers to draw conclusions about broader trends.
A third group of studies has focused specifically on the use of different types of
force by nonstate actors and, in particular, the use of terrorist tactics and civilian
victimization. Wood (2010), Wood, Kathman, and Gent (2012), and Ottmann (2017)
all provide insight into the determinants of civilian victimization by rebels. In a
similar vein, Stanton (2013) analyzes the circumstances under which rebel groups
use terrorist strategies, while Thomas (2014) and Fortna (2015) investigate the
effects of adopting terrorist strategies on rebels’ ability to get to the negotiating
table, obtain concessions, and win civil wars. Among the most recent studies, Greig,
David Mason, and Hamner (2016) examine where and when rebels choose to fight.
2210 Journal of Conflict Resolution 63(9)

These studies reveal important insights about rebel strategies and their effects but
have limited ability to account for the other side of the strategic interaction—gov-
ernment tactics and strategic approaches to counter the insurgency.
Finally, a small set of studies has examined the effects of specific aspects of
government or intervening state COIN strategy on campaign outcomes. Lyall
and Wilson (2009) show that the mechanization of counterinsurgent forces is
associated with lower odds of winning COIN campaigns. They argue that
increasing mechanization changes a military’s approach to COIN, reducing con-
tact with the local population in a way that hinders information collection and
makes it more difficult for COIN forces to apply force selectively. Enterline,
Stull, and Magagnoli (2013) analyze the impact of broad strategy shifts on the
success of foreign powers’ COIN campaigns, finding that switching to a hearts-
and-minds (HaM) strategy modestly increases the likelihood of COIN success.
Both studies are motivated by an interest in assessing what does and does not
lead to successful COIN outcomes. While they advance our understanding of the
relationship between strategic approaches and armed conflict outcomes, neither
study is able to evaluate the breadth of COIN strategies and tactics used by
governments or interveners. Enterline, Stull, and Magagnoli specifically call for
more data, noting that their “treatment of strategy shifts is very simple, and
future research may evaluate the qualitative differences across various HaM
implementations and link these with outcomes” (p. 193).
This nonexhaustive list of studies demonstrates a strong interest in specific infor-
mation on trends, causes, and effects of strategies and tactics in internal armed
conflicts. However, currently available data cannot provide answers to many of the
big questions in this research area. Accordingly, Shelton, Stojek, and Sullivan
(2013) conclude that “an insufficient number of studies empirically evaluate the
connection between civil war COIN strategy and overall conflict outcomes across
the universe of civil war cases” (p. 526).

Available Data on Strategies and Tactics in Intrastate Conflict

Several quantitative data sets contain some information on the strategies and tactics
used in internal armed conflicts, but none of them encompass both (a) the population
of internal armed conflicts over several decades and (b) the variety of strategies and
tactics that counterinsurgents employ, which range from different types of brute
force—aerial bombing, decapitation, and mechanized assaults—to HaM strategies
like providing population security.

 Balcells and Kalyvas (2014) categorize internal armed conflicts in the UCDP/
PRIO Armed Conflict data set as conventional, irregular, or symmetric non-
conventional. The annual data on “technologies of rebellion” are available for
128 conflicts between 1946 and 2008. This classification provides some
Sullivan and Karreth 2211

information about the warfighting strategies of governments and insurgents

but does not capture variation in the specific strategies and tactics employed.
 Valentino, Huth, and Balch-Lindsay (2004) provide original data on mass
killings of civilians during 147 interstate and intrastate wars that occurred
between 1945 and 2000. Their dichotomous mass killing variable indicates
that there were at least 50,000 intentional civilian deaths within five years.
 The UCDP one-sided violence data set (Allansson, Melander, and Themnér
2017; Eck and Hultman 2007) codes direct, intentional killing of civilians by
governments and formally organized armed groups in noncombat contexts.
The annual, actor-level data go back to 1989. They exclude civilian deaths
attributed to collateral damage.
 Enterline, Stull, and Magagnoli (2013) provide data on broad strategy shifts in
sixty six COIN wars fought by foreign powers in the twentieth century.
Specifically, they record whether the foreign power’s strategy changed during
the course of the war and, if so, whether it shifted toward a HaM emphasis.
 A study from the RAND Corporation (Paul, Clarke, and Grill 2010) lists the
presence or absence of a wide range of “good” and “bad” tactics employed by
counterinsurgents in thirty insurgencies between 1978 and 2008.
 For some country-specific cases, scholars have compiled a wealth of infor-
mation on government-dissident interactions. One example of such data is the
“Government Actions in a Terror Environment data” project (Dugan and
Chenoweth 2012). This project covers Israeli government responses to ter-
rorist attacks from 1987 to 2004. Like similar projects, the event-specific
depth of such data comes at the cost of limited coverage as far as actors
(e.g., nonstate actors), types of actions, and time and space are concerned.

The STAC data advance beyond these sources by providing conflict-specific

information on a variety of strategies and tactics employed by governments and
external interveners across the population of internal armed conflicts that began after
1944 and ended between 1947 and 2013.

Data Set Construction

We define internal armed conflict as a conflict between an incumbent government
and an armed, organized, domestic opponent that resulted in at least twenty-five
deaths directly attributable to one of the actors within one year. Domestic unrest
(e.g., demonstrations, protests, rioting) that cannot be linked to an organized group
with a political objective does not qualify as an armed conflict for the purposes of
this study, even if the unrest results intwenty-five or more deaths.
We began constructing the data set by aggregating conflicts identified in the
dyadic version of the UCDP armed conflicts data set (Harbom, Melander, and
Wallensteen 2008), Correlates of War Intrastate Wars data (Sarkees and Wayman
2010), the insurgencies data set created by Lyall and Wilson (2009), the Fearon and
2212 Journal of Conflict Resolution 63(9)

Laitin (2003) list of violent civil conflicts, and Nonviolent and Violent Campaigns
and Outcomes (NAVCO) data set (Chenoweth 2011). Using data available in these
data sets, a list of sources to be consulted for all cases, and case-specific sources,
coders eliminated duplicate cases and cases that did not fit our definition of internal
armed conflict. We also revised start and end dates for conflicts when the dates
coded in the extant data sets were not consistent with our definition based on further
research into fatalities on each side of the conflict or other information about, for
example, when a peace treaty took effect. Finally, we combined some of the conflict
dyads identified by UCDP because the rebel groups had substantially similar objec-
tives, claimed to represent the same population, were active during the same time
period, and were either formally allied, part of the same umbrella organization, or
engaging in significant cooperation on the ground. In these cases, we code variables
for the primary rebel group, operationalized as the group that had the greatest
number of troops for the majority of the conflict.1 This process, which is documen-
ted in a spreadsheet available online, resulted in a final set of 197 cases. Online
Appendix A contains a list of these cases.
For each internal conflict, we coded external interventions on behalf of the
government based on existing databases and, where applicable, further research.
Sources included work by Regan and collaborators (Regan 2002; Regan and Aydin
2006), Pickering and Kisangani (2009), and Sullivan and Koch (2009). A foreign
regime maintenance (FRM) intervention was coded if an external state sent at least
500 regular,2 combat-ready troops (ground, air, or naval) to the location of the
conflict with the intent to defend or otherwise assist the incumbent government in
their fight against the insurgents.
The core of the STAC data is information on strategies and tactics employed by
counterinsurgents—governments and, when present, intervening states. This infor-
mation was coded from systematic searches of a wide range of sources. Coders
investigated whether each actor employed specific tactics and the extent to which
these tactics were emphasized in the actors’ COIN strategy. These tactics are
described in more detail below.
All coding was conducted between 2009 and 2017 by two faculty investigators
and a team of undergraduate and graduate research assistants. In addition to identi-
fying the armed conflicts to include as cases, the first eighteen months were spent in
an iterative process of defining variables, developing operational definitions, writing
coding rules, identifying sources, and coding cases. Research assistants underwent
training, including several trial runs of coding cases, and participated in biweekly
team meetings with the principle investigators in which ambiguous cases and diffi-
cult coding decisions were discussed. Coders were provided with coding procedures,
a common list of approved sources, a detailed codebook, and a template for record-
ing case notes. In the beginning, all cases were coded by two coders and discussed in
meetings to identify weaknesses in our coding rules and procedures. Once firm
coding rules were established, “practice” cases were recoded following the revised
codebook.
Sullivan and Karreth 2213

Table 1. Government COIN Strategies and Tactics.

Variable N Mean Minimum Min (%) Maximum Max (%) a AC

Force model 195 1.31 1 (light) 69 2 (heavy) 31 .90 .95

Strategic bombing 195 1.51 1 (none) 67 4 (extensive) 5 .83 .90
Civilian projects 193 1.29 1 (none) 82 4 (extensive) 2 .74 .95
Civilian protection 193 1.44 1 (none) 69 4 (extensive) 2 .80 .93
Forced 190 1.51 1 (none) 74 4 (extensive) 9 .68 .83
Resettlement
Decapitation 191 1.83 1 (none) 50 4 (extensive) 8 .80 .86
Civilian targeting 195 0.90 0 (rare) 44 2 (extensive) 33 .84 .90
Mass killing 196 1.02 0 (no) 88 1 (yes) 12 .72 .93

Note: Alpha and AC measure intercoder reliability for our assessment with two independent codings of
one-third of the data. Both measure interrater agreement correcting for expected agreement due to
chance. Alpha is the Krippendorff’s a reliability statistic. AC is Gwet’s (2014) alternative measure for
variables with rare categories.

Each research assistant consulted at least three approved sources for each
case—including peer-reviewed articles and academic books, chronologies of
international events, newspapers, and reports issued by governmental and non-
governmental organizations. When the common list of sources did not provide
sufficient information, research assistants identified case-specific sources for
approval by the faculty investigators. The codebook contains a full bibliography
of all sources and the data set, and case notes indicate which of the over 400
primary and secondary sources were used to code each case. The written instruc-
tions given to research assistants and the codebook are available in the Online
Appendix.
To access intercoder reliability, we assigned a second, independent coder to
one-third of the cases. The second coder was instructed to recode all of the
COIN strategies and tactics variables without access to the first coder’s ratings.
We then calculated chance-corrected interrater agreement coefficients for each
of these variables. The Krippendorff’s a and Gwet’s AC for each of the COIN
strategies and tactics variables are listed with other descriptive statistics in
Tables 1 and 2 (Gwet 2014; Krippendorff 2013). There is no consensus in the
social sciences about what constitutes an acceptable level of intercoder agree-
ment. Recent scholarship on intercoder reliability for content analysis has sug-
gested that a Krippendorff’s a of .8 or above indicates high reliability, while
coefficients between .667 and .8 are appropriate for more tentative conclusions
(Krippendorff 2013). However, the Krippendorff’s a is known to produce low
coefficients for variables with skewed distributions—even when intercoder
agreement is high (Feng 2015; Lacy et al. 2015). It is not surprising, therefore,
that the a coefficients for some of our tactical variables are in the more tentative
range. Gwet’s AC, an alternative measure recommended for skewed variables, is
2214 Journal of Conflict Resolution 63(9)

Table 2. Intervening State COIN Strategies and Tactics.

N Mean Minimum Min (%) Maximum Max (%) a AC

Force model 42 1.47 1 (light) 53 2 (heavy) 47 .92 .96

Ground combat 195 0.169 0 (no) 83 1 (yes) 17 .79 .93
Ground troops 197 0.197 0 (no) 80 1 (yes) 20 .86 .90
Force type 42 4.08 1 (display) 10 5 (combat) 65 .78 .88
Civilian projects 39 1.36 1 (none) 79 4 (extensive) 5 .70 .93
Civilian protection 39 1.51 1 (none) 67 4 (extensive) 5 .76 .83
Civilian targeting 38 1.50 1 (rare) 71 3 (extensive) 17 .80 .86
Decapitation 39 1.23 1 (none) 79 3 (moderate) 3 .71 .83
Strategic bombing 41 1.925 1 (none) 58 4 (extensive) 18 .84 .91

Note: Alpha is the Krippendorff’s a reliability statistic. AC is Gwet’s (2014) alternative measure for
variables with rare categories. With the exception of ground troops and ground combat, intervention
variables are only coded for conflicts in which a third-party state intervened to defend the incumbent
regime.

above .8 for all of the variables. In order to avoid introducing any personal bias
into the data, when coders disagreed on the value of a variable, we randomly
assigned one of the values (Lacy et al. 2015).3
The complete data set was also subjected to a series of intercase consistency
reviews. The principal investigators conducted analyses of variable codings
within cases that had a relatively low probability of co-occurrence and flagged
each of these cases for review by a graduate student or postdoc research assis-
tant. We flagged, for example, conflicts in which the government was coded as
engaging in both high levels of civilian targeting and high levels of civilian
protection, cases coded as having little to no evidence of civilian targeting and
extensive strategic bombing, and conflicts coded as ending in government mil-
itary victories in which the postconflict government composition was coded as
representing the opposition.4

Data Description
Like many existing data sets, STAC codes conflict initiation and termination
dates, estimates of troop levels on each side, and estimates of government and
rebel troops killed in each conflict. Nominal variables record the leader or party
affiliation of the incumbent regime and the name of the primary opposition
group. The remainder of this section focuses on variables that are either unique
to the STAC data set or coded in a more complete and detailed way than in
existing data sets.
Sullivan and Karreth 2215

Strategies and Tactics in COIN

Eight core variables—force model, strategic bombing, civilian projects, civilian
protection, forced resettlement, decapitation, civilian targeting, and mass kill-
ing—provide information about the government’s COIN tactics and strategy. Table
1 displays descriptive statistics for these variables. Our coding rules consider the
extent to which the government emphasized each tactic relative to other tactics in its
overall strategic approach to counter the threat posed by the rebel group rather than
relative to the prevalence of that tactic in other armed conflicts. While the precise
coding rules are dependent on the particular attributes of each tactic, most variables
are coded on a four-point scale, where 1 indicates that there is no evidence the tactic
was employed and 4 indicates that the government engaged in use of the tactic
frequently, deliberately, and as one of its primary tactics to counter the threat posed
by the rebel group. For the intermediate categories, the coding rules stipulate that 2
indicates minimal, infrequent government efforts and/or weak evidence of govern-
ment efforts to employ a tactic; 3 is defined as moderate use of a tactic and is coded
for cases that fall between minimal and extensive.5
Force model codes the predominant type of force used by government troops as
either light force, defined as primary reliance on ground combat units that do not
include large military equipment, or heavy force, defined as significant reliance on
large, concentrated-firepower combat with mechanized units. A second variable
(strategic bombing) categorizes the role of strategic bombing in the government’s
military strategy from 1, indicating no evidence of strategic bombing by government
forces, to 4, indicating that the government engaged in strategic bombing often,
throughout the conflict, or relied on bombing as a primary tactic for at least one-
third of the duration of the conflict. Strategic bombing is defined as centrally coor-
dinated aerial bombardment of military and civilian targets designed to destroy the
opposition’s will and ability to fight. A light force model is employed in 69 percent
of conflicts. In the majority of conflicts (67 percent), there is no strategic bombing
and extensive strategic bombing occurs in only 5 percent of conflicts.
Two variables, civilian projects and civilian protection, code what we consider to
be government tactics to win the HaM of the civilian population. Civilian projects
codes the role of “projects intended to improve the material well-being of civilians”
in the government’s COIN strategy. Coders were instructed to consider activities
such as building schools, roads, or hospitals; providing medical care, clean water, or
electricity; and creating jobs for local residents. A code of “extensive” required
evidence that the government engaged in these activities “often and/or heavily” and
“as a deliberate strategy.” Civilian protection captures the role in the government’s
military strategy of “attempts to protect noncombatants from harm caused by the
insurgents.” A code of “extensive” required “evidence that the government consis-
tently tried to protect civilians from harm and/or deliberately prioritized civilian
protection.” The government engaged in efforts to improve the material well-being
of civilians in only 18 percent of the internal armed conflicts in our data set. Efforts
2216 Journal of Conflict Resolution 63(9)

to protect civilians were more common, playing at least a minor role in government
COIN efforts in 31 percent of conflicts.
A more forceful approach is captured by the variables forced resettlement and
decapitation. Resettlement is defined as the forced relocation of civilian popula-
tions to deny an armed group access to resources, recruits, sanctuary, and other
types of support and/or to separate combatants from noncombatants. Governments
forcibly relocated civilian populations in just over a quarter of armed conflicts. In 9
percent of the conflicts, forcible resettlement was extensive. The decapitation
variable measures the degree to which the government focused on capturing and
killing top insurgent leaders. Coders found weak evidence and/or infrequent,
unsuccessful attempts at decapitation (coded as 2, minor/rare) in 25 percent of
conflicts. Decapitation played a moderate to extensive role in 24 percent of the
conflicts in our data set.
The final government COIN variables, civilian targeting, and mass killing code
whether government armed forces intentionally selected civilians (noncombatants)
as direct targets of attack or regularly conducted military operations without
attempting to discriminate between combatants and noncombatants. Mass killing
is simply a dummy variable that adopts the Valentino, Huth, and Balch-Lindsay’s
(2004) definition of mass killing as more than 50,000 intentional civilian deaths in a
campaign within five years. Our operational definition of civilian targeting differs
from other measures of civilian victimization in civil wars in that it attempts to
identify the role the use of force against noncombatants played in the government’s
COIN strategy rather than the number of civilians killed by government forces. The
variable is coded on a three-point scale indicating the extent to which the govern-
ment targeted civilians from rarely or not at all (1) to extensively (3). This oper-
ationalization is consistent with the other STAC variables, which focus on actors’
strategic and tactical choices. Although this approach is more subjective than quan-
titative estimates of the number of noncombatants killed, it has several key advan-
tages. First, considering evidence of the extent to which civilian deaths were
systematic, deliberate, and intended as a tactic to combat the insurgency creates a
measure particularly well-suited for analyses of the conditions under which govern-
ments choose to employ particular tactics. Outcome-focused operational definitions
like the number of civilians killed conflate tactical and strategic emphasis with
government capacity (e.g., troop strength and firepower). At the same time, coding
the relative role of a particular tactic in a government’s overall approach is one way
to address the difficulty of making comparisons across armed conflicts with an
almost unlimited number of distinguishing characteristics (e.g., conflict duration,
geographic spread, population density, size of rebel forces, terrain) that could impact
observable outcomes. Finally, our coding considers civilian deaths attributed to the
government as evidence the government failed to discriminate between combatants
and noncombatants but is not dependent on finding precise numerical data. Conse-
quently, our measure is available for a much longer time span than the UCDP one-
sided violence data set.
Sullivan and Karreth 2217

The government engages in civilian targeting to a moderate extent or greater in

just over half of the armed conflicts in the STAC data set. While civilian targeting is
rare in about 43 percent of the armed conflicts, moderate to extensive civilian
targeting occurred in a slightly higher percentage of cases (44 percent) and mass
killing occurred in just over 12 percent.
Considered together, the tactical variables can provide insight into the govern-
ment’s overall strategic approach to countering an insurgency—even when the
government did not consciously employ a coherent strategy throughout the conflict.
The Chinese government’s campaign against Tibetan insurgents from 1956 to 1959,
for example, is coded as consisting of a heavy force model with minor employment
of civilian projects, no evidence of decapitation or civilian protection, moderate
strategic bombing, and extensive civilian targeting. In the coding notes, the coder
states that the Chinese government “implemented many civilian projects, such as
education and infrastructure projects, but the majority of these occurred before the
outbreak of organized resistance in 1965 . . . [T]hese programs ended due to extreme
Tibetan resistance as the conflict came to a head in 1957 and did not resume until
after the fall of the resistance movement in April 1957.” As evidence for moderate
strategic bombing, the coder states that “Throughout the conflict, the Chinese army
shelled Tibetan villages and monasteries using artillery and bomber aircraft to
support troops on the ground conducting raids. Because the air attacks were quite
frequent, but the emphasis of the COIN was based on ground raids and killing
suspected insurgency sympathizers, strategicair has been coded 3.” Extensive civil-
ian targeting is coded because “The Chinese government operated with great brutal-
ity against civilians in Tibetan villages, executing hundreds. It also attacked civilians
and monasteries throughout the conflict who were suspected of helping the insur-
gents. Additionally, when insurgents and civilians gathered in Lhasa to protect the
Dalai Lama in March 1959, the Chinese army opened fire on the crowd using
mortars and machine guns, killing thousands of insurgents and civilians alike.”
In contrast, the 1963 to 1964 Sudanese campaign against the Anya-Nya rebels in
southern Sudan is characterized as employing a light force model, moderate civilian
targeting, and no evidence of civilian projects, civilian protection, or decapitation
tactics. The coder notes that “Against the Anya-Nya rebels in southern Sudan, the
use of small-arms ground fighting was the only tactic used by the government. No
other COIN strategies were employed, leading all variables to be coded as 1. How-
ever, for [civilian targeting], although there were not orders to attack civilians,
troops often ‘took out their frustrations on the civilian population’ and attacked
them in their struggle to differentiate combatants” (Ciment 2015).
Each of the COIN strategies and tactics variables is also coded for the primary
intervening state if a third-party state deployed at least 500 military troops in an
attempt to counter the insurgent threat and maintain the governing authority of the
central government. There are forty-two FRM interventions in the data set. In
addition to the strategies and tactics variables coded for the government, several
additional variables provide a more complete picture of the nature of each FRM
2218 Journal of Conflict Resolution 63(9)

intervention. Variables indicate, for example, the number of foreign troops, the
number of intervening state fatalities, and the intervention’s start and end dates.
Table 2 provides descriptive statistics for the intervention variables. The highest
category of force employed by the intervening state is coded from (1) display to (5)
ground combat operations with more than 2,000 intervening state troops. Foreign
ground troops intervene on the government’s side in fewer than 20 percent of armed
conflicts, and foreign troops engage in large-scale, direct ground combat against
rebel forces in just thirty-three cases. External interveners were more likely than
local governments to employ a heavy force model and engage in strategic bombing,
but they were also more likely to provide civilian protection and focus on civilian
projects. Civilian targeting was nonexistent or rare in 71 percent, and extensive in
only 17 percent, of the foreign interventions for regime maintenance in the data set.

Application: Ethnicity and Government

Strategies in Armed Conflict
This section demonstrates the utility of the STAC data for answering a research
question that scholars have primarily explored in case studies or with quantitative
data that were limited to specific conflicts: what determines the strategic approach a
government takes to counter an internal armed threat to the regime? When do
governments adopt the “population-centric,” HaM approach advocated by current
US military doctrine,6 and when do they adopt a more forceful approach? Are the
two approaches truly mutually exclusive? While there are lively debates in the
academic and practitioner literatures on the effectiveness of various COIN strategies
and tactics, to date scholars have not even had sufficient data on the extent to which
various strategies and tactics are employed across space and time.
Here, we develop an argument about the conditions under which governments are
likely to use violence against noncombatants and forcibly resettle civilian popula-
tions and the conditions under which they are likely to employ tactics consistent with
a HaM COIN strategy—providing security and other public goods to win the alle-
giance of the civilian population.
While there is little systematic research on the determinants of COIN strategy and
tactics, a growing literature explores the conditions under which governments use
one particular tactic—indiscriminate violence against civilians. Mason and Krane
(1989) were among the first to tackle the question of government treatment of
civilians in civil war in a systematic way. Their theoretical model predicts that
carefully targeted repressive violence can temporarily reduce active support for an
opposition movement. In contrast, more indiscriminate violence can drive civilians
to actively support the rebels because, if the government will target them regardless
of their participation status, joining the opposition could increase their security by
providing some measure of protection. They argue that state weakness, rather than
strategic utility, drives governments to resort to indiscriminate violence.
Sullivan and Karreth 2219

The model Mason and Krane present is consistent with current US military
doctrine on COIN operations, which maintains that harming civilians is counter-
productive (US Department of the Army 2006; US Marine Corps 2006). It is also
consistent with the work of scholars like Kalyvas (2006) who argue that indiscri-
minate violence increases grievances, alienates the civilian population, and provides
noncombatants with incentives to join the insurgency. Other scholars, however,
maintain that there is a strategic logic to targeting civilians and that, sometimes,
“barbarism works” (Arreguı́n-Toft 2001, 41). Or even that using overwhelming
force against civilians is necessary for COIN success (Hazelton 2017; Luttwak
2007; Trinquier 1964). As Valentino (2014) notes, “scholars have increasingly come
to recognize that large-scale violence against civilians during interstate and civil
wars is neither arbitrary, unintended, nor distinct from the central logic of war itself”
(p. 94).
Proceeding from the assumption that violence against civilians is strategic, mul-
tiple studies have found evidence that governments are more likely to engage in
mass killing to counter guerilla insurgents, as opposed to conventionally structured
opposition forces (Valentino 2014). Because the support of the population is so
critical to irregular forces, and guerilla forces themselves can be difficult to target,
governments may target civilians in terror campaigns designed to deter them from
providing materials, protection, and intelligence to the rebels (Balcells and Kalyvas
2014; Downes 2008; Valentino, Huth, and Balch-Lindsay 2004). Krcmaric (2018),
however, argues that governments should be less likely to victimize civilians in
guerilla wars precisely because these wars are a contest between the government
and the rebels for the “HaM” of the population. In support of his argument, he finds
that mass killing is more likely in civil wars fought with conventional military
strategies than in armed conflicts against guerilla forces.
There is also mixed evidence about the impact of ethnic identity on civilian
targeting. Articles by Fjelde and Hultman (2014) and Valentino (2014) note the
conspicuous lack of evidence that ethnicity drives violence again civilians in armed
conflicts. Although Montalvo and Reynal-Querol (2008) find that ethnic polariza-
tion is positively correlated with the incidence of genocide in a country, most
cross-national studies have found little connection between ethnic diversity and the
likelihood of mass killing (Azam and Hoeffler 2002; Harff 2003; Kim 2010; Rum-
mel 1995; Valentino 2004; Valentino, Huth, and Balch-Lindsay 2004; Wood 2010).
As Valentino (2004) notes, “Some of the bloodiest mass killing in history have
occurred in relatively homogeneous societies, between groups of the same or closely
related ethnicity, nationality, religion, or class” (p. 2).
Departing from an analysis of genocide specifically, Fjelde and Hultman (2014)
attempt to resolve the dissonance between case studies that emphasize the role of
ethnic identity in explaining violence against civilians (Horowitz 1985; Kaldor
2001; Kaufmann 1996; Posen 1993; Sullivan 2012) and quantitative analyses that
have failed to find a significant relationship. They argue that cross-country compar-
isons may not capture the association between ethnic divisions and civilian
2220 Journal of Conflict Resolution 63(9)

victimization because the relevant variation takes place at a lower level of analysis.
Using disaggregated, georeferenced data on one-sided violence against civilians in
sub-Saharan Africa between 1989 and 2009, they find that both governments and
rebels engage in higher levels of civilian targeting in areas of the country inhabited
predominately by co-ethnics of the opposing side.
In the following analysis, we broaden the focus beyond violence against civilians
to explore the correlates of four COIN tactics: civilian protection, civilian welfare
projects, forcible resettlement of civilian populations, and civilian targeting. We
argue that the HaM approach to COIN—improving material conditions, providing
population security, and avoiding harm to civilians—has more strategic utility if the
government and the rebels are seeking the support of the same constituency. This is
likely to be the case when the opposition is mobilized on the basis of ideology rather
than ethnicity. If mobilization is based on class or ideology, the government may be
able to discredit the rebels’ grievance narrative by providing public goods (Berman,
Shapiro, and Felter 2011; Galula and Nagl 2006; Nagl 2002; US Department of the
Army 2006). Moreover, victimizing the civilian population runs the risk of driving
uncommitted civilians to support the rebels (Kalyvas 2006; Lichbach 1987; Mason
and Krane 1989; Petersen 2001).
In contrast, when a rebel group is mobilized along ethnic lines with the aim of
overturning the ethnic balance of power in the country, the government may decide
that it can brutally suppress the ethnic population from which the rebels draw their
support without alienating their own base of support (Downes 2007; Fjelde and
Hultman 2014; Kaufmann 1996). In fact, scapegoating a marginalized ethnic group
and responding forcefully could rally support for the government (Bowen 1996;
Gagnon 1994; Gurr 2000; Horowitz 1985; Tir and Jasinski 2008). Moreover, if a
rebellion draws its support from a marginalized ethnic community, the government
may have little access to intelligence that would allow its forces to selectively target
combatants (Kalyvas 2006). Because ethnic populations frequently live in concen-
trated geographic areas or are associated through visible ascriptive characteristics,
ethnicity becomes a convenient criterion for collective targeting of the opposition’s
support base (Fjelde and Hultman 2014).
We expect, therefore, that forcible resettlement and civilian targeting will be
more likely in ethnic conflicts. The government’s COIN strategy will be more likely
to include efforts to protect civilians and improve the material welfare of the pop-
ulation when the rebels draw support across ethnic lines.
The STAC data allow for testing this argument across a broad spatial and tem-
poral domain. Ethnic mobilization is coded following a conventional definition,
whereby ethnic conflicts involve “groups that identify with a distinct ethnic or
cultural heritage” (Regan 1996, 338) “ . . . who are in conflict over the power rela-
tionship that exists between those communities and the state” (Sambanis 2001, 261).
Based on this definition, almost 40 percent of conflicts in the STAC data set were
coded as ethnic conflicts, drawing on the primary and secondary sources described
above and identified in the codebook and data set.
Sullivan and Karreth 2221

Table 3. Ethnic Conflict and COIN Tactics.

Model 1 Model 2 Model 3 Model 4 Model 5

Civilian Civilian Welfare Forced Civilian Mass

Protection Projects Resettlement Targeting Killing

Ethnic conflict 0.83 (0.29) 0.88 (0.36) 3.53 (1.23)** 2.80 (0.95)** 4.34 (1.89)**
Cut 1 constant 0.74 (0.22) 1.45 (0.24) 1.64 (0.27) 0.11 (0.18) 0.06 (0.02)
Cut 2 constant 1.98 (0.29) 2.28 (0.29) 2.37 (0.31) 1.15 (0.19)
Cut 3 constant 3.78 (0.51) 3.78 (39.08) 3.00 (0.37)
N 192 192 189 194 194
Note: Models 1 to 4 are estimated with ordered logistic regression. Model 5 is estimated with logistic
regression. Coefficients reported as odds ratios. Robust standard errors clustering on conflict country in
parentheses.
*p < .05.
**p < .01.

Results
Table 3 displays a series of models estimating the effects of ethnic conflict on the
odds that the government’s COIN strategy will include efforts to protect civilians,
material welfare projects, civilian targeting, or forced population resettlement. The
dependent variables in models 1 through 3 are our ordinal measures of the extent to
which the government employed civilian protection, material welfare projects, or
population resettlement in its COIN campaign. Each dependent variable has four
categories: none, minor/rare, moderate, and extensive. The dependent variable in
model 4, civilian targeting, has three ordered categories indicating that government
forces rarely, moderately, or extensively engaged in civilian targeting in their cam-
paign to combat the insurgency. The first four models are estimated with ordered
logit equations. Model 5 fits a logit model for a binary indicator of government mass
killing, defined as conflicts with at least 50,000 intentional civilian deaths in a five-
year period.
The results suggest that there is no difference between ethnic and nonethnic
conflicts in the extent to which material welfare projects and civilian protection
efforts are employed. The estimated odds that either tactic is employed are lower
in ethnic conflicts, but the differences are not statistically significant. In contrast,
forcible resettlement and civilian targeting play a significantly greater role in ethnic
conflicts. Most notably, the odds of state-sponsored mass killing are 4.3 times
greater when insurgents are mobilized along ethnic lines.
In Table 4, we fit ordered logit models for forced resettlement and civilian
targeting with additional independent variables to control for potential confounding
factors. The dependent variable in models 3a and b is our four-category measure of
the role of forced resettlement in the government’s COIN strategy. The dependent
variable in models 4a through 4c use is four-category measure of civilian targeting in
2222 Journal of Conflict Resolution 63(9)

which the highest category is mass killing. Not all control variables are included in
every model because missing observations on some variables result in the loss of a
significant number of cases when we include these variables.
One of the most likely alternative explanations for the use of COIN tactics that
target noncombatants is desperation; governments use force against civilians when a
conflict is exacting a high price and they are desperate to bring an end to it (Downes
2008; Harff 2003; Kalyvas 2006; Valentino 2004; Valentino, Huth, and Balch-
Lindsay 2004). To control for factors likely to increase the costs of conflict for the
government, we include categorical measures of government and rebel troop
strength and the proportion of the geographic area of the country that experienced
significant conflict-related violence. In addition, we add a continuous measure of
conflict duration—the natural log of conflict duration in days. In one of the models
predicting civilian targeting by the government, we control for civilian targeting by
rebel forces with a dichotomous variable. All variables are from the STAC data set
and are described in more detail in the codebook.
To control for the possibility that civilian victimization and forced population
resettlement are more likely in ethnic conflicts because rebels in these conflicts are
more likely to employ guerilla warfare strategies, as opposed to conventional war-
fighting strategies, we import a measure of the “technology of rebellion” from
Balcells and Kalyvas (2014). Their measure codes civil wars in the UCDP/PRIO
armed conflict database (Gleditsch et al. 2002) as irregular, conventional, or sym-
metrical nonconventional on an annual basis.7 In our models, a dummy variable
indicates whether each armed conflict was fought predominately as a conventional
war. We also test a model with a dummy variable indicating that the conflict began
after 1989 because some scholars, including Kalyvas and Balcells, maintain that
post–Cold War civil wars are distinct from those that occurred previously (Fortna
2013; Kaldor 2001; Kalyvas 2001; Kalyvas and Balcells 2010).
A dummy variable for secessionist conflicts is included because many ethnic
conflicts have secessionist aims and several scholars find lower levels of violence
and civilian victimization in wars over territorial control than wars fought for control
of the central government (Eck and Hultman 2007; Heger and Salehyan 2007). The
inclusion of the secessionist conflict indicator allows us to determine whether the
ethnic identity of the rebels, or their war aims, drives the increase in civilian victi-
mization by counterinsurgent forces.
Finally, some prior studies suggest that democratic institutions can constrain
governments from victimizing civilians in armed conflicts (Davenport and Arm-
strong 2004; Rummel 1995), although others find democratic governments just as
likely to target civilians when warfighting becomes costly (Downes 2008). We use a
continuous measure of the conflict country’s level of democracy in the year the
armed conflict began from the Polity IV data set (Marshall and Jaggers 2002).
In models with the control variables, we continue to find that ethnic conflict has a
statistically significant effect on levels of forcible resettlement and civilian target-
ing. We use model 3b to calculate the marginal effects of ethnic conflict on the role
Sullivan and Karreth 2223

Table 4. Ordered Logit Models Predicting Forced Resettlement and Civilian Targeting.

Model 3a Model 3b Model 4a Model 4b Model 4c

Forced Forced Civilian Civilian Civilian

Resettlement Resettlement Targeting Targeting Targeting

Ethnic conflict 5.00 (3.55)* 2.75 (1.34)* 9.04 (4.66)** 6.90 (3.12)** 6.17 (2.43)**
Government troops
3–10k 0.10 (0.12) 0.26 (0.23) 1.73 (1.66) 1.44 (1.18) 1.44 (1.05)
10–30k 0.78 (0.70) 0.54 (0.46) 0.98 (1.13) 0.45 (0.34) 0.54 (0.36)
>30k 0.14 (0.15) 0.28 (0.24) 1.02 (1.16) 0.78 (0.55) 0.96 (0.67)
Rebel troops
3–10k 0.24 (0.21) 0.32 (0.21)* 1.16 (1.00) 1.42 (0.83) 1.34 (0.74)
10–30k 0.22 (0.22) 0.72 (0.46) 2.52 (2.07) 4.08 (2.47)* 3.06 (1.76)
>30k 0.63 (0.64) 0.83 (0.58) 5.58 (4.36)* 4.94 (2.90)** 3.19 (1.67)*
Territorial spread
25–49% of 1.13 (0.99) 0.84 (0.43) 1.04 (0.54) 0.91 (0.39) 1.11 (0.45)
country
50–75% of 1.45 (1.51) 1.18 (1.12) 3.57 (2.66) 2.13 (1.42) 2.29 (1.38)
country
>75% of 0.96 (0.84) 0.82 (0.52) 7.04 (3.56)** 2.27 (1.11) 3.43 (1.57)**
country
Conflict duration 2.02 (0.50)** 1.89 (0.33)** 1.12 (0.17) 1.29 (0.15)* 1.18 (0.14)
Secessionist aims 1.76 (1.18) 1.61 (0.91) 1.13 (0.57) 0.58 (0.27) 0.87 (0.37)
Conventional war 0.60 (0.42) 0.71 (0.31)
Reb civ targeting 0.99 (0.32)
Cut 1 constant 4.45 (1.91) 4.43 (1.36) 2.62 (1.31) 2.61 (1.08) 2.29 (0.90)
Cut 2 constant 5.51 (1.92) 5.29 (1.37) 3.92 (1.30) 3.94 (1.09) 3.58 (0.91)
Cut 3 constant 6.19 (2.01) 6.15 (1.39) 5.92 (1.34) 5.56 (1.15) 5.11 (0.97)
Observations 103 156 106 141 161
Note: Coefficients reported as odds ratios. Robust standard errors clustered on the conflict country in
parentheses.
*p < .05.
**p < .01.

of population resettlement and model 4c to calculate the effects of ethnic conflict on

the level of civilian targeting, holding all other independent variables at their means.
The probability that forcibly resettling segments of the civilian population will play
at least a minor role in the government’s COIN strategy is more than twice as high in
ethnic conflicts—increasing from just 15 percent in nonethnic conflicts to 33 percent
in ethnic conflicts. The only other variable that is consistently statistically significant
is conflict duration, which is also positively correlated with forced resettlement. A
one standard deviation increase in conflict duration, centered on the mean, increases
the probability population resettlement will play a minor or moderate role in the
government’s strategy from 10 percent to 24 percent. The probability that deliberate,
2224 Journal of Conflict Resolution 63(9)

Figure 1. Marginal effects of ethnic conflict on extent of civilian targeting. The y-axis shows
the difference in the predicted probability of each level of civilian targeting in ethnic versus
nonethnic conflicts. Negative numbers mean that a particular level of civilian targeting is less
likely in ethnic conflicts; positive numbers indicate that that level of targeting is more likely in
ethnic conflicts. Capped bars show the 95 percent confidence interval around each point
prediction. Predictions are generated from model 4c holding all covariates constant at their
sample means.

forcible relocation of civilians will be extensively employed increases from 2 percent

to 8 percent with a one standard deviation increase in conflict duration.
Ethnic conflict is also positively correlated with civilian targeting by the
government in every model specification. While the predicted probability of
at least moderate levels of civilian targeting is approximately 43 percent when
the opposition is not mobilized on the basis of ethnicity (all other variables set
to their means), the probability of at least moderate levels of civilian targeting
rises to 82 percent in ethnic conflicts. The marginal effect of ethnic conflict on
the probability of each level of civilian targeting is plotted in Figure 1. Com-
pared to conflicts in which the opposition is not mobilized along ethnic lines,
civilian targeting is forty percentage points less likely to be rare (or nonexistent)
and twenty-two percentage points more likely to be extensive in ethnic conflicts.
Ethnic conflicts are no more likely to involve moderate levels of civilian target-
ing than nonethnic conflicts, but the probability of mass killing is seventeen
percentage points higher in ethnic conflicts. When all other variables are held
constant at their mean values, the predicted probability of mass killing by the
government is just 4 percent for nonethnic conflicts. The probability of mass
killing rises to 21 percent for ethnic conflicts.
Sullivan and Karreth 2225

There is also evidence that desperation drives governments to target civilians.

Civilian targeting plays a more prominent role in campaigns against rebels with
greater troop strength and in conflicts that have spread to more than 75 percent of the
geographic area of the country. Governments rarely engage in extensive civilian
targeting when conflicts are confined to less than a quarter of the country’s territory.
When significant conflict-related violence affects more than three-quarters of the
country, the likelihood of extensive civilian targeting or mass killing approaches 50
percent. Similarly, the predicted probability of extensive civilian targeting or mass
killing is over 40 percent if opposition forces have more than 10,000 troops. In
contrast, the probability the government will engage in more than moderate levels
of civilian targeting is less than 23 percent when rebel troop strength is under 10,000
troops. Of course, we are not able to determine the direction of causality with these
data. Governments may become more likely to victimize civilians as opposition
forces gain strength, or civilian victimization could boost rebel recruitment.
Conflict duration has a statistically significant effect on civilian victimization
levels in model 4b, but the relationship is not statistically significant in models 4a
and c. All other control variables are uncorrelated with civilian targeting, including
government troop strength, civilian victimization by the rebels, the technology of
rebellion, a post–Cold War indicator, and the regime’s level of democracy.8 Seces-
sionist aims also have no effect on the likelihood the government will target civi-
lians. Although many secessionist conflicts are also ethnic conflicts, civilian
victimization by the government is no more likely in conflicts in which the rebels
seek to create a separate state outside the current regime’s authority.
More research is clearly warranted, but this difference in COIN approaches may
be explained by an internal security dilemma; government actors do not view
reconciliation, disarmament, and reintegration as a possibility when ethnicity has
been used to mobilize citizens to take up arms against the regime. Although civilian
targeting will not win the HaM of the targeted population, in ethnic civil wars, rebels
and the government may be trying to win the support of two distinct populations.

Conclusion
The STAC data set introduces seventeen new measures of the strategies and tactics
employed by belligerents in 197 intrastate conflicts between 1945 and 2013, pro-
viding scholars with a rich new source of information to facilitate investigations into
how regimes and their foreign supporters have responded to insurgent threats. We
demonstrate the utility of the STAC data with an analysis of the impact of rebel
mobilization on the basis of ethnicity on government tactics. Our analysis demon-
strates that governments are much more likely to both forcibly resettle and use
indiscriminate violence against civilians in ethnic conflicts. Although some of the
most brutal mass killing campaigns in history have taken place during civil wars in
which ideology rather than ethnicity was the predominant division, governments are
almost twice as likely to victimize civilians, and three times as likely to engage in
2226 Journal of Conflict Resolution 63(9)

mass killing, when the armed conflict is “among communities (ethnicities) who are
in conflict over the power relationship that exists between those communities and
the state” (Sambanis 2001, 261).
Future research can use the STAC data to address a wide range of questions of
interest to both academics and policy makers. In addition to the research questions
raised in this article, the STAC data offer insight into broader trends in the use of
strategies and tactics over time, as well as variation in the COIN approaches of
different types of states. The STAC data can also be used to investigate the effects
of actors’ tactical choices or strategic approach. Examples of such questions include
whether attempts to court civilians prolong conflicts, whether mixed strategies are
more successful than a pure brute force approach, and whether the effectiveness of
strategic approaches varies across time.
One obvious limitation of the STAC data is the use of the conflict as the level of
analysis, rather than a more fine-grained focus on temporal or spatial subunits, such
as the conflict-year or geographic regions within countries. This limitation reflects
the trade-off between breadth and depth common to all data collection efforts. While
more disaggregated data sets exist, we do not know of any other data set that
provides multiple unique measures of the warfighting behavior of belligerents in
such a large number of intrastate wars. One approach for researchers interested in
time-varying effects, or concerned about identifying causal mechanisms at a lower
level of analysis, would be to take a mixed-methods approach. Just as one might
integrate quantitative and qualitative analyses, scholars can employ the STAC data
to test hypotheses about macrolevel outcomes (e.g., strategic approach, armed con-
flict duration, or termination type) and test hypotheses about the microlevel causal
mechanisms underlying those macrolevel outcomes with disaggregated data on a
smaller number of conflicts. We hope that extensions of this project, including
information at a more granular level for a more limited set of cases, will be facili-
tated by the coding notes and source material citations released with the data set.

Authors’ Note
A full list of contributors to this project is available at http://plsullivan.web.unc.edu/. This
data feature article was greatly improved by exceptionally helpful feedback from the journal
editor and anonymous reviewers. Any opinions, findings, conclusions, or recommendations
expressed are solely the responsibility of the authors and do not necessarily reflect the views
of the funding agencies, reviewers, or any other contributors.

Acknowledgments
Research for this project was made possible by all the talented undergraduate and graduate
research assistants at the University of Georgia and the University of North Carolina at Chapel
Hill who collected and coded data for this project. We are especially grateful to Frances
Duffy, Menevis Cilizoglu, and Ghazal Dezfuli for exceptional research assistance.
Sullivan and Karreth 2227

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, author-
ship, and/or publication of this article.

Funding
The author(s) disclosed receipt of the following financial support for the research, authorship,
and/or publication of this article: This study has been supported by grants from the Office of
Naval Research, US Department of the Navy [N00014-09-1-0557], and the Carnegie Corpo-
ration of New York [D 15126].

ORCID iD
Patricia Lynne Sullivan https://orcid.org/0000-0002-0028-9452
Johannes Karreth https://orcid.org/0000-0003-4586-7153

Supplemental Material
Supplemental material for this article is available online.

Notes
1. If this could not be determined, we considered how often each group was referenced in our
sources and the territorial range of each group’s attacks.
2. Our definition of “regular” troops includes special forces but excludes covert operatives
from, for example, a state’s intelligence service, pro-government militias, and proxy forces
from another state or nonstate actor.
3. Cross-tabulations of the first and second coding of each of the strategy and tactics variables
reveal that most of the disagreement between coders occurs in the intermediate categories,
indicating that researchers could increase the reliability of the least reliable variables by
collapsing the intermediate categories or by creating dichotomous variables. The raw data
with the first and second coding of each variable, the cross-tabulations, and a suite of
interrater reliability statistics are available at http://plsullivan.web.unc.edu/.
4. A list of the cases that underwent additional review, RA notes, final coding decisions, and
sources consulted in the review are also available at http://plsullivan.web.unc.edu/.
5. During the initial iterative process of writing coding rules and coding practice cases, we
found that intercoder agreement was higher, and fewer of the variables were left with
missing values, if we used four categories rather than three and the “moderate” category
was only defined as falling between “minor/rare” and “extensive.”
6. COIN manual FM 3–24 advocates a population-centric approach that emphasizes provid-
ing security for the civilian population and avoiding noncombatant deaths—even at the
expense of short-term military objectives (US Department of the Army 2006; US Marine
Corps 2006).
7. Unfortunately, this variable from Balcells and Kalyvas is only available for 68 percent of
the cases in our data set.
8. Results for models that include control variables not shown in the text are available in the
Online Appendix.
2228 Journal of Conflict Resolution 63(9)

References
Allansson, Marie, Erik Melander, and Lotta Themnér. 2017. “Organized Violence, 1989–
2016.” Journal of Peace Research 54 (4): 574-87.
Arreguı́n-Toft, Ivan. 2001. “How the Weak Win Wars: A Theory of Asymmetric Conflict.”
International Security 26 (1): 93-128.
Azam, Jean-Paul, and Anke Hoeffler. 2002. “Violence Against Civilians in Civil Wars:
Looting or Terror?” Journal of Peace Research 39 (4): 461-85.
Balcells, Laia, and Stathis N. Kalyvas. 2014. “Does Warfare Matter? Severity, Duration, and
Outcomes of Civil Wars.” Journal of Conflict Resolution 58 (8): 1390-418.
Berman, Eli, Jacob N. Shapiro, and Joseph H. Felter. 2011. “Can Hearts and Minds Be
Bought? The Economics of Counterinsurgency in Iraq.” Journal of Political Economy
119 (4): 766-819.
Biddle, Stephen, Jeffrey A. Friedman, and Stephen Long. 2012. “Civil War Intervention and
the Problem of Iraq.” International Studies Quarterly 56 (1): 85-98.
Bowen, John Richard. 1996. “The Myth of Global Ethnic Conflict.” Journal of Democracy 7
(4): 3-14.
Chenoweth, Erica. 2011. Nonviolent and Violent Campaigns and Outcomes Dataset, Vol. 1.1.
Denver, CO: University of Denver.
Ciment, James. 2015. World Terrorism: An Encyclopedia of Political Violence from Ancient
Times to the Post-9/11 Era. Hoboken, NJ: Taylor and Francis.
Condra, Luke N., Joseph H. Felter, Radha K. Iyengar, and Jacob N. Shapiro. 2010. The Effect
of Civilian Casualties in Afghanistan and Iraq. National Bureau of Economic Research.
Working Paper Series No. 16152. doi: 10.3386/w16152.
Condra, Luke N., and Jacob Shapiro. 2012. “Who Takes the Blame? The Strategic Effects of
Collateral Damage.” American Journal of Political Science 56 (1): 167-87.
Davenport, Christian, and David A. Armstrong, II. 2004. “Democracy and the Violation of
Human Rights: A Statistical Analysis from 1976 to 1996.” American Journal of Political
Science 48 (3): 538-54.
Downes, Alexander B. 2007. “Draining the Sea by Filling the Graves: Investigating the
Effectiveness of Indiscriminate Violence as a Counterinsurgency Strategy.” Civil Wars
9 (4): 420-44.
Downes, Alexander B. 2008. Targeting Civilians in War. Ithaca, NY: Cornell University
Press.
Dugan, Laura, and Erica Chenoweth. 2012. “Moving beyond Deterrence: The Effectiveness
of Raising the Expected Utility of Abstaining from Terrorism in Israel.” American Socio-
logical Review 77 (4): 597-624.
Eck, Kristine, and Lisa Hultman. 2007. “One-sided Violence against Civilians in War:
Insights from New Fatality Data.” Journal of Peace Research 44 (2): 233-46.
Enterline, Andrew J., Emily Stull, and Joseph Magagnoli. 2013. “Reversal of Fortune? Strategy
Change and Counterinsurgency Success by Foreign Powers in the Twentieth Century.”
International Studies Perspectives 14 (2): 176-98.
Sullivan and Karreth 2229

Fearon, James D., and David D. Laitin. 2003. “Ethnicity, Insurgency, and Civil War.” Amer-
ican Political Science Review 97 (1): 75-90.
Feng, Guangchao Charles. 2015. “Mistakes and How to Avoid Mistakes in Using Intercoder
Reliability Indices.” Methodology: European Journal of Research Methods for the Beha-
vioral and Social Sciences 11 (1): 13-22.
Fjelde, Hanne, and Lisa Hultman. 2014. “Weakening the Enemy: A Disaggregated Study of
Violence against Civilians in Africa.” Journal of Conflict Resolution 58 (7): 1230-57.
Fortna, Page. 2013. “Has Violence Declined in World Politics?” Perspectives on Politics 11
(2): 566-70.
Fortna, Virginia Page. 2015. “Do Terrorists Win? Rebels’ Use of Terrorism and Civil War
Outcomes.” International Organization 69 (3): 519-56.
Gagnon, V. P. 1994. “Ethnic Nationalism and International Conflict: The Case of Serbia.”
International Security 19 (3): 130-66.
Galula, David, and John A. Nagl. 2006. Counterinsurgency Warfare: Theory and Practice.
Westport, CT: Praeger Security International.
Gleditsch, Nils Petter, Peter Wallensteen, Mikael Eriksson, Margareta Sollenberg, and
Håvard Strand. 2002. “Armed Conflict 1946–2001: A New Dataset.” Journal of Peace
Research 39 (5): 615-37.
Greig, J. Michael, T. David Mason, and Jesse Hamner. 2016. “Win, Lose, or Draw in the Fog
of Civil War.” Conflict Management and Peace Science 35 (5): 523-43. doi: 10.1177/
0738894216649343.
Gurr, Ted Robert. 2000. Peoples Versus States: Minorities at Risk in the New Century.
Washington, DC: US Institute of Peace Press.
Gwet, Kilem L. 2014. Handbook of Inter-rater Reliability: The Definitive Guide to Measuring
the Extent of Agreement among Multiple Raters. Gaithersburg: Advanced Analytics.
Harbom, Lotta, Erik Melander, and Peter Wallensteen. 2008. “Dyadic Dimensions of Armed
Conflict, 1946–2007.” Journal of Peace Research 45 (5): 697-710.
Harff, Barbara. 2003. “No Lessons Learned from the Holocaust? Assessing Risks of Genocide
and Political Mass Murder since 1955.” American Political Science Review 97 (1): 57-73.
Hazelton, Jacqueline L. 2017. “The “Hearts and Minds” Fallacy: Violence, Coercion, and
Success in Counterinsurgency Warfare.” International Security 42 (1): 80-113.
Heger, Lindsay, and Idean Salehyan. 2007. “Ruthless Rulers: Coalition Size and the Severity
of Civil Conflict.” International Studies Quarterly 51 (2): 385-403.
Horowitz, Donald L. 1985. Ethnic Groups in Conflict. Berkeley: University of California
Press.
Hultman, Lisa. 2012. “Military Offensives in Afghanistan: A Double-edged Sword.” Inter-
national Area Studies Review 15 (3): 230-48.
Kaldor, Mary. 2001. New and Old Wars: Organized Violence in a Global Era. Cambridge,
UK: Polity.
Kalyvas, Stathis N. 2001. ““New” and “Old” Civil Wars.” World Politics 54:99.
Kalyvas, Stathis N. 2006. The Logic of Violence in Civil War. Cambridge, UK: Cambridge
University Press.
2230 Journal of Conflict Resolution 63(9)

Kalyvas, S. N., and M. A. Kocher. 2009. “The Dynamics of Violence in Vietnam: An Analysis
of the Hamlet Evaluation System (HES).” Journal of Peace Research 46 (3): 335-55.
Kalyvas, Stathis, and Laia Balcells. 2010. “International System and Technologies of Rebel-
lion: The End of the Cold War Shaped Internal Conflict.” American Political Science
Review 104 (3): 415-29.
Kaufmann, Chaim. 1996. “Possible and Impossible Solutions to Ethnic Civil Wars.” Inter-
national Security 20 (4): 136-75.
Kilcullen, David. 2009. The Accidental Guerrilla: Fighting Small Wars in the Midst of a Big
One. Oxford, UK: Oxford University Press.
Kim, Dongsuk. 2010. “What Makes State Leaders Brutal? Examining Grievances and Mass
Killing during Civil War.” Civil Wars 12 (3): 237-60.
Kocher, Matthew Adam, Thomas B. Pepinsky, and Stathis N. Kalyvas. 2011. “Aerial Bomb-
ing and Counterinsurgency in the Vietnam War.” American Journal of Political Science 55
(2): 201-18.
Krcmaric, Daniel. 2018. “Varieties of Civil War and Mass Killing: Reassessing the Relation-
ship between Guerrilla Warfare and Civilian Victimization.” Journal of Peace Research
55 (1): 18-31.
Krippendorff, K. 2013. Content Analysis: An Introduction to Its Methodology, 3rd ed. Thou-
sand Oaks, CA: Sage.
Lacy, Stephen, Brendan R. Watson, Daniel Riffe, and Jennette Lovejoy. 2015. “Issues and
Best Practices in Content Analysis.” Journalism & Mass Communication Quarterly 92 (4):
791-811.
Lichbach, Mark Irving. 1987. “Deterrence or Escalation? The Puzzle of Aggregate Studies of
Repression and Dissent.” Journal of Conflict Resolution 31 (2): 266-97.
Luttwak, Edward. 2007, February. “Dead End: Counterinsurgency Warfare as Military Mal-
practice.” Harper’s Magazine, 33-42.
Lyall, Jason. 2009. “Does Indiscriminate Violence Incite Insurgent Attacks? Evidence from
Chechnya.” Journal of Conflict Resolution 53 (3): 331-62.
Lyall, Jason, and Isaiah Wilson. 2009. “Rage against the Machines: Explaining Outcomes in
Counterinsurgency Wars.” International Organization 63 (1): 67-106.
Marshall, Monty G., and Jaggers Keith. 2002. Polity IV Dataset [Computer file; version
p4v2002]. Center for International Development and Conflict Management. College Park,
MD: University of Maryland.
Mason, David T., and Dale A. Krane. 1989. “The Political Economy of Death Squads: Toward
a Theory of the Impact of State-sanctioned Terror.” International Studies Quarterly 33 (2):
175-98.
Montalvo, Jose G., and Marta Reynal-Querol. 2008. “Discrete Polarisation with an Applica-
tion to the Determinants of Genocides*.” The Economic Journal 118 (533): 1835-65.
Nagl, John A. 2002. Counterinsurgency Lessons from Malaya and Vietnam: Learning to Eat
Soup with a Knife. Westport, CT: Praeger.
Ottmann, Martin. 2017. “Rebel Constituencies and Rebel Violence against Civilians in Civil
Conflicts.” Conflict Management and Peace Science 34 (1): 27-51.
Sullivan and Karreth 2231

Paul, Christopher, Colin P. Clarke, and Beth Grill. 2010. Victory Has a Thousand Fathers:
Sources of Success in Counterinsurgency. Santa Monica, CA: RAND Corporation.
Petersen, Roger D. 2001. Resistance and Rebellion: Lessons from Eastern Europe. Cam-
bridge, MA: Cambridge University Press.
Pickering, Jeffrey, and Emizet F. Kisangani. 2009. “The International Military Intervention
Dataset: An Updated Resource for Conflict Scholars.” Journal of Peace Research 46 (4):
589-99.
Posen, Barry R. 1993. “The Security Dilemma and Ethnic Conflict.” Survival 35 (1): 27-47.
Regan, Patrick M. 1996. “Conditions of Successful Third-party Intervention in Intrastate
Conflicts.” Journal of Conflict Resolution 40 (2): 336-59.
Regan, Patrick M. 2002. “Third-party Interventions and the Duration of Intrastate Conflicts.”
Journal of Conflict Resolution 46 (1): 55-73.
Regan, Patrick M., and Aysegul Aydin. 2006. “Diplomacy and Other Forms of Intervention in
Civil Wars.” Journal of Conflict Resolution 50 (5): 736-56.
Rummel, Rudolph J. 1995. “Democracy, Power, Genocide, and Mass Murder.” Journal of
Conflict Resolution 39 (1): 3-26.
Sambanis, Nicholas. 2001. “Do Ethnic and Nonethnic Civil Wars Have the Same Causes?”
Journal of Conflict Resolution 45 (3): 259-82.
Sarkees, Meredith Reid, and Frank Whelon Wayman. 2010. Resort to War: A Data Guide to
Inter-state, Extra-state, Intra-state, and Non-state Wars, 1816–2007, Correlates of War
Series. Washington, DC: CQ Press.
Sexton, Renard. 2016. “Aid as a Tool against Insurgency: Evidence from Contested and
Controlled Territory in Afghanistan.” American Political Science Review 110 (4): 731-49.
Shelton, A. M., S. M. Stojek, and P. L. Sullivan. 2013. “What Do We Know about Civil War
Outcomes?” International Studies Review 15 (4): 515-38.
Stanton, Jessica A. 2013. “Terrorism in the Context of Civil War.” The Journal of Politics 75
(4): 1009-22.
Sullivan, Christopher Michael. 2012. “Blood in the Village: A Local-level Investigation of
State Massacres.” Conflict Management and Peace Science 29 (4): 373-96.
Sullivan, Patricia L., and Michael T. Koch. 2009. “Military Intervention by Powerful States,
1945–2003.” Journal of Peace Research 46 (5): 707-18.
Thomas, Jakana. 2014. “Rewarding Bad Behavior: How Governments Respond to Terrorism
in Civil War.” American Journal of Political Science 58 (4): 804-18.
Tir, Jaroslav, and Michael Jasinski. 2008. “Domestic-level Diversionary Theory of War:
Targeting Ethnic Minorities.” Journal of Conflict Resolution 52 (5): 641-64.
Toft, Monica Duffy, and Yuri M. Zhukov. 2012. “Denial and Punishment in the North
Caucasus: Evaluating the Effectiveness of Coercive Counter-insurgency.” Journal of
Peace Research 49 (6): 785-800.
Trinquier, Roger. 1964. Modern Warfare: A French View of Counterinsurgency. New York:
Praeger.
US Department of the Army. 2006. Field Manual 3–24: Counterinsurgency. Washington,
DC: GPO.
US Marine Corps. 2006. Warfighting Publication No. 3–33.5. Washington, DC: GPO.
2232 Journal of Conflict Resolution 63(9)

Valentino, Benjamin A. 2004. Final Solutions: Mass Killing and Genocide in the Twentieth
Century. Ithaca, NY: Cornell University Press.
Valentino, Benjamin A. 2014. “Why We Kill: The Political Science of Political Violence
against Civilians.” Annual Review of Political Science 17 (1): 89-103.
Valentino, Benjamin, Paul Huth, and Dylan Balch-Lindsay. 2004. ““Draining the Sea”: Mass
Killing and Guerrilla Warfare.” International Organization 58 (2): 375-407.
Wood, Reed M. 2010. “Rebel Capability and Strategic Violence against Civilians.” Journal of
Peace Research 47 (5): 601-14.
Wood, Reed M., Jacob D. Kathman, and Stephen E. Gent. 2012. “Armed Intervention and
Civilian Victimization in Intrastate Conflicts.” Journal of Peace Research 49 (5): 647-60.