Cornering the Market: Optimal Governmental

Responses to Competitive Political Violence

William Spaniel
October 17, 2016

Abstract
I develop a model that endogenizes entry into the market of competitive political
violence. In equilibrium, an existing group may overproduce violence to corner
their market and deter entry by a potential competitor. Contrary to some hypotheses about outbidding, violence can therefore be greater with only a single
group than when the second group enters the market. I then investigate four
manners by which a target government might mitigate the violence: offensive
measures that undermine the lead groups marginal cost of violence, defensive
measures that absorb a portion of all violence, deterrent measures that increase
the cost of group formation, and concessions to the groups audience to reduce
grievances. Of these, only defensive measures are guaranteed to decrease violence;
increasing the burden of entry and decreasing grievances can counterintuitively
increase violence.

Assistant Professor,
Department of Political Science,
([email protected], http://williamspaniel.com).

Uniersity

of

Pittsburgh.

Introduction

Over the last decade, scholars of political violence have noted that the motivation of
many attacks is not policy change (Cronin 2011, 40). Instead, organizations sometimes
use violence to communicate with each other and audiences sympathetic to their ideology. Correspondingly, the theory of outbidding conceptualizes violence as a means
through which groups compete for recruits and resources.1 Violence serves as an advertisement, and greater competition forces each group to spend more to broadcast their
message. This literature often concludes that as the number of active groups increases,
so does this competition, and thus so does the violence produced (e.g., Bloom 2005,
95).
That said, critics wonder about the generality of outbidding. Moghadam (2008, 36),
for instance, notes that the Tamil Tigers began employing especially violent suicide
missions in 1987, after direct competition had subsided. Furthermore, some of the
most deadly terrorist and insurgent groups have been relatively hegemonic during their
peak violence periods: Euskadi Ta Askatasuna (ETA), the Irish Republican Army,
al-Qaeda in the late 1990s and early 2000s, and the Islamic State in the mid 2010s.
Correspondingly, large-n statistical analyses have found an inconsistent relationship
between groups and violence, with some scholars finding a positive relationship and
others finding no relationship.2
In this paper, I argue that the theory of outbidding needs to expand beyond observed competition. Organizations in the market for political violence must fear not
only present competitors but also competitors on the horizon. Consequently, I develop
a model of competitive political violence between an existing group and a potential entrant into the market. In equilibrium, existing organizations have incentive to saturate
the market in violence, capture available resources, and deny potential competitors a
profitable entry. Surprisingly, this strategy of cornering the market can be more violent than when groups actively compete with one another. The theory of outbidding
therefore does not guarantee that more observed groups implies more violence, which
makes sense of the contradictory empirical findings.
1

See Crenshaw 1985; Horowitz 1985; Oots 1989; Kydd and Walter 2006; McCauley and Moskalenko
2008.
2
See Clauset et al 2010; Findley and Young 2012; Stanton 2013; Nemeth 2014; Jaeger et al 2015;
Fortna 2015; Conrad and Greene 2015.

While this finding is interesting in its own right, the model also produces a number
of empirical and policy implications. Policymakers have suggested or tried at least four
different means to reduce violence against target governments: (1) increasing the cost of
group formation, (2) decreasing grievances among those who might support an organization, (3) launching offensive measures to destroy the infrastructure of existing groups,
and (4) hardening potential targets of attacks.3 I show that one cannot assess the effectiveness of counterterrorism and counterinsurgency strategies without understanding
the second-order effects of group competition. Indeed, only one of these strategies is
fully effective, and two may counterintuitively increase the prevalence of violence if the
target does not fully commit to them. These discrepancies help explain inconsistencies
in the empirical literature on counterterrorism (Lum, Kennedy, and Sherley 2006).
First, increasing the cost of group formation can backfire. When the costs of formation are low, cornering the market looks unattractive for the lead group. This is because
it is easy for its competitors to enter, thereby requiring the lead group to exert too much
effort to deter another from entering. When the costs of formation are high, however,
the lead group finds cornering the market attractive. Because cornering can require
more violence than the competitive level, overall violence can increase at this transition
point. Violence ultimately declines as the cost grows sufficiently largemeaning that
the target may have to fully commit to the strategy to see any positive effects.
Second, decreasing grievances among those who might support an organization can
also lead to an increase in violence. The logic is similar to the previous case. When those
wishing to provide membership and support to political violence producers is low, the
market is less likely to support multiple competing groups. This incentivizes the lead
group to corner the market. When the pool of membership and support is large, the lead
group may permit competitors to enter. As before, because competitive markets can
have less violence overall, shifting from the competitive market to the corner market
can spike the level of violence. This helps explain the unclear relationship between
reducing grievances and violence that the literature has previously uncovered (Brancati
2006; Dugan and Chenoweth 2012). As before, fully committing to reducing grievances
eventually leads to a decrease in violence.
Third, I show that defensive measures are an assuredly effective strategy, as they
3

See, for example, the National Strategy for Combating Terrorism (http://20012009.state.gov/s/ct/rls/wh/71803.htm), Pillar 2001 (29-40), and Netanyahu 1995 (132-147).

lead to a guaranteed reduction in violence. Because defensive measures do not target a

single group, they reduce the violence for both the leader and any potential competitors.
In turn, and in contrast to the previous two cases, defensive measures do not make
cornering the market more attractive. And regardless of whether the lead group corners
the market or the groups compete, the defensive measures counter a portion of overall
violence as intended.
Finally, offensive measures to destroy the infrastructure of existing groups has
weakly positive effects. Such tactics increase the amount of effort a lead group to
produce violence. When this shifts the lead groups preference from cornering the market to allowing competition, violence overall violence decreases; this strategy does not
lead to a spike like in the previous two cases because it directly reduces the lead groups
violence output even when the market is competitive. It similarly leads to lower overall violence when another group would have entered the market in the absence of an
intervention. However, if the lead group would corner the market without an intervention and still prefers to corner given the (possibly minimal) intervention, violence
remains static. This is because the lead group calculates its level of violence to deter
competition, and the intervention does not alter a competitors payoff for entry.
The optimal offensive strategy has a second key policy implication. If the target
wishes to offensively intervene against a cornering group, it must do so in a manner that
leads to other groups competing. Critics of the War on Terror argue that such measures are ineffective because they resemble Whac-a-Moleknock one terrorist group
down, and another simply springs up in its place. The model validates this replacement
effect. However, critics overlook how debilitating the lead group disincentivizes high
levels of cornering violence and cripples that lead group in future competition. Thus,
whack-a-terrorist-group can be effective even if it inspires more group formation.

A Model of Cornering and Competition

This section begins the analysis with a game featuring a pair of violent non-state actors;
later, I will add a target government that can manipulate competition between them.
I begin by describing the sequence of play, then I solve for the equilibrium and analyze
the prevalence of violence across the observable outcomes.

2.1

The Game

Players. The two-player baseline model contains a pair of violent non-state actors.
Group 1 is an existing producer of violence seeking to keep its support flowing. Group
2 is considering whether to enter the market and compete for Group 1s resources.
I feature only two groups for transparency in the results. Nevertheless, it is worth
stressing that the fundamental theoretical results I present are robust to interactions
featuring multiple existing groups and multiple potential groups deciding whether to
enter the market. This is because cornering the market to deny all potential competitors
entry allows all existing groups to share a larger pie, making the existing groups more
willing to overproduce violence to obtain this result.
Actions and Timing. As Figure 1 illustrates, play begins with Group 1 choosing a
level of violence v1 0. Larger values represent greater effort exerted, which in turn
leads to more violence against the target government that I will introduce later. Group
2 then observes Group 1s selection and decides whether to enter the market or not.
Entering the market costs c (0, 1), which represents the fixed costs of creating the
organization.4 If Group 2 enters, it selects a level of violence of its own v2 0. The
game ends.
Preferences and Payoffs. I use a contest success function to map group effort into a
1
m1 v1 , where m1 > 0
division of audience resources. Consequently, Group 1 earns v1v+v
2
represents Group 1s marginal cost for producing violence.5 Note that if Group 2 does
not enter, v2 = 0 and therefore Group 1s payoff simplifies to 1m1 v1 . Group 2s payoff
2
is similar, but it potentially pays the cost c to enter. Thus, it earns v1v+v
m2 v2 c if
2
it enters, where m2 > 0 represents its marginal cost of violence. This simplifies to 0 if
Group 2 stays out.
Note that in this setup clarifies the groups motives as maximizing marketshare. This
is the underlying motivation in the outbidding literature (Bloom 2005) and reflects the
4

I constrain c below 1 because the value of benefits is standardized to 1. Thus, if c > 1, Group 2
would want to stay out under all conditions. The rest of the analysis would be trivial and not be of
theoretical relevance, so I omit it.
5
Contest success functions like this are undefined when v1 + v2 = 0. This is not problematic,
however, because regardless of how the benefits are divided, either party could profitably deviate to
some arbitrarily small amount and capture the entire quantity.

Group 1

v1
0

Group 2

Quit

Enter

Group 2

1 m1 v1 , 0

v2
0
v1
v1 +v2

2
m1 v1 , v1v+v
m2 v2 c
2

Figure 1: The two-player extensive form game.

idea that some terrorist groups are not primarily motivated by policy change (Cronin
2011, 40). It also shows how groups can credibly threaten to use costly violence even
in the absence of political goals. Including such preferences into the utility functions
does not substantially alter the results presented below. Specifically, it does not change
the instances in which a cornered market features more violence than a competitive
market, nor does it affect how interventions into the market can increase or decrease
violence.6

2.2

Solving for the Optimal Levels of Violence

Because this is a sequential game of complete information, I search for its subgame
perfect equilibria (SPE). SPE refines Nash equilibrium by ensuring that all threats are
crediblei.e., actions are optimal given the history of the game.
Although the interaction has few moves, it is complex to fully solve for due to the
lack of restrictions on quantities of violence that the actors can produce. I therefore
discuss the intuition behind each decision one step at a time. Moreover, I emphasize
the intuition for these choices; the appendix contains full proofs wherever applicable.
2.2.1

Group 2s Violence Decision

Group 2s violence decision is straightforward. At this stage, Group 1 has already

selected its level of violence, and Group 2 has already paid the cost of entry. It therefore
only needs to optimize its payoff for the contest. Group
q 2s objective function for the
v2
contest is v1 +v2 m2 v2 . Optimizing this yields v2 mv12 v1 .
Note that this optimal level of violence is decreasing in Group 2s marginal cost m2 .
This makes sensethe more expensive violence is for Group 2, the less Group 2 should
be inclined to commit violence. In addition, if Group 1s allocation is sufficiently large,
the Group 2s marginal utility for each unit of violence begins to decline. Intuitively,
producing 1 unit of violence produces a greater return for Group 2 when Group 1 has
produced 1 unit of violence than when Group 1 has produced 1 million units. In fact,
if Group 2s marginal cost and Group 1s level of violence are too high (i.e., m2 v1 > 1),
Group 2 produces v2 = 0.
6

It does, however, lead to more violence overall than compared to this model. This is because the
marginal cost of violence is effectively lower if more violence is more likely to lead to a favorable policy
change.

2.2.2

Group 2s Entry Decision

There are two cases to consider here. First, suppose that combination of Group 1s
violence and Group 2s marginal cost is sufficiently high. Then Group 2 will optimally
produce v2 = 0 and receive a payoff of 0 from the contest. If it enters under these
circumstances, its payoff is c. If it quits, it receives 0 instead. Group 2 therefore
quitsit makes no sense to pay fixed costs of entry and then not exert any effort.
Second, suppose that the combination of Group 1s violence and Group 2s marginal
cost is sufficiently low. Now Group 2 will optimally produce a positive amount of effort,
namely v2 . It may nevertheless not wish to play the contest if the cost of entry is too
great. Specifically, recall that Group 2s overall utility for entering and then choosing
v2

v2 equals v1 +v
m2 v2 c. Group 2 therefore enters if this amount is greater than 0, its
2
q
payoff for quitting. Setting up this inequality, substituting v2 = mv12 v1 and solving
for v1 yields:
v1 >

v1

(1 c)2

m2

Thus, Group 2 quits if v1 > v1 , enters if v1 < v1 , and is indifferent between the two
if v1 = v1 . Again, this makes sense: as Group 1 produces more violence, Group 2s
payoff for competition goes down. In turn, if Group 1s production is sufficiently high,
Group 2 prefers quitting to paying a cost to enter a competition that will not end well.
But if Group 1s production is sufficiently low, it is worth paying the cost of entry to
obtain the benefits from the contest.7 Note as well that Group 2 is less likely to enter
when the cost of entry or its marginal cost is high.
2.2.3

Group 1s Violence Decision: To Corner or to Compete

The above showed that Group 1 can control Group 2s entryproducing large amounts
of violence will corner the market, while choosing a smaller value of v1 will induce
competition. Although the appendix shows that the precise details of the strategy are
7

This critical value for v1 provides insight into situations with multiple potential
entrants. Suppose
(1 ci )2
those entrants have differing costs of entry ci and mi . Then the maximum of
is enough to
mi
corner the market. This is because if the group with the highest expected utility for competing is
unwilling to enter the market, all others would stay out as well. It is also true that violence spikes
when Group 1 produces that amount instead of the optimal quantity to compete this best opposing
group, and this spike drives all the key results of the extensions I develop below.

more involved, Group 1s decision-making process basically takes the following form:
1. Calculate the minimum v1 such that Group 2 would want to quit, which is v1 .8
(Choosing anything above the minimum is an unnecessary expense for Group 1,
as Group 2 is already staying out and Group 1 is taking all of the benefits.)
2. Calculate Group 1s payoff for cornering the market in the manner described
in (1).
3. Calculate the optimal value of v1 conditional on Group 2 entering, which is

m2
.
4m21

4. Calculate Group 1s payoff for choosing the strategy in (3).

5. Compare the payoffs from (2) and (4). If (2) is larger, play (1); if (4) is larger,
play (3).
Whether the payoff from (2) is larger than the payoff from (4) depends on c. Specifq

2
m22
m2

ically if c > c 1 m1 4m2 , Group 1 prefers to corner the market. This is

1
because higher costs of entry mean that Group 1 can produce less violence and still
deter Group 2 from entering. This makes cornering the market look more attractive,
causing Group 1 to be more likely to pursue that strategy.
Meanwhile, if c < c , cornering the market is too expensive to Group 1. In turn,
Group 1 produces the optimal amount of violence as if it were expecting Group 2 to
enter, and Group 2 indeed enters.

2.3

When Is Violence Most Prevalent?

Outbidding would seem to predict that violence peaks when multiple groups compete
for their audiences resources. The model shows this intuition is not guaranteed to
hold. There are only two outcomes equilibrium outcomes to compare: cornered and
2
competitive markets. When Group 1 corners, it produces v1 = (1m2c) to convince
Group 2 to not enter. This is the sum total of violence because Group 2 produces
m2
nothing when it stays out. In contrast, in a competitive market, Group 1 produces 4m
2
1
rm
and Group 2 responds with

2
4m2
1
m2

m2
.
4m21

Violence is greater in the first case if:

This assumes that Group 2 quits when indifferent between entering and quitting. The appendix
shows that equilibrium conditions require this.

Equilibrium Outcomes
Cost of Entry (c)

.4

Cornered Market

Competitive Market

0
0

Lead Groups Marginal Cost of Violence (m1 )

Figure 2: Substantive outcome of the games equilibrium, with m2 = 1. The lead

group chooses to corner the market when its marginal cost of violence is low and the
competitors cost of entering is high.

(1 c)
m2
+
>
m2
4m21
m1 >

m2
4m21

m2

m2
4m21

m2

2(1 c)2

As the extensions below visualize and the appendix proves, this can hold for certain
parameter values while still inducing the proper equilibrium outcome. Indeed, violence
is greater right as Group 1 prefers to corner than right as Group 1 prefers to compete.
Increasing c to sufficiently large levels eventually makes the equilibrium quantity of violence lower with the cornered market. This is because Group 1 can commit a relatively
low level of violence and still convince Group 2 to not enter.
Why does peak with only one group? A numerical example illustrates the logic.
Suppose that m1 = m2 = 1 and that Group 2 would assuredly
enter. Then Group 1
q
m2
v1
1
optimally selects 4m2 = 4 . Group 2 responds by selecting m2 v1 = 14 . Because both
1
select the same amount of violence for these parameters, they split the contest and each

receive 12 .9 After subtracting out their costs, each receives a final payoff of 14 , with a
sum violence of 12 .
Note that out of the whole value of 1, Group 1 only extracts a quarter of it. Consequently, Group 1 is willing to commit up to 34 in violence to push Group 2 out of the
market; this would give Group 1 the entire pie and leave it with more leftover than if
it had competed. Put differently, it is willing to use more violence by itself to corner
the market than the groups combined violence in a competitive market. And, indeed,
1
, then
this is exactly what happens for some parameter spaces. For example, if c = 25
2
(1 c)
16
Group 1 can produce v1 = m2 = 25 in violence to deter Group 2s entry. This leaves
16
9
Group 1 with a payoff of 1 25
= 25
, which is better than its payoff inducing entry. As
such, Group 1 corners the market by choosing a level of violence well above what both
groups would produce together when in competition.
These results help explain the inconsistent relationship between groups and violence
that empirical studies of outbidding have uncovered.10 Knowing how many groups exist
is insufficient to understand the understand the relationship between group numbers
and violence. A market with only a single group may outproduce (ceteris paribus) a
competitive market because that one group needs to deter entry of others. Potential
groups are equally important. But even knowing that information still can lead to
mixed results depending on the research design; although conditions exist where the
single group outproduces competitive markets, a single group may produce less than a
large competitive market if the barriers to entry are great.
The results also suggest a need to think more holistically about competition. Kydd
and Walter (2006, 78), for example, argue that one solution to the problem of outbidding would be to eliminate the struggle for power by encouraging competing groups to
consolidate into a unified opposition. The model demonstrates an important hidden
assumption underlying that claim: if consolidation creates new opportunities for other
groups to enter the market, unification may backfire. Indeed, these types of manipulations can have strange effects on the market for competitive violence. I therefore
formally extend the model below to allow for interventions.
9

The fact that both produce the same amount here is a quirk of these particular parameters and
does not generally occur.
10
Clauset et al 2010; Findley and Young 2012; Stanton 2013; Nemeth 2014; Jaeger et al 2015; Fortna
2015; Conrad and Greene 2015.

10

Manipulating the Market

If targets of violence attempt to manipulate competitive incentives, they have a variety

of options to choose from: (1) they may increase the barriers of entry for groups considering whether to form; (2) they can shrink the pool of potential support by reducing
grievances among the affected population; (3) they may increase the marginal cost of
violence for existing groups by taking offensive measures to gut those organizations;
and (4) they can increase defensive measures to mitigate violence from all producers.
To understand how these actions influence equilibrium play, I introduce a third
player to the game: the Target. The Target begins the new interactions by choosing
to pay a cost to shift these parameters in ways that would apparently handicap the
groups in some way. But the Target must be careful: the results below indicate that
these strategies can sometimes increase violence. Intuitively, this because the new
barriers can shift a competitive market into a cornered market, which is more violent.
Before beginning, a word of caution is appropriate. Although these categories are
useful for conceptualizing counterterrorism and counterinsurgency strategies, certain
tactics may affect multiple components simultaneously. For example, broad-scope domestic communications surveillance can increase the difficulty of both planning an attack and coordinating the formation of an organization. Readers therefore ought to
exercise caution in translating these results to specific policy recommendations, noting
that the actual affect might be a convex combination of these multiple extensions.

3.1

Increasing the Barriers to Entry

First, consider the cost of the competing group to enter, c. Targets can influence this
cost in a number of ways. One major component of the National Strategy for Combatting Terrorism is to deny organizations sanctuary in rogue states. As such rogue
states crumble, would-be groups must seek asylum in less desirable and more remote
locations, increasing the burden of the initial outlay to establish an organization. Meanwhile, the September 11 attacks created a push to create a norm against terrorism. This
process began with United Nations Security Council Resolution 1373, which instructs
countries to codify anti-terrorism laws and ratify anti-terrorism conventions. Political
violence entrepreneurs face increased hurdles in coordination the formation of a group
under such conditions. Improving economic conditions can raise the opportunity cost
11

of abandoning the civilian sector as well (Blomberg, Hess, and Weerapana 2004).
To incorporate these efforts by potential victims of violence into the model, suppose
that the Target begins the game by choosing c [c, ). The value c > 0 represents
the cost of entering without any intervention by the Target. Let the Targets payoff
be the negative of sum of the violence it suffers minus a function of how much effort
the Target expends to increase the cost of entry. The appendix fully solves the game
in which this specific utility function equals (v1 + v2 ) (c c), where measures
how much the Target values effort versus violence; smaller values of reflect a greater
fear of violence, as the Target finds the per-unit increase of changing the cost to be less
important. However, the key results follow as long as the effort function is convex.11
Broadly, the Target can pursue one of two strategies. Recall that the cost of entry c
partially determines whether Group 1 corners the market. If c is great, Group 1 deters
Group 2 from entering because doing so only requires a modest amount of violence; if c
is small, on the other hand, Group 1 prefers competing with Group 2 because cornering
q

2
m22
m2

requires too much violence. The value c = 1 m1 4m2 reflects this cutpoint.
1
Thus, as long as c < c , the Targets decision determines whether Group 1 will corner
or compete.12
The fact that the Target can turn a competitive market into a cornered market leads
to the following result:
Proposition 1. Increasing the fixed costs of entry (c) can lead to an increase in violence.
Figure 3 helps communicate the logic. When c < c , the parties commit a fixed
amount of violence in a competitive market. The precise value of c does not impact
the groups production choices because, conditional on entering the market, the cost c
is sunk for Group 2 and therefore does not affect its violence decision. Pushing past c ,
however, spikes the violence because the market shifts from competitive to cornered;
Group 1 now overproduces to shut out Group 2. Further increases c decrease equilibrium
violence on this range because Group 1 can produce less violence to convince Group 2
stay out as the cost of entry increases.
11

Thus, taking the negative of is concave, leaving a strictly concave utility function overall. The
specific loss functions for more effort from the target are generally unimportant in these extensions, so
I save most of the details for the appendix.
12
If c > c , then the cost of entry is already so expensive that Group 2 will not enter the market.
Increases to c will not change that.

12

Equilibrium Violence

0
0

Cost of Entry (c)

.2

Figure 3: Equilibrium violence as a function of the cost of entry c, with m1 = m2 = 1.

When the cost of entry is low, the competitor enters the market, and the groups produce
the same amount of violence regardless of that cost of entry. When the cost of entry is
high, the lead group produces enough violence to corner the market, which decreases
in that cost of entry. Increasing the cost of entry may lead to an increase in violence.

The shape of the violence levels has a number of important implications for counterterrorism and counterinsurgency. First, and most apparent, increasing barriers to
entry can backfire. Starting in a world where a competitive market would result and
shifting it cornered market can spike violence. To make matters worse, the Target
would also waste its resources to shift c in the process. Consequently, strategic Targets
never alter the cost of entry in this way.
Second, if a competitive market is the result of inaction and the Target chooses to
increase the cost of entry, it must create a cornered market. This is because a shift
from c to another value still below c does not alter equilibrium violence. Meaningful
change requires moving to a cornered market, with the caveat from above that this can
do more harm than good.
As the appendix details, these facts result in the Target playing a go big or go
home strategy in equilibrium. In particular, it calculates the optimal tradeoff between
its effort to change c and violence conditional on inducing a cornered market. If its
utility for that is less than than maintaining c, it chooses c. If it is greater, then it
picks that optimal c. But note from Figure 3 that this optimal c must be well above
13

c . That is because the equilibrium violence in the optimal cornered outcome must be
below the amount of violence from a competitive market. Due to the discontinuous
jump in violence at c , the Target must place c well above c to see any net decrease in
violence. In short, the Target can never choose a half measure.
Consequently, the Lucas (1976) critique urges caution when considering the empirical implications of the results for Propositions 1. Suppose that the full empirical record
reveals a monotonic decrease in violence as targets increase the barriers of entry for
groups. It would be tempting to then conclude that governments seeking to reduce
violence should increase the costs of entry. However, the Target never chooses a value
for c that increases equilibrium violence. This strategic decision therefore obscures the
overall effects of c in the empirical record. Blindly advocating for increased barriers
based on the data could in turn lead to an inadvertent jump in violence if the Target
picks one of the half measures described above.

3.2

Does Resolving Grievances Reduce Violence?

A softer approach to counterterrorism involves reducing grievances among those who

would otherwise wish to lend support to an organization. Isolationists in the United
States, for example, argue that abandoning foreign entanglements and deconstructing
military bases in the Middle East will result in fewer attacks from radical Islamicists.
France and Germany in particular cited fear of terrorism caused by increased grievances
as a reason to stay out of the Iraq War (Pauly 2013, 12-13; Dettke 2009, 157-158).
On the domestic font, separatist insurgents would seem less prone to commit violence
against their home governments if granted regional autonomy. Pakistan adopted this
strategy in a similar scenario against the Taliban, relinquishing control over parts of
their country to the group until American pressure in 2007 made the effort worth the
cost (Hoyt 2015; Rashid 2009, 385). Meanwhile, the National Strategy for Combatting
Terrorism argues that democratic governance can reduce grievances by giving citizens
the opportunity to address issues through their power to vote.
Nevertheless, the model demonstrates that the relationship between grievances and
violence is not straightforward. Consider the following extension. Previously, the value
of the prize equaled 1. Now, let the size of that pie equal [0, 1]. The Target
begins the game by selecting . One could conceptualize selecting a greater value

14

Equilibrium Violence

Size of Aggrieved Population ()

Figure 4: Equilibrium violence as a function of the percentage of supporters entering
the market, with m1 = m2 = 1 and c = .01. Larger pools of support turn a cornered
market into a competitive market, as the challenging entrant wishes to capture some
market share. Counterintuitively, this means that increasing grievances can cause a
decrease in overall violence.

of as demanding a larger extractive policy, broader international involvement, or

increased repression of a minority group. Greater demands lead to a larger aggrieved
population, either through direct injury or indirect economic distortions (Bueno de
Mesquita 2005a). Altering in this manner reflects that. Let the Targets utility
function equal (v1 + v2 ), where > 0 is a scalar that measures the Targets value
for the aggrieving policy versus violence.
Meanwhile,
the value of the contest for Group


vi
i given Group js violence is now vi +vj mi vi .
As before, the appendix contains a detailed explanation of the games equilibrium
behavior under the conditions of interest. However, the following proposition explains
the key finding:
Proposition 2. Decreasing grievances (i.e., decreasing ) can lead to an increase in
violence.
Figure 4 illustrates the central intuition. When the size of the pie is small, the
market cannot support multiple groups; there simply are not enough available resources
for Group 2 to justify entry. Even as the pie increases to where it could support both
15

groups, Group 1 can easily corner the market. Only after there is sufficient support
q
2

m22
2

) is Group 2 willing enough to enter that Group 1

(i.e., > c 1 m
m1
4m21
no longer wishes to overproduce to corner the market.
But this puts the Target in a familiar dilemma. If it reduces grievances among
the groups audience (that is, it moves from right to left on Figure 4), it switches a
competitive market to a cornered market. And because cornering the market near the
cutpoint requires Group 1 to produce more violence than the sum of their violence
in a competitive market, reducing grievances can backfire.
That said, outside the discontinuity, equilibrium violence increases within the segregated parameter spaces as grievances increase. Consequently, unlike with endogenous
costs of entry, the Target may wish to alter grievances in a manner that does not change
a cornered market to a competitive one or vice versa. However, if it wishes to change a
potentially competitive market to a cornered market, half measures will not work once
again. This is because the Target must place well below to drop the violence below
where it would otherwise be just to the right of .13
Indeed, there are many substantive examples of organizations increasing violence following concessions. Dugan and Chenoweth (2012), for example, find Israel experienced
upticks in violence First Intifada and Oslo Lull following small amounts of concessions.
And offering decentralization in ethnic conflicts has an inconsistent track record (Brancati 2006). Bueno de Mesquita (2005b) explains the counterintuitive relationship as the
result of moderates pulling out of the organizational hierarchy; the remaining extremists correspondingly push the agenda harder than before. Alternatively, the conciliator
may expect this as the cost of doing business, knowing that some individuals may wish
to spoil the peaceful path forward (Stedman 1997; Kydd and Walter 2002; Findley and
Young 2015).14
My mechanism is complementary to these and indicates that future empirical research ought to consider the counterfactualthis violence can be the result of what
would happen in its absence, making it difficult to identify its cause. A steady decline
13

In Figure 4, equilibrium violence goes to 0 as goes to 0. This is due to the functional form
that maps the Targets choice to the pool of support. One might imagine instead that some segment
of the population would still want to contribute even if = 0. Here, equilibrium violence would be
bound strictly above 0 as goes to 0. This does not affect the main result on the discontinuity at ,
however.
14
A reputation mechanism (Walter 2006) can also explain a spike in violence from outside parties.

16

in grievances ultimately reduces violence, but the benefits may not be immediately
apparent.

3.3

Does Whac-a-Mole Work?

Offensive measures are a third strategy to reduce political violence (Pillar 2001, 33-34).
Here, the target government actively pursues existing groups, attempting to reduce
their fighters, funds, and infrastructure. Many operations fit this category: demolition
of operative housing (Benmelech, Berrebi, and Klor 2015), bombing of camps and
convoys, assassinating leadership, attacking state sponsors, seizing financial assets held
abroad, and crackdown on black market commerce, like opiates in Afghanistan and oil
with ISIS. The target government may also massively broaden its intelligence netboth
domestic and abroadto assist with these tasks. Offensive measures rarely eradicate
entire organizations. However, they make continued violence more expensive, as the
group must alter the means of attack (Phillips 2015), operate with fewer agents, and a
draw from a smaller budget to accomplish its goals.
The Whac-A-Mole theory of counterterrorism suggests that these results could
backfire. These critics suggest that destroying one group is not helpful because another
group will arise to capture the marketshare. This might be especially concerning if a
strong organization replaces an otherwise enfeebled group. Nevertheless, equilibrium
results indicate that Whac-A-Mole is effective.
To investigate that claim, suppose that the target begins the game by choosing
m1 [m1 , ), where m1 > 0 represents Group 1s marginal cost of violence if the
Target takes no action. As always, the Targets payoff is the negative sum of the
violence it suffers minus a function of the effort the Target expends to increase Group
1s marginal cost. The appendix fully solves the game in which this specific utility
function equals (v1 + v2 ) (m1 m1 ), where measures how much the Target values
effort versus violence; smaller values of reflect a greater fear of violence, as the Target
finds the per-unit increase of changing the marginal cost to be less important. Like
before, the key results follow as long as the effort function is convex.
The Targets decision ultimately hinges on whether to choose a marginal cost above
or below a particular cutpoint. Recall that m1 partially determines whether Group 1
q
2

m22
2
corners the market. In terms of c, Group 1 cornered if c > 1 m

. Solving
m1
4m2
1

17

Equilibrium Violence

0
m1

Lead Groups Marginal Cost of Violence

Figure 5: Equilibrium violence as a function of the lead groups marginal cost of violence, with m2 = 1 and c = .05. Increasing the marginal cost switches the outcome
from a cornered to a competitive market. Violence consequently diminishes because
the lead group no longer overproduces to suppress the potential entrant.

m1

m2 +m2 2 cc 15

.
2(1 c)2

for m1 yields m1 <

Thus, if m1 is less than m1 , Group 1 corners
because the necessary overproduction of violence to keep out Group 2 appears relatively
cheap. But if m1 > m1 , cornering is too expensive, and thus Group 1 choose to compete
with Group 2.16
Because increasing the marginal cost cannot turn an otherwise competitive market
into a cornered one, the following proposition states that offensive measures are mostly
effective:
Proposition 3. Increasing the lead groups marginal cost of violence (m1 ) weakly decreases violence.
Figure 5 explains the relationship. Selecting an m1 < m1 induces Group 1 to corner
the market. This leads to a fixed level of violence because Group 1s violence choice
depends on Group 2s incentives (i.e., c and m2 ) and not Group 1s specific marginal
15

Solving for m1 requires using the quadratic formula, which generates two solutions. The requirements of this parameter space rule out the smaller of the two solutions, thereby generating a single
relevant cutpoint.
16
If m1 > m1 , then the Target can only select a value that leads to a competitive market.

18

cost. Furthermore, the level of violence is high because Group 1 must overproduce to
corner the market. Pushing past m1 leads to a discontinuous drop in violence because
the outcome switches to a competitive market. Equilibrium levels of violence decline
as m1 increases here because Group 1 wishes to produce less in a competitive contest
as its per unit cost increases.17
In turn, if m1 < m1 , the Targets decision works as follows.18 It begins by calculating
its overall payoff for maintaining a cornered market by selecting m1 . The Target will not
choose a value between m1 and m1 because doing so requires costly effort but maintains
an identical amount of violence as m1 .19 It then calculates the optimal tradeoff between
increasing m1 and the Targets own marginal cost of effort assuming that the market
will be competitive. If the optimal m1 in this calculation is greater than m1 , then
Group 1 compares its utility for selecting that to its utility for keeping m1 and chooses
the option that is greater.
If the optimal m1 in this calculation is less than m1 , the Target may still wish to
increase m1 to benefit from the discontinuous dropoff. It therefore compares its payoff
for m1 to that of m1 (assuming that the indifferent Group 1 pursues the competitive
outcome) and picks the strategy that produces the greater payoff.20
Consequently, increasing the lead groups marginal cost mostly has positive effects.
The lone issue is that exerting effort but choosing a value less than m1 (instead of
sticking with m1 ) has no net effect on violence. Thus, if the Target wishes to manipulate
Group 1s marginal cost of violence in an otherwise cornered market, it must shift m1 to
a value that produces a competitive market. Unlike the manipulation of c, though, the
Target might not want to go deep past m1 because altering the market cannot increase
violence. On the contrary, Group 1 might wish to go exactly to m1 to experience the
17

A careful reader will note that increasing

m1 !
can increase Group 2s production. However, overall
r m

violence production equals

m2
4m21

2
4m2
1

m2

m2
4m21

1
2m1 ,

which is decreasing in m1 .

If m1 > m1 , the the Target can only induce a competitive outcome. It therefore chooses the value
of m1 that optimizes the reduction in violence versus , its marginal cost effort.
19
Long 2014 provides empirical support for thishe finds that leadership targeting in Afghanistan
and Iraq had limited effects when leadership is well-institutionalized (i.e., when m1 is low) but reduces
violence when leadership is poorly institutionalized (i.e., when m1 is high). The nonlinear relationship
in Figure 5 matches this.
20
In fact, in any equilibrium in which the Target chooses m1 , Group 1 must produce the competitive
level of violence with probability 1. This is because the Target could profitably deviate to a slightly
larger value if Group 1 were to select the cornering level of violence with positive probability.
18

19

discontinuous dropoff and leave it at that.

What do the Whac-a-Mole critics miss? Their main concern is that destroying
one group merely allows another organization to arise to capture the marketshare.
The equilibrium results sympathize with thisincreasing m1 across the m1 threshold
indeed produces another group. But the introduction of another group is exactly what
the Target wishes to induceif no intervention would result a single group cornering
the market, the Targets manipulation must cause an additional group to form. This
pleases the Target, as it switches the game from the cornered market outcome with an
overproduction of violence to a competitive market with a comparatively smaller level
of violence.
Still, Whac-a-Mole correctly cautions against the efficacy of large-scale offensive
operations. Because another organization waits in the wings, there are decreasing
marginal returns to enfeebling the lead group.21 Although the Target may wish to
intervene in a competitive market, the competition nevertheless limits the effectiveness
of that intervention.
Offensive operations interact with previously described manipulations in interesting
ways as well. To begin, the critical value c in Figure 7 drifts to the right as m1
increases. Thus, enfeebling Group 1 means that the Target has to increase the cost
of entry at a higher level than before to causes a decline in violence. Second, I have
assumed that increasing m1 does not increase grievances. This might not be the case
for some operations, as the provocation literature highlights.22 The bright spot here
is that, per Proposition 2, the grievances facilitate switching an otherwise cornered
market into a competitive one, thereby compounding the effect.
The discontinuous dropoff in violence at m1 also has an important implication for
collective counterterrorism. Throughout, I have described the Target as a single entity.
Arce and Sandler (2005), however, note that transnational terrorism often strikes multiple entities. Further, they show that counterterrorism under these circumstances can
be a collective action problem. Individual incentives encourage states to bolster their
own defense. But this has negative externalities on other states, as those less-defended
countries are more tempting terrorist targets. This leaves all states worse off than if
1
Specifically, the sum violence in a competitive market is 2m
.
1
See See Fromkin 1975, 962-964; Price 1977; Crenshaw 1981, 387; Berry 1987, 8-10; Laqueur 1987;
Lake 2002; Bloom 2005, 107-110; Kydd and Walter 2006, 69-72; Carter 2016.
21

22

20

they offensively attacked the organization because proactive measures have positive externalities. Despite the temptation to free ride, a state may wish to choose contribute to
the proactive effort under the conditions of my model so that it may capture a portion
of the discontinuous drop in violence.23
On the other hand, the model also gives an alternative explanation for a lack of
counterterrorism. When the lead groups marginal cost of violence is very low, the
counterterrorist must exert a great deal of effort to reach the nonmonotonic dropoff in
violence. As such, rather than contributions failing due to a collective action problem,
moving the needle may simply be too expensive.

3.4

The Effectiveness of Defensive Measures

Finally, a target government can opt for defensive measures: hardening potential targets, stationing extra security at large gatherings, and increasing the scope of searches
at airports (Pillar 2001, 37-40). The best defense may not make the target invulnerable,
but it can force the organization to divert to other targets (Lum, Kennedy, and Sherley 2006), which may be less worrisome. Defense has an interesting property: it does
not discriminate. Thus, in contrast of offensive measures, reinforcing security directly
impacts the violence output of all organizations, formed or unformed.
To model the consequences of defense, suppose that the game begins with the Target
choosing [0, 1], where represents the portion of violence that is successful.24 Thus,
the original game is the special case where = 1. A value of = 23 means only that twothirds of the groups effort turns into punishing violence, with the defensive measures
absorbing the remaining third. The Target pays a cost that is strictly decreasing in
and strictly convex.25 This means that cutting violence by larger margins is more
costly, and the first measures to reduce violence are less costly than the next. From
there, the game proceeds as usual except that Group is contribution to the contest is
now vi instead of vi .
A surprising result follows immediately from the groups utility functions. For ex23

Of course, this does not give a clear indication of which state should exert that effort. Nevertheless,
efficient equilibria exist in volunteers dilemmas like this, which is not the case for standard prisoners
dilemmas.
24
Powell 2007 unpacks the blackbox of defense I describe here.
25
As before, the negative of this function is strictly concave, giving the Target a strictly concave
utility function overall.

21

Equilibrium Violence

=1
=

1
2

0
0

Cost of Entry (c)

.2

Figure 6: Equilibrium violence as a function of the lead groups cost of entry for two
different levels of defensive efforts (). Increasing defensive measures does not impact
the lead groups decision to corner the market, and violence levels are lower regardless
of that choice.

1
m1 v1 . The values immediately cancel. The
ample, Group 1s utility equals v1v+v
2
same is true for Group 2s utility function, meaning that has no impact on the remaining subgame. Group 2 still uses the same entry decision rule, and both groups choose
their violence levels exactly as before. This is because the defensive measures impact
both groups equally, and they choose their levels of violence where their marginal gain
equals their marginal cost. The violence strategies of the original game hit this point
for this extension precisely because has no affect on the contest.
In turn, the Target merely needs to optimize a straightforward tradeoff between
the sum of violence realized violence (v1 + v2 ) and its cost of effort. The following
proposition summarizes how the Target can think about the benefits of action:

Proposition 4. Increasing defensive measures (i.e., decreasing ) strictly decreases

equilibrium violence.
Figure 6 showcases the relationship, showing how equilibrium violence changes as
a function of c for two levels of . The solid line graphs the = 1 case, which is a
duplicate of plot in Figure 3. The dotted line tracks = 21 . The cutpoint for c remains
unchanged, and violence decreases by exactly half across the board. Generalizing form
22

this, decreasing further will only force equilibrium violence to decline. The figure
also shows that defense is most useful in just-barely cornered markets. Violence peaks
under those circumstances, making the marginal cost of defense look less onerous.
The monotonic reduction in violence here means that defensive measures have an
attractive quality that the others lack: non-strategic increases to defense always pay off
for the target. This ease of implementation may help explain the roughly half of U.S.
counterterrorism spending funds target hardening (Rosendorff and Sandler 2004, 658).
However, the takeaway here is not that governments should exclusively rely on defense.
Other strategies can work as well, and their marginal utility may be superior due to
beneficial nonmontonic declines in violence. Policymakers must exert more effort to
properly calibrate those strategies to ensure the benefits, though.

Conclusion

This paper explored the hidden role of potential entry into a market for political violence. Even when not facing immediate competition, hegemonic groups may still wish
to produce high levels of violence to convince other political violence entrepreneurs to
stay on the sidelines. Under such conditions, potential entrants observe that they cannot capture a large share of its audiences resources and therefore do not wish to pay
the costs necessary to create an organization; the hegemonic group ultimately benefits
by maintaining hegemony over the resources, though it may have to pay a high price
to dominate the marketplace.
Indeed, the model revealed that total violence is often greater when the second
group refuses to compete. This led to a number of unexpected policy implications.
Actions that make entering market look less appealingsuch as increasing the barriers
of entry and decreasing the pool of potential supportcan counterintuitively increase
violence by switching the equilibrium outcome from a competitive market to a cornered
market. Target governments must therefore greatly increase the barriers of entry and
greatly reduce grievances to receive any benefits. In contrast, other counterterrorism
measuresoffensive operations designed to destroy the infrastructure of existing groups
and defensive operations designed to reduce realized violence from all groupsdo not
backfire in this manner. This is because they do not cause the lead group to switch from
a competitive market to a cornered market, which in turn halts the jump in violence.
23

Overall, the model serves as a reminder that counterterrorism operations are a

part of a larger strategic environment. Failure to account for optimal responses to
group competition can lead to inaccurate empirical predictions and dangerous policy
implications. This paper only addressed a single manner of competitioni.e., deterring
market entry. The outbidding literature remains underdeveloped in thinking about
how non-group actors affect competitive behavior. Future research ought to address
whether similar unexpected relationships hold for alternative environments.

24

Notation
vi
mi
c

m1

g()

Description
Group is level of violence
Group is marginal cost of violence
Group 2s cost of entry
Targets marginal cost to increase c
Minimum possible value of c
Scalar measuring Targets value for policy versus violence
Level of grievances among the groups audience
Targets marginal cost to increase m1
Minimum possible value of m1
Portion of violence that Targets defense does not stop
Function mapping level of to cost of defensive measures
Table 1: Notation of the Many Extensions

Appendix

This appendix gives thorough proofs for the main game and the propositions for all the
extensions. It also solves for the Targets equilibrium actions in those extensions. For
clarity, Table 1 contains a glossary of the notation of the various models.

5.1

Proof of the Main Model

As in the paper, I break this down sequentially.

5.1.1

Group 2s Violence Decision

q
Lemma 1. Let v2 mv12 v1 . In all SPE, Group 2 selects max{0, v2 }.
Proof: Group 2s decision comes at the end of the game, so it optimizes its payoff
by choosing the value of v2 that maximizes:
v2
m2 v2
v1 + v2
.
The first order condition is:

25

v1 + v2 v2
m2 = 0
(v1 + v2 )2
m2 v22 + 2m2 v1 v2 + v1 (m2 v1 1) = 0
Applying the quadratic formula to obtain the roots yields:
2m2 v1

p
4m22 v12 4m2 v1 (m2 v1 1)
2m2

v1
v1

m2

Because v2 0, m12 v1 cannot be the solution because it is always negative.

This leaves m12 v1 , which itself may be negative if v1 >

2 selects the maximum of 0 and v2 .
5.1.2

1 26
.
m2

Consequently, Group

Group 2s Entry Decision

Now to the entry decision. I base the cutpoint on Group 1s chosen violence level v1
because Group 1 selects v1 in the next move that we need to solve for. As the following
lemma shows, that level of violence determines Group 2s entry decision:

Lemma 2. Let v1 (1m2c) . Group 2 enters if v1 < v1 , stays out if v1 > v1 , and is
indifferent between the two choices if v1 = v1 .
Proof: Consider four cases. First, suppose v1 m12 . By Lemma 1, Group 2 would
select v2 = 0 if it were to enter, giving it a payoff of 0 for the competition phase.
However, to reach that point, it would have to pay a cost of c. Because not entering
generates a payoff of 0, Group 2 must not enter.
Second, suppose v1 (v1 , m12 ). If Group 2 enters, it selects v2 = v2 . Working through
the contest success function, Group 2s payoff for the competition phase equals:
q
v1 +

v1
m2

v1
v1
m2

v1

1
26

v1
v1
m2

m2 v1

2
This is a maximum because the second derivative of the objective function is (v1 +v
3.
2)

26

Therefore, Group 2 stays out if:

m2 v1 c < 0

(1 c)2
v1 >
m2

This is true because the second requires this exact inequality.

2
Third, suppose v1 < v1 . By analogous argument, Group 2 enters if v1 < (1m2c) .
This holds here, so Group 2 enters.
Finally, suppose v1 = v1 . By analogous argument, Group 2 is indifferent between
entering and not entering, so it may mix freely between the two strategies.
5.1.3

Group 1s Violence Decision

q

2
m22
2
Proposition 5. Suppose m1 > m22 . Let c 1 m

. In the unique SPE,

2
m1
4m1
m2

Group 1 chooses v1 = v1 if c c and v1 = 4m2 if c < c . Group 2 quits in the first

1
case and enters and produces v2 in the second case.
Proof: I will first prove existence by assuming that Group 2 quits if indifferent
between quitting and entering. Broadly, Group 1 can select from two categories of
violence: an amount at least as great as the cutpoint c or an amount below. If Group
1
1 picks from the larger set, Group 2 quits. Group 1s payoff is therefore v1v+0
m1 v1 =
1 m1 v1 . Note that this amount is strictly decreasing in v1 that is, any extra violence
here serves no purpose to Group 1 but is costly. Consequently, no equilibria exist in
which Group 1 picks a v1 > v1 .
Now consider an amount
q from the smaller set. Group 2 enters, and (by Lemma
1) produces v2 = max{0, mv12 v1 } in violence. Using the contest success function,
q
provided that mv12 v1 > 0, Group 1s payoff equals:

v1 +

v
q1

v1
m2

v1

m1 v1

m2 v1 m1 v1
The first order condition of that objective function is:

27


m2
m1 = 0
2 v1
v1 =

m2
4m21
1

3
2

m2 v
2 41

This a maximum because the second derivative of the objective

function is
,
q
m2
v1
which is negative. Further, note that v1 = 4m2 implies that m2 v1 > 0 because the
1
proposition assumes m1 > m22 . This in turn means that Group 1 maximizes its payoff
with Group 2 still playing a violence strategy on the interior.
Now compare Group 1s utility for the minimum amount of violence necessary to
drive Group 2 out to Group 1s optimal competitive quantity. If Group 1 drives Group 2
out at the lowest price, it earns 1 m1 v1 . Meanwhile, choosing the optimal competitive
amount produces a payoff of

m2
4m2
1
m2
+v2
2
4m1

m2
m1 4m
2 =
1

m2
.
4m1

Thus, Group 1 induces entry if:

m2
(1 c)2
> 1 m1
4m1
m2
s
!2
m2
m22
c> 1

m1 4m21

This is the cutpoint given in the proposition.27

v1

m2
m1

m22
4m21

to corner the market if c < 1

q

2
The argument also shows that it is an equilibrium to corner the market if c = 1 m

m1
28
I will now prove uniqueness. Consider the same two cases as before, beginning
m2
with v1 4m
2 . The above proof showed that Group 1 cannot choose an amount other
By analogous argument, Group 1 chooses

Note that the value inside of the radical is positive if m1 > m42 . This must be true because the
parameter space has the more stringent requirement that m1 > m22 . Thus, although the causes of
terrorism are complex, the cutpoint is not. One might also be concerned that the optimal competitive
level of violence is greater than the minimum necessary to keep out Group 2. (This could be the
case if c is high.) However, the inequality still produces the correct result because the left hand side
represents a value that gives less than the whole prize and at a greater cost than the right hand side.
Thus, the inequality would tell us that Group 1 would pick the quantity to keep out Group 2.
28
The structure of these uniqueness proofs is similar to the uniqueness proof the ultimatum game.
The receiver (Group 2) is indifferent when the proposer (Group 1) selects a particular value. But
because rejecting (entering the market) leads to a dropoff in utility for the proposer (Group 1), and
because the proposer (Group 1) could deviate to an offer (level of violence) slightly greater to break
indifference, equilibrium constraints guarantee a unique equilibrium.
27

28

2

m22
4m21

.
2

than v1 in equilibrium. It remains to be seen whether equilibria exist in which Group

1 chooses v1 and Group 2 enters with positive probability. (This is rational for Group
2 because it is indifferent between entering and quitting when v1 = v1 .) So suppose
that Group 2 enters with probability (0, 1]. Group 1s utility for selecting v1 now
equals:
"

(1 c)2
m2

(1 c)2
+
m2


v2

Note that because

m1



#

(1 c)2
(1 c)2
+ (1 ) 1 m1
m2
m2

(1 c)2
m2

(1 c)2
+v2
m2

< 1, this is less than Group 1s utility when Group 2

quits with probability 1. Also note that Group 1s utility is continuous (and decreasing)
2
as it selects a level of violence greater than (1m2c) . Thus, there exists an  > 0 such
that
"

(1 c)2
m2

(1 c)2
+
m2



#

(1 c)2
(1 c)2
m1
+ (1 ) 1 m1
m2
m2
v2



(1 c)2

< 1 m1
m2


holds. This means that if Group 2 enters with positive probability, Group 1 can profm2
itably deviate. In turn, the equilibrium in Proposition 5 is unique when v1 4m
2.
1
m2
m2

Now consider the second case, in which v1 > 4m2 . If Group 1 selects 4m2 , the
1
1
existence proof showed that Group 2 has a unique best response to enter and produce v2 .
2
If Group 1 selects (1m2c) instead, Group 2 is indifferent between entering and quitting.
But if Group 2 enters with positive probability, the same uniqueness argument as above
2
appliesi.e., Group 1 has a profitable deviation to some v1 = (1m2c) + . Therefore,
the strategies presented in Proposition 5 are unique.
Proposition 6. Suppose m1
Group 2 quits.

m2
.
2

In the unique SPE, Group 1 chooses v1 = v1 and

Proof: By Lemma 2, if Group 1 selects v1 > v1 , Group 2 quits. No equilibria

exist for v1 > v1 because Group 1 can profitably deviating to a slightly smaller level of
violence still greater than v1 .
If Group 1 selects v1 (0, v1 ), Group 2 enters and produces v2 violence. Group 1s
29

1
utility equals v1v+v
m1 v1 . The first order condition from Proposition 1 demonstrates
2
m2
that this function is strictly increasing until v1 = 4m
2 . Note that for this propositions
1
m2
m2

parameter space (m1 2 ), 4m2 is greater than v1 . Thus, Group 1s utility is strictly
1
increasing on the interval v1 (0, v1 ). In turn, no equilibrium can exist on that range
because Group 1 could profitably deviate to a level of violence slightly greater while
still below v1 .
The only case left to check is when Group 1 produces exactly v1 . Group 2 is
indifferent between entering and quitting. The proof for Proposition 5 showed that
Group 1 receives strictly more when Group 2 quits under these circumstances. As
such, this is an equilibrium. Furthermore, the proof for Proposition 5 also showed that
Group 1 could profitably deviate to a slightly greater level of violence if Group 2 were
to enter with positive probability when indifferent. The aforementioned equilibrium is
therefore unique.

5.2

Endogenous Fixed Costs of Entry

Proposition 7. Let c (m211)2 . If m1 < m22 , the Target chooses max{c, c }. If

m1 > m22 , the following four cases describe the Targets unique equilibrium action:
1. If c > c and c > c , the Target chooses c.
2. If c > c and c < c , the Target chooses c .
3. If c < c and c > c , the Target chooses c.

)2

4. If c < c and c < c , the Target chooses c if (1mc2

and c if

2
(1mc2 )

(c c) > 2m1 1

(c c) < 2m1 1 .

Proof : Recall that the Targets objective function is (v1 + v2 ) c. From Propom2

sitions 5 and 6, Group 1 produces v1 = 4m

2 if c < c . Group 2 enters and produces v2 ,
1

generating a total of
in violence. If c c , Group 1 produces v1 = (1m2c) violence
to force out Group 2.
Note that increasing the cost of entry does not affect equilibrium levels of violence
if the chosen cost is low enough to fall in the competitive case. If the Target wishes to
push the cost of entry into the greater case, it must maximize:
1
2m1

30


(1 c)2
(c c)

m2
The first order condition is:


1
c

m2

c = c =

=0

1
(m2 1)2

Recall that Proposition 6 showed that there is no cost for which Group 2 enters
under its parameters. Thus, Group 2 does not enter regardless of the Targets decision.
In turn, the Target only has to worry about the above first order condition. It therefore
chooses the solution to the optimization problem c if it is greater than the minimum
cost c. If c < c, then the first order condition showed that any further cost increases
lead to a decrease in utility, so the Target chooses c.
Now to the cases. First, suppose c > c and c > c . This means that the minimum
cost of entry is greater than both the highest possible cost for a competitive equilibrium
and the Targets optimal cost in the cornering equilibrium. Since the first order condition shows that increasing c any further leads to a lower utility, the Target optimally
picks c.
Second, suppose c > c and c < c . The minimum cost of entry remains greater
that the highest possible cost for a competitive equilibrium but is less than the Targets optimal cost in the cornering equilibrium. The first order condition showed that
selecting c is optimal here.
Third, suppose c < c and c > c . The minimum cost of entry is less than
the highest possible cost for a competitive equilibrium but is now greater than the
Targets optimal cost in the cornering equilibrium. The optimal cost for maintaining
the competitive equilibrium remans c because anything greater needlessly exerts effort.
Its payoff for choosing that amount equals 2m1 1 . The optimal cost to shift into the
cornering equilibrium switches to c because the first order condition showed that the
Targets payoff
is decreasing going away from c . Its payoff for choosing that amount
2

equals (1m2c ) (c c). The Target therefore chooses c if

31


1
(1 c )2

>
(c c)
2m1
m2
I can show this holds by instead demonstrating

1
(1 c )2

>
2m1
m2
Substitution and substantial algebraic manipulation yields m1 > 2m1 1 , which holds.
So the Target chooses c.
Finally, suppose c < c and c < c . Now the minimum cost of entry is less than
the highest possible cost for a competitive equilibrium and is also less than the Targets
optimal cost in the cornering equilibrium. Thus, the Target can manipulate whether
Group 1 will induce Group 2 to enter. If the Target wishes to deter entry, the first
order condition
showed the optimal cost for doing so is c . This generates a payoff

2
of (1mc2 ) (c c). Alternatively, the Target can maintain the competitive
equilibrium; the optimal cost to do this is c because any additional cost lowers the
Targets payoff without manipulating the remainder of the game. The Target earns
2
2m1 1 for this. Thus, it prefers selecting c if (1mc2 ) (c c) > 2m1 1 and

)2

prefers selecting c if (1mc2

5.3

(c c) < 2m1 1 .

Endogenous Grievances

I begin by solving the two player interaction with Group 1 and Group 2. The proof
strategy follows the strategy of the original game, as it is a special case of this version
with = 1.
Group 2s Violence Decision. At the end of the game, if Group 2 has entered, its
objective function is:
v2
() m2 v2
v1 + v2
The first order condition is:
v1
() m2 = 0
(v1 + v2 )2
32

r
v2

,
m2

This is negative if v1 >

v1
v1
m2

so Group 2 chooses max{

v1
m2

v1 , 0}.29

Group 2s Entry Decision. If v1 m2 , Group 2 produces 0 for the contest and

thus receives 0 as its payoff for the contest. Since entering costs c, Group 2 will not
enter if v1 > m2 .
q
1
v1 . Routing this through its utility function
If v1 < m2 , Group 2 produces v
m2
q v
q

1 v
1
m2
v1
q
for the contest, entering produces
() m2
v1 at cost c. Quitting
v1
m2
v1 +

m2

v1

generates 0. Group 2 therefore enters if:

q
v1 +

v1
m2

v1

v1
m2

v1

r
() m2


v1
v1 c > 0
m2

c < ( m2 v1 )2

Analogously, Group 2 quits if c > ( m2 v1 )2 and is indifferent when c =

( m2 v1 )2 .
Group 1s Violence Decision. The minimum cost necessary to drive Group 2 is
2

c)
. Thus, Group 2 enters if Group 1
c = ( m2 v1 )2 .30 Solving for v1 yields (
2 2
2
m
( c)
( c) 31
selects v1 <
and stays out if v1
. In turn, Group 1s optimal level
m2
m2
( c)2
of violence to force Group 2 out equals
m2
If Group 1 wishes to induce Group 2s entry, its objective function is:

v1 +

v
q1

v1
m2

v1

() m1 v1

The first order condition is:

2
This is maximum because the second derivative is (v1 +v
3.
2)
30
I assume that Group 2 stays out when indifferent, but this must be true in equilibrium as before.
31
The out decision covers the case where v1 > m2 , which would otherwise result in Group 2 producing
0 violence.
29

33


m2
m1 = 0
2 v1
v1 =

m2
4m21

Selecting this value induces Group 2 to choose a strictly positive quantity of violence
2
c)
if m1 > m22 . If m1 < m22 instead, then Group 1 selects (
. (This is analogous to
m2
Proposition 6.)
2
c)
Now for the m1 > m22 case. If Group 1 drives out Group 2, it earns (m1 ) (
.
m2
If it produces the optimal competitive amount and Group 2 enters, Group 1s utility is
2
the above objective function with v1 = m
. It therefore forces out Group 2 if:
4m2
1

( c)2
(m1 )
>
m2

m2
4m21

r
m2
4m21

< 

m
() 22
4m1

m2

() m1

m2
4m21

m2
4m21

c
1

m2
m1

m22
4m21

2

By analogous argument, Group 1 induces entry if > and is indifferent when

= .32
Violence at the Cutpoint. As with the other extensions, the quantity of violence
near the cutpoint drives the Targets decision. As approaches from the left,
2
c)
in violence. As approaches from the right,
Group 1 corners and produces (
m2
s  
m2
4m21

m2
4m2
1

2
the parties compete; Group 1 produces
and Group 2 produces
m
.
m2
4m21

Substituting = , violence in the cornering case is greater than the sum of violence
in the competitive case if:

32

As with the baseline model, this covers the case where the optimal amount from the first order
condition exceeds the minimum necessary amount to force out Group 2; if this were the case, the
utility on the right hand side would produce less than from the contest and at a greater cost than
the left hand side, even though the left hand side gives the full to Group 1.

34

c
s
1

m2
m2
22
m1
4m1

!2

!2

m2
m2
2

m1
4m2
1

s
1

s
1

>

m2
v
u
u
u
u
u
u
u
u
u
u
u
u
t

4m21

c
v
u
um
m2
2
1t m2
1 4m2
1
4m21

!2 m2

m2
m2
22
m1
4m1

2
m2

c
s
1

m2

m2
m2
22
m1
4m1

!2 m2

4m21

In what is arguably one of the most amazing applications of algebraic manipulation

in human history, this monstrosity reduces to m1 > m22 . This must be true at the
cutpoint, otherwise Group 1 would have a strict preference to corner.
The Targets Grievance Decision. We are now ready for the main proposition of
this extension:
Proposition 8. Let =
unique equilibrium action:
1
2m1

1. If >
and

2. If >

1
2m1

4. If <

1
2m1

<

( c)2
.
m2

c)2
m2

and > , the Target chooses = 1.

1
and
2m1

2
( c)
and
m2

3. If <

The following four cases describe the Targets

and < , the Target chooses = 1 if 2m1 1 > (

if

1
2m1

c
.
(1m2 )2

< , the Target chooses = if

= if

2m1

<

( c)2
.
m2

2m1

>

and > , the Target chooses = .

Proof: The Targets utility function for a chosen depends on whether it is greater

than or less than . For values greater than , that function is 2m

. The first
1
derivative of this is:
35

1
2m1

Therefore, the Target maximizes its utility on this region at = 1 if > 2m1 1 and
at = if < 2m1 1 .33
2
c)
For values less than , the Targets utility function is (
. The first order
m2
condition of this is:
p
1 c

=0
m2
c
= =
(1 m2 )2
Therefore, the Target maximizes its utility on this region at if < and at
if > .
Now consider the four cases. If > 2m1 1 and < , the Target most prefers
selecting = 1 in the higher region and in the lower region. It therefore chooses
whichever producesthe greater expected utility. That is, the
Target chooses = 1 if

c)2
c)2

(
(
1
1

2m1 >
and

if

<

.
m2
2m1
m2
1

If > 2m1 and < , the Target still most prefers selecting = 1 in the
higher region but now prefers in the lower region. Notice, however, that the Targets
expected utility for for the cornered market is less than its expected utility at
in the competitive market. Furthermore, the Targets utility is strictly increasing in
above . This implies that the utility for = 1 must be greater than the utility for
in a cornered market. Therefore, the Target must select = 1.
If < 2m1 1 and < , the Target now most prefers selecting from the higher
region and from the lower region. For simplicity, assume that Group 1 chooses
a competitive market when indifferent between cornering and competing.34 Then the
Target chooses whichever of the two produces the
greater expected utility. That is, the

( c)2

Target chooses = if 2m
>

and = if 2m
<
m2
1
1

( m2 c) .
Finally, if <

1
2m1

and > , the Target still most prefers selecting from

33

There is no interior solution because the mapping of to violence is affine in this region.
As before, and for the same reasons, this assumption is actually required for equilibria to exist.
That is, if Group 1 cornered with positive probability when = , Group 1 could profitably deviate
to some slightly smaller amount.
34

36

the higher region and now also prefers selecting from the lower region. It therefore
must select . Following this, as an equilibrium condition, Group 1 must select a
competitive market.35

5.4

Endogenous Marginal Costs of Violence for the Lead Group

m2 +m2 2 cc

2(1 c)2

and m
Proposition 9. Let

1
describe the Targets unique equilibrium action:
m1

1 .
2

The following five cases

1. If m1 > m1 and m1 > m

1 , the Target chooses m1 .

2. If m1 > m1 and m1 < m

1 , the Target chooses m1 .

(1 c)
3. If m1 < m1 and m1 > m
> 2m1 (m1 m1 )
1 , the Target chooses m1 if m2

and m1 if (1m2c) < 2m1 (m1 m1 ).

1

4. If m1 < m1 , m1 < m
1 , and m1 < m1 , the Target chooses according to the rule
in (3).

m1 , the Target chooses m1 if (1m2c) <

5. If m1 < m1 , m1 < m
1 , and m1 >
(1 c)2

2m1 (m
> 2m1 (m
1 m1 ) and m1 if m2
1 m1 ).
1

Proof: To begin, we must restate Proposition 5 in terms of m1 . Recall from the

proof that the Group 1 prefers inducing entry if:

m2
(1 c)2
> 1 m1
4m1
m2
Analogously, Group 1 prefers cornering the market if the inequality is flipped and is
indifferent when those two values are equal.
We can rewrite this inequality as:
4(1

c)2 m21 4m2 m1 + m22 > 0

Applying the quadratic formula yields the following roots:

35

Again, this is because the Target could profitably deviate to a slightly greater amount otherwise.

37

m2

(4m2 )2 4[4(1 c)2 ](m22 )

2[4(1 c)]
p
m2 m2 2 c c

2(1 c)2

Because the coefficient on

lead term of the polynomial is negative, Group 1
the

m2 m2 2 cc

. However, recall that Proposition 5 only applies if

induces entry if m1 <
2(1 c)2
m1 > m22 . The smaller root is therefore irrelevant if:
p
m2 m2 2 c c
m2
2
<
2
2(1 c)
c<1
This is true. Thus, Group 1 corners the market up until m1 and induces competition
afterward. As such, the Targets choice is whether to keep m1 small and allow for
Group 1 to corner or force Group 1 to compete with Group 2 by raising m1 . In the
2
cornered market case, equilibrium violence equals (1m2c) . In the competitive market
case, equilibrium violence equals 2m1 1 . If in this second case, the Targets utility equals
2m1 1 (m1 m1 ). Thus, the first order condition is:
1
=0
2m21
1
m
1 =
2
Without constraints, this is the best marginal cost the Target can create if Group
1 induces competition.36
Now to the cases. First, suppose m1 > m1 and m1 > m
1 . This means that the
minimum marginal cost is greater than the highest possible cost for the cornered market
outcome and also higher than the Targets optimal marginal cost within the competitive
outcome. Consequently, the Target can only induce a competitive equilibrium, and any
additional marginal costs only push the Target further from its optimal m1 . In turn,
the Target chooses m1 .
Second, suppose m1 > m1 and m1 > m
1 . The minimum marginal cost is still
36

It is a maximum because the second order condition is m13 .

1

38

greater than the highest possible cost for the cornered market outcome but is now less
than the Targets optimal marginal cost within the competitive outcome. The first
order condition above showed that m
1 is the Targets optimal choice here.
Third, suppose m1 < m1 and m1 > m
1 . Now the minimum marginal cost is less
than the highest possible cost for the cornered market outcome but is greater than the
optimal marginal cost for a competitive outcome. Increasing m1 within the cornered
range cannot be optimal because it increases the Targets effort without changing the
total violence. Increasing past m1 cannot be optimal either because the Targets utility
for a competitive equilibrium is strictly decreasing past m
1 . Therefore, the possible
equilibrium amounts are m1 and m1 .
Although m1 is past the Targets optimal competitive market marginal cost, it may
still yet be optimal. This is because there is a discontinuous dropoff between violence
in the cornered market case and in the competitive case if:

(1 c)2
1
>
m2
2m1

c<2
This is true and is what guarantees that increasing m1 leads to a decrease in violence
even though it may switch the outcome from a cornered to a competitive market. Since
Group 1 is indifferent between the cornering the market and the competitive outcome
at m1 , assume that it opts for the competitive market with probability 1 in this case.37
In turn, the Target prefers keeping the marginal cost at m1 if

(1 c)2
1

> (m1 m1 )
m2
2m1
Analogously, the Target shifts the marginal cost to m1 if the inequality is reversed.

Fourth, suppose m1 < m1 , m1 < m

1 , and m1 < m1 . The minimum marginal
cost remains less than the highest possible cost for the cornered market, but it is now
less than the optimal marginal cost for a competitive outcome. Further, the optimal
marginal cost for a competitive outcome falls in the region where Group 1 would corner
the market. The Target again has no incentive to change the marginal cost within the
37

In fact, Group 1 must induce entry when indifferent in equilibrium. This is for the familiar reason:
if Group 1 were to choose the monopoly amount of violence, the Target could profitably deviate to
m1 . This same uniqueness logic prevails for the other cases.

39

cornered market region. If it were to push to the competitive outcome, it would not
want to increase the marginal cost past the cutpoint m1 because this is already beyond
the optimal competitive marginal cost. Thus, the optimization problem is identical to
the third case.

Finally, suppose m1 < m1 , m1 < m

1 , and m1 > m1 . This is the same as the
fourth, except now the optimal competitive marginal cost is greater than the cutpoint
that separates the outcomes. Thus, if the Target wished to increase the marginal cost
to induce a competitive market, it would select m
1 . Keeping it at m1 is optimal if

(1 c)2
1

< (m
1 m1 )
m2
2m1
Analogously, the Target shifts the marginal cost to m
1 if the inequality is reversed.

5.5

Endogenous Defense

Proposition 10. Let V = v1 +v2 , where v1 and v2 equal the equilibrium violence levels
given by Propositions 5 and 6. The following three cases describe the Targets unique
equilibrium action:
1. If g 0 () = V for some [0, 1], choose the that solves the equation. (The
solution is unique.)
2. If g 0 () > V for all [0, 1], choose = 1.
3. If g 0 () < V for all [0, 1], choose = 0.
If > 0, Groups 1 and 2 then play strategies according to Propositions 5 and 6,
which are not a function of . If = 0, the groups produce no violence.
The main paper showed that the decisions from Groups 1 and 2 are not a function
of . (The value of cancels out in the contest function.) The only exception is when
= 0, as this creates a divide by 0 issue. The utilities for each group depend on how
one defines the results of the contest when all effort equals 0, but this is inconsequential
for the Targets decision.

40

So consider the Targets objective function, (v1 + v2 ) g(), where g(1) = 0,

g 0 () > 0 for [0, 1], and g 00 () > 0 for [0, 1]. Substantively, this means that
defensive measures are not costly when none are taken ( = 1), less defense is cheaper
than more, and that the each defensive measure is more expensive to implement than
the last. Because v1 + v2 is not a function of (unless = 0), we can rewrite it as
V > 0. The first order condition is:
V g 0 () = 0
If g 0 () = V for some [0, 1], the Target picks this value. It must be unique
because g 0 () is strictly decreasing, and it is a maximum because the second derivative
is g 00 (), which is negative.
If g 0 () > V for all [0, 1], then the solution for the optimization problem lies
beyond = 1, and the Targets utility is strictly increasing on the interval. It therefore
chooses the maximum of that interval, = 1.
If g 0 () < V for all [0, 1], then the solution for the optimization problem lies
before = 0, and the Targets utility is strictly decreasing on the interval. It therefore
chooses the minimum of that interval, = 0.

Works Cited

Arce M, Daniel G. and Todd Sandler. 2005. Counterterrorism: A Game-Theoretic

Analysis. Journal of Conflict Resolution 49 (2): 183-200.
Benmelech, Efraim, Claude Berrebi, and Esteban F. Klor. 2015. Counter-SuicideTerrorism: Evidence from House Demolitions. Journal of Politics 77 (1): 27-43.
Berry, Nicholas O. 1987. Theories on the Efficacy of Terrorism. Journal of Conflict
Studies 7 (1): 7-20.
Blomberg, S. Brock, Gregory D. Hess, and Akila Weerapana. 2004. Economic Conditions and Terrorism. European Journal of Political Economy 20 (2): 463-478.
Bloom, Mia. 2005. Dying to Kill: The Allure of Suicide Terror. New York: Columbia
University Press.
Brancati, Dawn. 2006. Decentralization: Fueling the Fire or Dampening the Flames
of Ethnic Conflict and Secessionism? International Organization 60 (3): 651-685.
Bueno de Mesquita, Ethan. 2005a. The Quality of Terror. American Journal of
Political Science 49 (3): 515-530.
41

Bueno de Mesquita, Ethan. 2005b. Conciliation, Counterterrorism, and Patterns of

Terrorist Violence. International Organization 59 (1): 145-176.
Carter, David. 2016. Provocation and the Strategy of Terrorist and Guerilla Attacks.
International Organization, Forthcoming.
Clauset, Aaron, Lindsay Heger, Maxwell Young, and Kristian Skrede Gleditsch. 2010.
The Strategic Calculus of Terrorism: Substitution and Competition in the IsraelPalestine Conflict. Conflict and Cooperation 45 (1): 6-33.
Conrad, Justin and Kevin Greene. 2015. Competition, Differentiation, and the Severity of Terrorist Attacks. Journal of Politics 77 (2): 546-561.
Crenshaw, Martha. 1981. The Causes of Terrorism. Comparative Politics 13 (4):
379-399.
Cronin, Audrey Kurth. 2011. How Terrorism Ends: Understanding the Decline and
Demise of Terrorist Campaigns. Princeton: Princeton University Press.
Dettke, Dieter. 2009. Germany Says No: The Iraq War and the Future of German
Foreign and Security Policy. Baltimore: Johns Hopkins 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 Sociological Review 77 (4): 597-624.
Findley, Michael G. and Joseph K. Young. 2012. More Combatant Groups, More Terror?: Empirical Tests of an Outbidding Logic. Terrorism and Political Violence
24 (5): 706-721.
Findley, Michael G. and Joseph K. Young. 2015. Terrorism, Spoiling, and the Resolution of Civil Wars. The Journal of Politics 77 (4): 1115-1128.
Fortna, Virginia Page. 2015. Do Terrorists Win? Rebels Use of Terrorism and Civil
War Outcomes. International Organization, Forthcoming.
Fromkin, David. 1975. The Strategy of Terrorism. Foreign Affairs 53 (4): 683-698.
Hoyt, Timothy D. 2015. Pakistan: A Tale of Two Allies. In Coalition Challenges
in Afghanistan: The Politics of Alliance Eds. Gale Mattox and Stephen Grenier.
Stanford: Stanford University Press.
Jaeger, David A., Esteban F. Klor, Sami H. Miaari, and M. Daniele Paserman. 2015.
Can Militants Use Violence to Win Public Support? Evidence from the Second
Intifada. Journal of Conflict Resolution 59 (3): 528-549.
Kydd, Andrew H. and Barbara F. Walter. 2002. Sabotaging the Peace: The Politics
of Extremist Violence. International Organization 56 (2): 263-296.
Kydd, Andrew H. and Barbara F. Walter. 2006. The Strategies of Terrorism. International Security 31 (1): 49-80.
Lake, David A. 2002. Rational Extremism: Understanding Terrorism in the TwentyFirst Century. Dialogue IO 1 (1): 15-29.
Laqueur, Walter. 1987. The Age of Terrorism. Boston: Little, Brown.
Lucas, Jr., Robert E. 1976. Econometric Policy Evaluation: A Critique. In The
Phillips Curve and Labor Markets Eds. Karl Brunner and Allan H. Meltzer. North
42

Holland: Elsevier.
Lum, Cynthia, Leslie W. Kennedy, and Alison Sherley. 2006. Are Counter-Terrorism
Strategies Effective? The Results of the Campbell Systematic Review on CounterTerrorism Evaluation Research. Journal of Experimental Criminology 2 (4): 489516.
Moghadam, Assaf. 2008. The Globalization of Martyrdom: Al Qaeda, Salafi Jihad, and
the Diffusion of Suicide Attacks. Baltimore: Johns Hopkins University Press.
Nemeth, Stephen. 2014. The Effect of Competition on Terrorist Group Operations.
Journal of Conflict Resolution 58 (2): 336-362.
Netanyahu, Benjamin. 1995. Fighting Terrorism: How Democracies Can Defeat Domestic and International Terrorists. New York: Farrar Straus Giroux.
Pauly, Jr., Robert J. 2013. French Security Agenda in the Post-9/11 World. In Old
Europe, New Europe, and the U.S.: Renegotiating Transatlantic Security in the
Post 9/11 Era Eds. Tom Lansford and Blagovest Tashev. Burlington: Ashgate.
Phillips, Brian J. 2015. Deadlier in the U.S.? On Lone Wolves, Terrorist Groups, and
Attack Lethality. Terrorism and Political Violence, Forthcoming.
Pillar, Paul R. 2001. Terrorism and U.S. Foreign Policy. Washington, DC: Brookings.
Powell, Robert. 2007. Allocating Defensive Resources with Private Information about
Vulnerability. American Political Science Review 101 (4): 799-809.
Price, Jr., H. Edward. 1977. The Strategy and Tactics of Revolutionary Terrorism.
Comparative Studies in Society and History 19 (1): 52-66.
Rashid, Ahmed. 2009. Descent into Chaos: The U.S. and the Disaster in Pakistan,
Afghanistan, and Central Asia. New York: Penguin.
Rosendorff, B. Peter and Todd Sandler. 2004. Too Much of a Good Thing: The
Proactive Response Dilemma. Journal of Conflict Resolution 48 (5): 657-671.
Stanton, Jessica A. 2013. Terrorism in the Context of Civil War. Journal of Politics
75 (4): 1009-1022.
Stedman, Stephen John. 1997. Spoiler Problems in Peace Processes. International
Security 22 (2): 5-53.
Walter, Barbara F. 2006. Building Reputation: Why Governments Fight Some Separatists but not Others. American Journal of Political Science 50 (2): 313-330.
Wilkinson, Paul. 1993. Politics, Diplomacy, and Peace Processes: Pathways Out of
Terrorism. Terrorism and Political Violence 11 (4): 66-82.

43