Negotiating with Rebel Governments: The Effect of Service Provision on ConflictNegotiations
By
Lindsay HegerOne Earth Future [email protected]
Danielle F. JungPrinceton University
PRELIMINARY DRAFT: PLEASE DO NOT CIRCULATE OR CITE WITHOUTPERMISSION
Abstract
When rebels provide social services, do they have more leverage negotiating terms of a peacedeal? The literature suggests that service-providing groups may, on average, have a wider base ofsupport and a more centralized organizational structure. We argue that these features deterpotential spoilers from breaking away from the organization during negotiation processes. This,in turn, makes governments more willing to engage in negotiations since the threat from spoilersis smaller. Thus, service providing rebels are more likely to engage in stable negotiationprocesses compared to non-providers. This paper analyzes these propositions by gatheringservice provision data on nearly 400 terrorist andgroups and their involvement in and behaviorduring peace talks. It also serves as an introduction to a larger project about the implications ofrebel service provision on conflict outcomes.
A December 2012 Guardian report on a meeting between rebel leaders in Aleppo described the proceedings asfollows:
…First on the agenda was the task of reintroducing the men to each other, as many had switched battalionssince their last meeting in the endless game of musical chairs of the Syrian revolution.
A who's who of the revolution followed, each commander stating his name and his unit. Some battalionswere huge, with hundreds of men, artillery pieces and tanks. Others consisted of fewer than 50 fighters.
"Haji, I thought you were with Halab al-Shaba'a brigade," Haji Marea said to one of the men. "No, wehave reformed. We are a new battalion," the man said.
"Brothers, we have a grave situation ahead of us," interjected Abdul-Jabbar Akidi, a defected colonel wholeads the military council of Aleppo. Formed to channel supplies to the rebels, the council was supposed tobe the overarching command structure for the Free Syrian Army in Aleppo. Instead, it soon became onemore faction among many competing for influence…1
Introduction
There is no place that exemplifies better the strategic pitfalls of organizational
mismanagement than the rebel movement in Syria. The relative disorganization of the Syrian
opposition, particularly the non-Islamic groups,is proving fundamentally untenable. Competition
between groups, ineffective attacks, few defenses, and very little international assistance plagues
the opposition’s progress. We view these pitfalls as largely symptomatic of a managerial
problem:specifically, the rebels have none. And the implications of such disarray are stark when
considering the government’s strategic choice to negotiate a political end to the violence or
continue fighting. In a speech in January 2013 Bashar al-Assad stated, “We never rejected a
political solution…but with whom should we talk?” Despite the Assad’s central role in the
horrific conflict, his point is not without merit. With very little in the way of an organized
opposition, negotiating a political solution to the conflict is an unattractive option. In this paper
we explore why this is the case and test how relatively unorganized rebels fare when
governments consider negotiations as a possible solution to conflict.
1See http://www.guardian.co.uk/world/2012/dec/27/syrian-rebels-scramble-spoils-war, accessed January 2013
Here we ask two fundamental questions. First, why do rebels and governments meet at
the negotiation tableonly sometimes and, second,why are some negotiationsstable while others
fall prey to spoilers?2Hamas, the Irish Republican Army (IRA), the Palestinian Liberation
Organization (PLO), and the Free Ache Movement (GAM) have all engaged in negotiations and
often these have been remarkably stable interactions. In contrast al-Mansoorain (a Kashmiri
separatist group), the Jenin Martyr’s Brigade (a nationalist Palestinian group), and the Real Irish
Republican Army have never negotiated with officials of their respective adversaries. Is it simply
the case that groups in this latter set are more resolved or more capable of holding out? Or are
these groupsmore extreme and, therefore, less open to negotiations? Are there strategic reasons
why some adversaries never talk and only fight?
Weexamine this issue in the context of spoiling behavior and its relationship to a rebel
group’s provision of goods and services. We argue when rebel organizations provide services
they are less likely to be affected by one of the factors that often derails negotiations: spoilers.
Service provision bolsters a group’s organizational coherence by increasing support, legitimacy
and organizational capacity associated with service provision. Taken together, greater
organizational coherence can work to deter potential extremist factions within rebels’ ranks from
breaking away. Therefore, service-providing groups are more capable of credibly committing to
negotiation processes, and governments are more interested in negotiating with groups capable
of controlling potential extremists. Negotiations are, therefore, both more likely to occur and
more likely to be stable when they involve groups that provide non-violent services within their
communities.
This analysis makes two major contributions. First, we generate a novel measure of
service provision. To generate this measure we collect and analyze data on service provision by
2 We use the term rebels interchangeably with terrorist and insurgent throughout this paper.
almost 400 terrorist groups. Todate, similar data have only existed for a handful of non-
representative organizations. Our approach uses both a larger and less biased sample, and our
data provide a richer indicator of a group’s governance activities: both the extent and type of
service provision are captured. Secondly, our approach represents a departure from much of the
literature on negotiations, service provision, and conflict outcomes. Prior literature has focused
largely on the odds of a stable peace under negotiated settlement and how the international
community should identify and deal with potential spoilers (Greenhill 2006 and Stedman 1997).
Instead, we focus on a selection issue that occurs prior to negotiations, when parties evaluate the
relative utility of negotiations. If the odds of spoiling are high, negotiations may never occur.
With the exception of Kydd and Walter (2002), the conditions under which this occurs are
relatively unknown. This project explores how rebel groups’ activities shape both rebel and
government expectations about the likelihood of spoiling, and thus, whether negotiations are
possible.
This article is part of a larger effort todevelopan organizational assessment of conflict.
There is growing consensus that when rebels provide services they engage in
measurablydifferent forms of violence, organizational techniques, and are capable of building a
larger baseof support vis-à-vis the legitimacy gained as pseudo-governments. As a first step in
this larger project we have developed a methodology for systematically coding the service
provision activities of more than 400 rebel organizations. Current data on service provision
arelimited largely to anecdotal evidence, case studies, and small-N research focusing on groups
designated as terrorist organizations by a limited number of states in the international system.
The US, UK, Russia, the EU, the UN, Australia, Canada, and India are, to the best of our
knowledge, the suppliers of such lists. While data for groups on these lists offera promising
start—and we takemany of the lessons learned from these projectsusing these lists to heart—they
are problematic because of the bias inherent in limiting the population to several dozen high
profile groups on any one country’s watch list. Ourdata, covering hundreds of groups worldwide,
are the most inclusive to date. Additionally, our data are suitable for developing number of
indicators of goods provision, including confidence measures and a service provision profile.
Both are intellectually appealing in a wide array of contexts, particularly as they give us a sense
of how violent non-state actors engage in governance over populations they purport to fight on
the behalf of.
In this analysis, we look to apply our measures of service provision to the project’s
general claim that service-providing rebel organizations are unique. In this article we focus
ondifference manifested in the conflict negotiation processes. When rebels and terrorist
organizations provide services to the communities in which they operate, they deter potential
defections and generate social capitaland grassroots support. This capital increases the ability of
groups to negotiate without fracturing and, simultaneously, that governments see them as
credible at the negotiation table, making them morewilling to enter into talks. We believe this
credibility extends to make the likelihood of stable talks greater whenservice-providing groups
are negotiating rather than non-service-providing groups.
We develop this argument in five parts below. First, to motivate the project wereview the
literature on the effects of service provision as it relates to rebel behavior. In the second section
we develop our argument and presentcompeting hypotheses about conflict negotiations and
service-providing rebels. In section three we describe our measure of service provision and the
data we use to test the effect of service provision on negotiation (stability). In section four we
estimate the relationship between service provision and both negotiations and stable negotiations,
finding support for our hypotheses: service-providing groups are more likely to enter
negotiations, and negotiations with service-providing groups are morelikely to be stable. The
final section concludes.
I. Toward a Better Understanding of Service Provision
Recent literature examining rebel behavior and conflict processes highlights the role
social service provision plays in how conflictsunfold, organizations cohere, andthe odds of rebel
success. Here, we examine the range of service provision and the state of our collective
knowledge about its effects. We conclude this section discussing the literature we use tomotivate
our study as applied to spoiling and organizational cohesion duringnegotiations.
Service provision in conflict zones takes a range of forms. Some groups provide welfare,
food, medical services, education, orreligious services. Anecdotal evidence suggests Hezbollah
and Hamas both provide an extensive amount of welfare, education, and medical services to their
constituents, but they are not outliers. For example, the Revolutionary Armed Forces (FARC) in
Colombia supplymedical services, the Tamil Tigers in Sri Lanka maintained mail delivery
(among other services), and the Communist Party of the Philippines (New People’s Army)
supports literacy programs and performs marriage ceremonies for their supporters. The IRA
provided transportation services within Republican neighborhoods during the Troubles and even
smaller Loyalist groups delivered milk for new mothers living in their communities. Amongst
service-providing violent groups, Hezbollahis one of the most famous.
While Hezbollah’s military wing, Al Moqawama al Islamia (The Islamic Resistance) hasattracted much of world’s attention, the various other community activities of Hezbollahare of equal if not greater importance at home. It runs a range of philanthropic andcommercial activities including hospitals, medical centers, schools, orphanages,rehabilitation centers for the handicapped, supermarkets, gas stations, constructioncompanies, a radio station (Nur) and public service television station (Al Manar). Up
until the middle 1990s Hezbollah was also responsible for public services and utilities inthe southern suburbs of Beirut.3
Hamas’ service sector is similarly sophisticated, covering a rangeof charitable operations
and not-for-profit community resources. Only occasionally does Hamas’ leadership divulge
information about the organization’s community links, but one particularly informative
presentation suggested that the group participates in a wide array of activities. Speaking at a
meeting of Hamas leaders from North America held in Philadelphia in 1993, visiting Hamas
operative Muim Shabib offered a presentation on “the situation in Palestine” and the status of
“Islamic works” tied to Hamas. According to the FBI transcript, Shabib described the institutions
tied to Hamas as falling under the following classifications: educational (schools, universities),
social and charitable (refugees, orphans, relief), cultural, health institutions (clinics, medical
centers), public syndicates, technical institutions, sports clubs, media, religious institutions, and
women’s institutions.4The list of services provided by Shabib includes activities in both Gaza
and the West Bank. Organizational growth in intervening years has almost certainly led to an
increase in the variety of services provided, although to the best of our knowledge an equivalent
open-source assessment of the precise nature of Hamas’ current operations isnot available.5
Not all rebels have as extensivenon-violent wings dedicated to service provision. One
service many groups do provide is dispute adjudication and resolution services through
something akin to a local police force (Asalet al. 2010). These operations are less capital
intensive than Hezbollah’s hospitals or Hamas’ educational institutions, but they do provide
rebels with strategically important information about communities. Services are frequently
3Hezbollah Website: http://almashriq.hiof.no/lebanon/300/320/324/324.2/hizballah/. Accessed 1/23/20084Levitt 2006.Page 81.5Author communiqué’ with Matthew Levitt. For a more current list gathered from open-source data, see the SimonWeisenhall Center Snider Social Action Institute Report (2003) athttp://www.wiesenthal.com/atf/cf/%7BDFD2AAC1-2ADE-428A-9263-35234229D8D8%7D/hydraofterror.pdf.
provided in territorieswhere groups operate through unofficial community centers. If the group
has a political wing, these centers often serve a dual purpose as distribution centers and party
headquarters.
Early literature on rebel organizations suggested non-state groups provide goods and
services when state institutions are inadequate or nonexistent. In doing so, non-state actors gain
power, legitimacy, and influence over the communities they served (Tuijl 1999; Hasenfeld
1987). Only very recently has the literature concretely identified the specific causal mechanisms
linking service provision to power and identified and tested the implications of service provision
for alternative forms of rebel behavior (seeGrynkewich 2008 and Flanigan 2006). Workin this
areafocuses most extensively on the relationship between service provision and violence, noting
robust relationships between services and particular forms of violence. Specifically, service
provision is linked to suicide bombing, attacks on civilian targets, and highly lethal attacks
(Heger 2010; Berman 2009;Hegeret al.2012). Recent work also explores the relationship
between service provision and rebel organization. Flanigan (2008) looks at the link between
services andrebel recruitment while Berman and Laitin (2008) find that services support “club”
forms of organization in which rebels distribute goods to counter problems associated with
defection. Berman et al. (2011) find counterinsurgent-provided services increase the inclination
of noncombatants to share information with authorities about rebels and Mampilly (2011)
examines the link between rebel service provision and international legitimacy.
Several points of empirical consensusabout service-providing rebel organizationsare
apparent. First, service-providing groups likely have a greater base of support. Service provision,
particularly when substituting for poor governance on the part of the state, makes recruitment
easier, reduces information shared with counterinsurgents, and increases perceptions of
legitimacy. All of these outcomes strongly suggest that service providers likely have a stronger
base of civilian support. Second, service provision implies a rebel organization is functionally
differentiated. Functionally differentiated groupsshould be more adept at fighting (Hegeret al.
2012)and at negotiating largely because they have clear lines of command and control, implying
a significant degree of control from the top of the organization to the operatives working on the
ground. Alternatively, these cohesive groups can be characterized as having fewer veto players,
reducing the duration of conflict (Cunningham 2006). Finally, consistent with the points above,
services are associated with greater severity of violence and more extreme tactics.The literature
linking suicide terrorism and civilian targeting to services rests largely on the argument that these
forms of violence are only possible when groups are internally strong and have a significant
degree of support from the population.
II. Spoiling, Services, and Negotiations
Together, the lessons paint a picture of service providing rebel organizations as
hierarchical, in which leadership retains a high degree of operational command and control. The
organization is relatively immune to defection problems while also enjoying support from a
relatively large swath of the population and is capable of conducting highly lethal violent
campaigns. In this section we outline our argument: that the conditions above suggest service
providers are less likely to suffer from spoiler problems during negotiations. The literature on
spoiling has focused on categorizing spoilers and offered recommendations about how best to
avert losses associated with spoiling. We argue services are associated with greater
organizational coherence and cohesion, and this is the critical link between services and
negotiations.
There are several types of spoilers (Stedman 1997). Our argument focuses on spoilers
that emerge during the negotiation process when the negotiating rebel group fractures and
extremists attempt to derail the process. These spoilers are a type of “inside spoiler,”
reflectingtheir position at the bargaining table. Spoilers distinct from the negotiating parties,
“outside spoilers,” exist when actors feel wrongly excluded from the peace process or if outside
entities have a stake in continued fighting. Our argument does not address explicitly outside
spoilers, although we suspect that when a group delivers services, resultingsupport from vast
sections of the community may also deter some outside spoiling.6 Breakaway dissent groups
have plagued negotiations and ceasefires in many prominent conflicts including those involving
Euskadi Ta Askatasuna (ETA) in Spain and various Palestinian nationalist groups.
The Decision to Talk and Stability of Negotiations
In early work on the subject, Stedman (1997) argues that spoilers emerge when actors,
“believe that peace emerging from negotiations threatens their power, worldview, and interests,
and use violence to undermine attempts to achieve it”(5).Stable negotiations are unaffected by
spoiling violence. Successful stable negotiations result in peace. Unstable negotiations frequently
break down because violence ruins trust between principals, affecting the ability of the parties to
negotiate an enforceable agreement (Kydd and Walter 2002).
Stedman and others have explored the ways in which third parties can serve as custodians
of the peace, ensuring that spoiler activity is effectively deterred through costly signaling, local
legitimacy-enhancing measures, and coherent strategy (Stedman 1997; Walter 2001; and Barnett
and Zürcher 2009; Khatib nd). Some focus on the efficacy of negotiations and strategies of those
6We see little reason to suspect that service delivery will encourage outside spoiling, thus, mitigating any empiricaleffect in our analysis. Theoretically, however, our principal focus is on internal spoilers that emerge when extremistsbreakaway from an organization.
involved. Downes (2004), for instance, examines the reasons why negotiations are less likely to
work in certain contexts, especially in the wake of ethnic civil wars. Beardsley, Quinn, Biswas,
and Wilkenfeld (2006) examine how different mediation styles affect the bargaining
environment and conflict settlement outcomes. Buhaug, Gates and Lujala (2009) suggest that
rebel groups on par with their adversaries may be more likely to engage in negotiations with a
government because the prospects of either side winning are small.
When parties enter negotiations they evaluate the relative costs. Zahar (2003) explains
potential spoilers consider the costs associated with returning to fighting and the loss of peace
dividends when assessing whether to act. We argue that a similar calculus occurs prior to talks
when adversaries are considering the negotiation option. Negotiating can be a costly option,
especially if talks fail. Failed negotiations can leave parties less likely to trust each other in the
future. Failed negotiations also transmit information about the relative resolve and constraints
that parties’ face, information parties would rather keep close to the vest and have only revealed
because they miscalculated the odds of a peaceful resolution.
Thus, when parties consider whether to enter negotiations they evaluate the probability
spoilers will hamper negotiations. How do parties evaluate this possibility? Governments
concerned about spoilers assess the group’s coherence, internal structure, community appeal, and
leadership’s commitment to negotiations and/or peace. To the extent that a group maintains
control over its ranks and control dissent, governments are less likely to be worried about
potential rebel spoilers. Newman and Richmond (2006), in analyzing the role of foreign
mediators, suggest a similar logic that emphasizes the importance of insider loyalty to successful
negotiation processes. They write that “[i]n order to secure the sustainability of peace, custodians
must not only ‘neutralize’ the environment in such a way as to prevent actors from accessing
resources that could reignite war, they must also steer peace in such a way as to consolidate
insider loyalty to the process.”7
We argue organizations that service providershave a larger array of tools to contain
potential spoilers. Work by Heger (2010) and Berman (2009) suggestsservices can be used to
generate a stronger organization, less vulnerable to defections and with a more stable cadre of
community supporters. By providing services to its operatives and community supporters, the
group can threaten potential spoilers from within the organization or the community whose
tactics may undermine the group with loss of access to benefits. This deters those who pursue
such actions. There are observable implications of this argument. Foremost is that governments
ought to be less likely to enter negotiations with groups they perceive as susceptible to
fracturing. Negotiations in such a situation will be more likely to be costly, and less likely to end
the conflict successfully. Conversely, governments should be more likely to negotiate with
service providers who are able to control their organization. This leads to our first hypothesis:
H1: Governments are more likely to enter negotiations with service providing groups.
In effect this means service providers are more likely to be involved in negotiation
processes than non-providers. We acknowledge that service providers may be more likely to be
involved in negotiations for any number of reasons. These groups may be substantially larger
than non-providers, they may be much more established groups, operating for longer, they may
be the primary representative of a larger movement, or they may be the only option for a
government seeking to negotiate with a legitimate authority of the rebellion and view service
provision to be an indicator of legitimacy.8 Our argument, although a newtheoretical
contribution, is observationally equivalent to these. Thus, identifying a unique implication of our
7 Newman and Richmond, 2006, page 10.8 Lake (2010) details the links between legitimacy and service delivery in his discussion of contemporary phases ofcounterinsurgency practice.
argument is important to substantiate our claims. To this end we focus a large portion of our
analysis on the implications of our argument on the stability of negotiations, once started.
Specifically, our argument implies negotiations with service providers, ceteris paribus, ought to
be more stable than those with non-providers. We see service provision, as opposed to group size
or legitimacy, as having a more direct effect on negotiation stability. Thus we will test the
following second hypothesis to assess the robustness of our theory.
H2: Negotiations involving rebel groups that provide goods and services are more likely to be
stable than negotiations that involve non-providers.
A potential counterargument is that service provision has the opposite effect on both
negotiation possibility and stability. Service provision may exacerbatetensions and lead to
factionalization. These problems result from inadequate distribution issues. If service distribution
favors one subsetof the organization’s membership over another, or is directed toward one sub-
population within the community, spoilers may be encouraged to derail negotiations as a
demonstration of their strength within the organization.
Service provision may also create conditions that favor unstable negotiations as a
consequence of the associated capital burdens placed on the organization to maintain service
delivery in the wake of fighting. Rebel groups (and governments) involved in fighting may use
negotiations as a way to gain information about the resolve of their opponent, buy time to rearm
or reassess strategic plans. Service-providing rebels may enter negotiations in order to create the
peaceful conditions necessary to build non-violent capital and networks. Negotiations to buy
time are inherently unstable as at least one party at the table is not dedicated to finding a
sustainable solution. To the extent that service providers have incentives to create periods of
peace to sustain their non-violent activities, negotiations with service providers are likely to fall
apart. When this occurs, governments are less likely to engage in negotiations with service
providers and, if they meet at the negotiating table, talks are less likely to be stable. The two
counter-hypotheses to our argument are:
H3: Governments are less likely to enter negotiations with service providers
H4: Negotiations involving rebel groups that provide goods and services are more likely to beunstable than negotiations that involve non-providers.
III. Assessing Service Provision and Negotiations
In this section, we assess the relationship between service provision and negotiations. As
previously mentioned, this article is part of a lager project examining the empirical differences
between service providers and non-providers in both the context of conflict and post-conflict
outcomes. To date the project has been principally concerned with data challenges, most
specifically identifying and collecting systematic measures of service provision by terrorist
groups. Here, we outline our strategy to collect this data and, in doing so, identify and describe
the major independent variable used for this analysis.
Identifying and Measuring Service Provision
Binary indicators of service provision are available for a variety of groups. Heger 2010
describes a dataset generated for approximately 50 groups designated by the US as Foreign
Terrorist Organizations. The MAROB dataset provides an ordinal measure of service provision
for groups in the Middle East and North Africa (MENA) from 1980-2004. To capture service
provision, we gathered data designed to capture the presence, type, and extent of rebel-provided
services. Starting with the universe of over 400 terrorist groups identified by Cronin (2011), we
search all related news reports associated with each organization for indicators of service
provision activities.9To create our measure of services, we first identified and collected all news
items for all available dates (coverage is complete from 1980 on) available on Lexis-Nexis
relating to each group. We captured any news report mentioning the specified group’s name,
erring on the side of over-collection.10 The range of coverage available for groups varied widely.
The group with the least coverage returned only 235 words of news coverage text while the
group with the most coverage returned 3,061,421words. The average group returned nearly
380,000 words in related stories. We will refine this subset of documents in the future and
expend the document pool for each group searching for group acronyms, aliases, and variations
on spelling.
We generated a list of service provision terms (see Table 1A in the appendix) through an
extensive examination of anecdotal evidence, similar small-N studies, and expert interviews. We
then compared each group’s news reports to our list of service provision terms. Following this
comparison, we generated frequencies of words and phrases (up to two-word phrases) related to
service provision in each group’s news coverage using the text analysis program Wordscores.11
Our fundamental underlying assumption is that news coverage for groups that provide services is
more likely to mention words associated with service provision than those groups who provide
fewer services, or do not provide them at all.
The raw, coded reports allow us to create several measures of service provision.
9 We use Cronin’s universe of cases because it represents one of the largest empirical investigations of rebel groupsto date. She identifies the set of groups that do not exclusively target military or property and show sustainedorganizational capabilities. To search for the group, we used the precise spelling and wording identified by Cronin,and required those words needed to be contained in a single phrase. Future collections will account for the variety ofaliases used by many groups.10 If it is likely to bias our measure of service provision, it will do so by biasing us against service provision if themeasure is constructed as a proportion of the total coverage. In raw counts, this strategy will not have a significanteffect.11Wordscores is text analysis software available at http://www.tcd.ie/Political_Science/wordscores/software.html.
First, we have a raw count of all words and phrases related to service provision (total
service mentions).
Second, we compare this count to the size of the coverage for that group, creating an
indicator of the relative frequency of service provision within the scope of its activities
(total coverage).
Third, we combine search terms into broad categories of service provision: Education,
Youth and Recreation; Health and Emergency; Security and Justice; Financial, Jobs,
Welfare, Subsidies; Natural Disaster; Public Services; Religious (the detailed sub-
categories are listed in Table 1A in the Appendix). This allows us to understand both the
larger trends of service provision across terrorist groups—something that we have been
unable to test to this point—and to address questions about how service provision will
impact the success of a group and its performance in negotiations using service profiles.
Table 1 reports summary statistics for these service provision measures. Services falling
under the public services category are most prevalent in the dataset, followed by
mentions of financial and educational services. Least frequent are mentions about
services related to natural disasters and healthcare.
We foresee several potential problems with our measures. First, news coverage of
terrorist groups is likely to be non-random. Older groups, for instance, likely have more
coverage. Groups operating in democracies where press reporting is relatively free are also likely
to have more coverage. To the best of our abilities, we attempted to include controls in our
models whenever we perceive coverage to be unbalanced trough an assessment of total coverage
quartiles.
Second, what is released and reported in the subset of newspapers available is likely to
under-report service provision. This may significantly diminish our population of service
providers identified in the universe. This will make it more difficult for us to find the relationship
between service provision and negotiations that we expect. This issue would be problematic if
the bias were in the other direction.
Third, we need to be very careful about assessing context of any service mentions to
ensure they are services attributed to the group, rather than the state, or even a mention of the
group not providing that particular service. To understand better the extent to which our
methodology is biasedby incorrect or ambiguous attribution of services to actors other than the
terrorist group, weconducted an attribution analysis in which coders scrutinized a significant
segment of the reports for a randomly selected subset of groups to assess any attribution biases
that may exist.12Coders overwhelmingly found that mentions of service provision were
attributable to the terrorist group, giving us confidence the measure is capturing the concept we
want to measure.
Normalizing the Service Measure
The distribution of mentions of service words is heavily weighted to very few mentions.
Frequency distributions of all mentions of service words across groups (left panel) and the public
service words (chosen for the sake of example) sub-category (right panel) are shown in Figure 1.
To address this problem we think about measuring service by any given group in two ways:
groups’ weighted service provision created by dividing the total number of service words for any
given group by the total number of service mentions across all groups (weighted service) and the
group’s rank based on service provision measured against all other groups (rank service). In both
cases, the higher the number, the more likely the group is providing services. We created similar
12We randomly selected 10 percent of groups for a backcheck.
measures for each sub-category. The weighted service measure is more likely to be influenced by
outliers (although the analysis in the subsequent section shows the results for both are similar),
thus, we prefer the ranked service indicator. For the sake of transparency, we show the results
from both indicators. Table 2A in the appendix shows the descriptive statistics across service
sub-categories for the weighted service measure.
Looking at the relative service profiles of groups helps to illustrate the variety in service
profiles these data are able to capture. In Figure 2 we display the relative service provision of
four groups, chosen to demonstrate service distribution for amongst large, medium, small, and
very small providers based on our ranking of their overall service provision. The first, Hezbollah,
is amongst the top overall service providers. Consistent with most descriptions of this group,
Hezbollah is heavily engaged in public services, disaster relief (probably mostly from the post
2006 Israeli-Lebanon war), and welfare. The IRA and Baloch Liberation Army are both second-
tier providers, while the May 15 Organization for the Liberation of Palestine is amongst the
lesser providers (note the change in the scale of the Y-axis). Reports for both the IRA and the
May 15 Organization emphasize security services, while the Baloch Liberation Army appears to
focus more on financial welfare.
IV. Analysis
Measuring Negotiation and Negotiation Stability
We draw measures of negotiation occurrence and stability from Cronin’s data on terrorist
organizations worldwide.13 To test Hypothesis 1, that governments are more likely to enter
negotiations with service providers, we use her dichotomous measure “Talks.” “Talks” is a
dichotomous measure indicating whether a group has had any negotiations with a government.
Negotiations count as occurring whether they are stable, succeed, or fail. Simply put, this
13See http://howterrorismends.com/ for more information and access to the data.
indicator captures any attempts at negotiations. To measure Hypothesis 2, thatnegotiations
involving rebel groups that provide goods and services are more likely to be stable than
negotiations that involve non-providers, we use Cronin’s measure “talksstable” indicating
whether the talks were either stable/conflict-ending or were unstable and failed. For both talks
and talksstable, the unit of observation is the terrorist group and not an individual instance of
negotiation (see our discussion about future project refinements below for more on this).
Controls
We included a host of control variables, all drawn from the Terrorist Organization
Profiles (TOPS) database hosted by the University of Maryland’s National Consortium for the
Study of Terrorism and Responses to Terrorism.14 Some are motivatedby the literature on
terrorism and negotiations and others are the result of our balance tests discussed above. The
controls can generally be divided into two categories: group-specific and country-specific. There
are five group-specific controls. First, based on others’ research, religious groups tend to be
significantly different in both lethality and tactical strategies utilized (Asal and Rethemeyer 2008
and Berman 2009). For reasons related to their extreme tactics, governments may be less inclined
to negotiate with religious groups, thus, we control for this group ideology. Second, we control
for an organization’s age on the assumption that both service coverage and a group’s strength
will positively correlate with age.15 As conflicts persist and the rebels prove themselves strong
enough to maintain a threat across a long period of time, negotiations may be more attractive
solutions in the context of hurting stalemates, something widely acknowledged as a pivotal
moment for negotiations (see, for instance, Sisk 2009). Third, we control for the number of a
group’s allies, factions, and enemies in group’s home country. We suspect this is appropriate for
14 See http://www.start.umd.edu/start/data_collections/tops/ for data and a more detailed description.15 This is the only group-level control variable that comes from Cronin’s 2011 dataset. We utilized Cronin’s “year”variable here, which is an estimate of the age of the group in years.
two reasons. First, to the extent that more groups may create disincentives for a government to
enter negotiations, this is an important control. Second, if there is a highly competitive
environment resulting from the presence of multiple rebel groups, insurgent outbidding may
occur. In this case groups may see negotiations as a signal of weakness and decline any and all
attempts at talks. We also added controls indicating whether there was evidence that the terrorist
group had a political party and whether it had bases in multiple countries. Political participation
may make groups likely to negotiate and governments likely to approach them to negotiate. We
code this as a binary indicator and use TOPS’ text description of the group as the data source.
Additionally, we add a binary control for multiple bases because it may proxy for groups that are
more likely to fracture due to multiple veto points.
We view country specific controls as necessary indicators of environmental conditions
affecting both rebel and government likelihood of engaging in negotiations. Because our unit of
observation is a terrorist group, the country-specific variables require some explanation. The
majority of terrorist attacks are domestic in nature, meaning that both the group’s home base and
where it uses violence are located in the same state. As a result all country-level variables
reference the group’s home base. For most groups, we expect this estimation to be accurate.
However, there are two exceptions. First, some groups (e.g. al-Qaeda Central) focus exclusively
on international terrorism. In these cases, the country-level indicators will not reflect
environmental conditions related to negotiations between adversaries. However, we expect that
this is largely the exception and without attack data that differentiate between domestic and
international attackers we are unable to empirically control for these anomalies. We remain
confident in our approach and don’t see any reason this should bias our findings. The second set
of groups for which country-controls may not be entirely appropriate are groups that have bases
in multiple countries. When TOPS reports multiple bases for a country, we assume the first base
listed is primary and is used to generate all controls. However, a group may split itself in ways
that make reporting on country characteristics where all bases are located as necessary to explain
a group’s behavior. For instance, Palestinian groups such as the Palestinian Islamic Jihad have
moved around the Middle East utilizing various countries as bases until they are pressured to
leave. While these groups are ultimately geared toward changing Israeli policy, the pressures
they face to move between bases most certainly influences their strength, abilities, and probably
willingness to negotiate. Outside of controlling for the presence of multiple bases, the country-
level controls we include do little to account for these pressures. We see few reasons to suspect
bias from this issue, although we acknowledge its presence.
We include three country controls. First, we include a variable for any given country’s
Composite Index of National Capabilities (CINC) score (logged and averaged over the time
period 1980-2001). We expect that the more capable a country, the less likely it will engage in
negotiations with a terrorist group. We also include controls for the countries log average total
population (1980-2001) and regime type. Because this is not a temporal dataset, controlling for a
country’s regime type is problematic. We decided to use a binary democracy variable indicating
whether the average POLITY2 (Marshall, Jaggers, and Gurr 2011) score from all years 1980-
2010 was above a 6. We also attempted to include controls indicating major regime changes or
any democratic regimes during the time period. None of these significantly impacted our results
and for the sake of parsimony we only report the findings from the average POLITY2 variable.
Table 2 reports the descriptive statistics for all controls.
Analysis
We estimatethe relationships using logisticregressions. We begin with an analysis of the
relationship between negations occurrence (“talks”) and service provision. Table 3 reports the
results. Model 1 reports the simple relationship between our weighted service provision indicator
and the occurrence of talks. Model 2 incorporates group-level controls and Model 3 incorporates
country-level controls. Model 4 includes country fixed-effects (omitting the previous country
controls). Models 5-8 are similarly specified using the ranked service provision indicator rather
than the weighted measure.
The results indicate a strong, positive relationship between service provision and the
probability of negotiations taking place (hypothesis 1). The models in Table 3 above show that
the general relationship between high levels of service provision of any kind and entering into
talks is robust to controlling for the group’s age, type, political participation, and strategic
environment. Moreover, the effects are substantial. For instance, estimating first differences
using model 6 specifications (see Tomz, Wittenberg, and King 2003 for information about the
CLARIFY program used in this estimation), there is nearly a 10% increase in the odds that
negotiations will occur between the 100th and 300th ranked groups. (Table 3A in the appendix
lists all the predicted probabilities for this estimation.)
Table 3 shows a number of other interesting results. Religious groups are significantly
less likelyto be engaged in negotiations. Above we theorized that may be due to their use of more
extreme tactics. These findings lend support to our suspicions. Also consistent with our
preliminary thoughts, older groups are more likely to be engaged in negotiations, as are groups
with political parties. Interestingly, neither the number of related groups nor democratic
opponents has a significant impact on the likelihood of negotiations. When a group is based in a
country with higher capabilities (i.e. higher CINC scores) negotiations are much less likely. This
may reflect a government’s ability to withstand the costs imposed through internecine violence
or its ability to achieve a military defeat (or the belief that this is likely and the consequent
unwillingness to negotiate).
We now turn to estimating the relationship between service provision and stable
negotiations (“talksstable”), Table 4 lists our logit results. The model specifications are similar to
Table 3. The results from these models strongly suggest that service providers are uniquely able
to engage in stable negotiations, confirming hypothesis 2. The substantive effect, again using
CLARIFY to estimate predicted probabilities, show negotiations are approximately 6% less
likely to be stable for lower-tiered organizations (100th rank versus 300th rank). The results are
consistent for both the weighted and ranked indicators of service provision.
There are a few major differences between the results for talks and stable talks. The odds
of stable talks are relatively unaffected by groups’ ages. In addition, democracy becomes
marginally significant, suggesting that when negotiations involve a democratic government they
are more likely to be stable.
Service Subcategories
Thus far we have focused on aggregate measures of service provision. We now turn our
attention to a discussion of service subcategories. Recall the subcategories of services include:
Education, Youth and Recreation; Health and Emergency; Security and Justice; Financial, Jobs,
Welfare, Subsidies; Natural Disaster; Public Services; Religious Services. Tables 5 and 6 below
show our findings, for negotiation occurrence (“talks”) and negotiation stability (“talksstable”)
respectively. Using the specifications from model 4 (8, 12, and 16), we estimated logit models
using the ranked service provision indicator for groups across each service category.
For both the likelihood of negotiations occurring (Table 5) and the stability of
negotiations (Table 6), three major service sectors appear to be most impactful: educational
services, disaster relief, and public services. There are a number of ways to interpret these
findings. In all cases, distribution is likely key. If these sectors are those more influential at
deterring spoiling attacks, which is what the data suggests, they may also be the sectors
associated with goods that are more easily distributed in a targeted manner. To the extent that the
group can control access to education, disaster relief, and public services, it may be able to more
easily target potential spoilers in order to deter them. Alternatively, or perhaps in addition to,
these sectors may have the most influence on people’s lives and are consequently the most
effective at maintaining group loyalty. Services under these sectors may also be the most easily
branded, allowing groups to take credit for their delivery more easily. At present we have neither
the space nor data to discern between these effects, however it is an issue that merits greater
understanding especially for counterinsurgents creating policies to counter violent service
providers.
IV. Discussion and Future Steps
The findings strongly support hypotheses 1 and 2, indicating that service providers are
both more likely to be involved in negotiations and those negotiations are likely to be stable. We
find no evidence to suggest that either hypothesis 3 or 4 is accurate. We are intrigued by the
strong positive findings in the specific service categories as well. Our analysis suggests that the
impact of all services is not equal. Educational services appear to have a different effect on
negotiations compared to either security or religious services. Theoretically these differences are
entirely underdeveloped. These issues merit further investigation, particularly if our attribution
analysis does not reveal any significantly different biases between service categories.
Several prominent features of this analysis are currently lacking. The first and most
glaring omission here is the lack of certain controls present in the analysis. To this end, there are
two major omissions. First is the lack of a control for the presence or willingness of a third party
to credibly guarantee any commitments made at the negotiation table. This omission is
something we hope to address in future iterations when we gather the data necessary to do a
cross-temporal analysis. To do so we intend to re-code service provision for groups incorporating
the date of the news report so we can generate a yearly measure of service provision.Temporal
data will allow us to account for specific periods of negotiations (we have already gathered data
on negotiations dates for all relevant groups in this sample) and the presence of third parties at
the negotiating table. We consider the work by Fortna (2004) and Walter (2001) as particularly
informative in this regard.
The next major data omission is a set of attack-specific group controls. In this regard, we
are provoked by our findings for religious groups. The literature suggests these groups are
largely more violent and use more extreme tactics yet our present analysis does not allow us to
truly assess the independent effects of religious identity and violent tactics. Thus, we view this as
a major addition to the future iterations of this project.
Finally, we plan to run the same analysis focusing on the issue of rebel success at
soliciting concessions from the government. If rebels do in fact gain more legitimacy for their
cause and more support from their communities by providing public goods and services,
governments may be more likely to capitulate to rebel demands. Insofar as a government’s key
allies consider the rebel group a dissenting voice in politics, service provision may serve to
increase the legitimacy of rebel claims. It can also grow the group’s base of support. As a
consequence of increased legitimacy or support base size (or both), international pressure may
encourage a government to concede to rebel demands. Internal pressure may be even more
effective. If service provision does increase rebel support, capitulation may become more likely
as the number of rebel supporters grows. On the other hand, there are reasons to suspect that
service providers may in fact be less likely to achieve their demands. Evidence suggests that
rebels are likely to use more horrific forms of violence as service providers. These include the
use of suicide attacks and attacks that intentionally target civilians with lethal force. While these
groups may be more coherent (and less likely to be plagued by spoilers), there may be a
significant downside related to their violent methods. As Abrahams (2006) argues, groups that
employ the most extreme violent methods are likely to be perceived as having maximalist
demands (despite the true relative nature of their demands). Populations and politicians are
unlikely to give in when they perceive their opponent in this manner. Thus, the correlation
between service delivery and extreme violence also implies that service providers are unlikely to
achieve their strategic goals. We plan on conducting this analysis in the near future using, again,
Cronin (2011) data on rebel success.
This paper demonstrates the power of service provision as influential on a state’s
interaction with violent non-state actors, and prospects for peace. When non-state actors such as
rebel groups or terrorists act like pseudo-governments by providing services to their constituents
they are much more likely to be involved in stable negotiation processes. Our results suggest that
service provision is in fact strongly tied to hierarchy and a wide base of support; these factors
discourage potential extremist factions from breaking away during negotiations. For
policymakers our findings suggestcounterinsurgency strategy should takethe structure and type
of violent threat into account. Non-state actors that actively provide governance may be much
more attractive as partners to a political settlement than non-state actors that use violence and do
little to provide for their communities. However, this also suggests a note of caution.
Governments engaging service providing rebels may have a more formidable opponent on their
hands. This analysis and argument suggests that these groups are less amenable to divide and
conquer tactics. This, coupled with other research confirming that service providing rebels are
more lethal (Heger et al. 2012) and less prone to defection (Berman 2009), suggests that
governments may have to give in to these rebel demands in order to avoid a costly violent
conflict.
Works Cited
Abrams, M. (2006) Why Terrorism Does Not Work.International Security. 31(2): 42-78.
Asal, V., Berman, E., Heger, L., & Laitin, D. (2010) The Institute on Global Conflict andCooperation at the University of California: Political Violence Expert Survey. Data availableupon request to Dr. Heger.
Asal, V., Rethemeyer, R. (2008) The Nature of the Beast: Organizational Structures and theLethality of Terrorist Attacks. Jouranl of Politics. 70 (2): 437-449
Barnett, M. & Zürcher, C. (2009) The Peacebuilder’s contract: How external statebuildingreinforces weak statehood. In The Dilemmas of Statebuilding: Confronting the contradictions ofpostwar peace operations Roland Paris and Timothy D Sisk eds. Routledge.
Beardsley, K. C., Quinn, D. M., Biswas, B., &Wilkenfeld, J. (2006) Mediation Style and CrisisOutcomes. Journal of Conflict Resolution. 50(1): 58-86.
Berman, E. & Laitin, D. (2008) Religion, terrorism and public goods: Testing the club model.Journal of Public Economics. 92(10-11): 1942-1967.
Berman, E. (2009) Radical, Religious and Violent: the New Economics of Terrorism. Cambridge,MA: MIT Press.
Berman, E., Shapiro, J. N., & Felter, J.H. (2011) Can Hearts and Minds be Bought? TheEconomics of Counterinsurgency in Iraq. Journal of Political Economy.119 (4): 766-819
Buhaug, H., Gates, S., & Lujala, P. (2009) Geography, Rebel Capability, and the Duration ofCivil Conflict. Journal of Conflict Resolution. 53 (4): 544-569
Cronin, A. K. (2011) How Terrorism Ends: Understanding the Decline and Demise of TerroristCampaigns. Princeton, NJ: Princeton University Press.
Cunningham, D. (2006) Veto players and Civil War Duration. American Journal of PoliticalScience. 50 (4): 875-892.
Cunningham, D., Gleditsch, K.S., & Salehyan, I. (2009) It Takes Two: A Dyadic Analysis ofCivil War Duration and Outcome. Journal of Conflict Resolution. 5 (4): 570-597.
Downes, A. B. (2004) The Problem with Negotiated Settlements to Ethnic Civil Wars. SecurityStudies. 13 (4): 230-279
Flanigan, S.T. (2006). Charity as Resistance: Connections between Charity, Contentious Politics,and Terror. Studies in Conflict & Terrorism. 29 (7): 641-655
Flanigan, S. T. (2008). Nonprofit Service Provision by Insurgent Organizations: The Cases ofHizballah and the Tamil Tigers. Studies in Conflict and Terrorism. 31(6): 499-519
Fortna, V. P. (2004) Peace Time: Cease-Fire Agreements and the Durability of Peace. Princeton,NJ: Princeton University Press.
Gould, E. D. & Klor, E.F. (2010) Does Terrorism Work? Quarterly Journal of Economics125(4): 1459-1510.
Greenhill, Kelly M. (2006) The Perils of Profiling: Civil War Spoilers and the Collapse of theIntrastate Peace Accords. International Security. 31 (3): 7-40
Grynkewich, Alexus G. (2008) Welfare as Warfare: How Violent Non-State Groups Use SocialServices to Attack the State. Studies in Conflict & Terrorism. 31 (4): 350-370
Hasenfeld, Y. (1987) Power in Social Work Practice. Social Service Review. 61(3): 469–483
Heger, L. (2010) In the Crosshairs: Explaining Violence Against Civilians. Doctoral dissertation,University of California, San Diego.
Heger, L., Jung, D., and Wong, W. (2012) Organizing for Success: How Structure AffectsTransnational Campaigns. Terrorism and Political Violence.
Hezbollah Website. (Accessed 2008). Hizbullah-The Party of Godhttp://almashriq.hiof.no/lebanon/300/320/324/324.2/hizballah/
Lake, D. (2010) The Practice and Theory of US Statebuilding. Journal of Intervention andStatebuilding. 4 (3): 257-284
Levitt, M. (2006) Hamas: Politics, Charity, and Terrorism in the Service of Jihad. WashingtonDC: The Washington Institute for Near East Policy.
Khatib, S. (nd) Spoiler Management During Algeria’s Civil War: Explaining the Peace.(Accessed 10/24/2012) http://www.stanford.edu/group/sjir/6.1.06_khatib.html
Kydd, A. & Walter, B. (2002) Sabotaging the Peace: The Politics of Extremist Violence.International Organization. 56 (2): 263-296
Mampilly, Z. C. (2011) Rebel Rulers: Insurgent Governance and Civilian Life During War.Ithaca, NY: Cornell University Press.
Marshall, M., Jaggers, K., and Gurr, T. (2011) “Polity IV Project Political RegimeCharacteristics and Transitions 1800-2010: Dataset User’s Manual” Center for Systemic Peacehttp://www.systemicpeace.org/polity/polity4.htm
Newman, E & Richmond, O. (2006) Obstacles to Peace Processes: Understanding Spoiling, inChallenges to Peacebuilding: Managing Spoilers During Conflict Resolution. United NationsUniversity Press
Simon Weisenhall Center Snider Social Action Institute Report. (2003) athttp://www.wiesenthal.com/atf/cf/%7BDFD2AAC1-2ADE-428A-9263-35234229D8D8%7D/hydraofterror.pdf (Accessed at various times throughout 2012)
Sisk, T. (2009) International Mediation in Civil War: Bargaining with Bullets. London:Routledge
Stedman, S.J. (1997) Spoiler Problems in Peace Processes.International Security. 22 (2): 5-53
Tomz, M., Wittenberg, J., & King, G. (2003) Clarify: Software for Interpreting and PresentingStatistical Results. Available at http://www.stanford.edu/~tomz/software/clarify.pdf.
Van Tuijl, P.(1999) NGOs and human rights: Sources of justice and democracy. Journal ofInternational Affairs. 52 (2): 493-504
Walter, B. (2001). Committing to Peace: The Successful Settlement of Civil Wars. Princeton, NJ:Princeton University Press.
Zahar, M. (2003) Reframing the Spoiler Debate in Peace Processes, in ContemporaryPeacemaking: Conflict, Violence and Peace Processes. New York, NY: Palgrave Macmillan
TABLES AND FIGURES
Figure 1: Distribution of service mentions (total and public services) across groups
Variable Observations Mean Std. Dev Min MaxEducation,Youth,Recreation
396 272 581 0 6942
Health,Emergency
396 158 338 0 3640
Security,Justice
396 257 511 0 4159
Financial,Jobs, Welfare,Subsidies
396 238 393 0 2384
NaturalDisaster
396 147 425 0 3677
PublicServices
396 1072 1463 0 7535
Religious 396 169 369 0 3868Table 1. Summary statistics of Service Mentions across categories
Variable Observations Mean Std. Dev Min Max
Talks 392 0.19 0.36 0 1
StableTalks
392 0.10 0.31 0 1
ReligiousGroup
390 0.31 0.46 0 1
Group Age 388 13.78 14.25 1 140
NumberOtherRebels
390 1.45 2.65 0 34
PoliticalParty
392 0.09 0.29 0 1
MultipleBases
394 0.26 0.44 0 1
Democracy 383 0.47 0.50 0 1
TotalPopulation
383 10.42 1.62 6.35 13.70
CINCScore
383 -5.16 1.57 -9.27 -1.96
Table 2: Descriptive Statistics for negotiations and controls
Variables Model 1(std.error)
Model2
Model3
Model4
Model5
Model6
Model7
Model8
WeightedService
Provision
3.70***(0.85)
3.07***(0.99)
4.52***(0.93)
4.48**(01.73)
RankedService
Provision
2.01***(0.48)
1.37**(0.56)
2.31***(0.51)
1.48*(0.88)
ReligiousGroup
-1.26***(0.38)
-1.81***(0.60)
-1.17***(0.37)
-1.61***(0.57)
Group Age 0.03***(0.01)
0.04***(0.01)
0.02***(0.01)
0.03**(0.01)
NumberOther
Rebels
0.02(0.05)
0.006(0.07)
0.02(0.04)
0.03(0.06)
PoliticalParty
1.29***(0.38)
1.70**(0.07)
1.34***(0.38)
1.87***(0.70)
MultipleBases
-0.01(0.32)
-0.96(0.62)
-0.03(0.31)
-0.95(0.61)
Democracy 0.41(0.31)
0.44(0.31)
TotalPopulation
0.95***(0.21)
0.91***(0.21)
CINC -1.11***(0.24)
-1.07***(0.23)
CountryFixed
Effects
No No No Yes No No No Yes
Constant -1.92(0.18)
-2.21(0.25)
-18.10(3.60)
-25.46(1.86)
-2.54(0.31)
-2.56(0.34)
-18.13(3.57)
-25.08(1.85)
Log-likelihood
-183.04 -158.61 -166.11 82.61 -183.25 -160.09 -167.14 -85.06
N 392 381 381 213 392 381 381 213
Pseudo X² 0.05 0.16 0.12 0.36 0.05 0.15 0.11 0.34
Table 3: Testing Negotiations Occurrence (“talks) using logit models. ***significant at 0.01, **significant at0.05, *significant at 0.10
Variables Model 9(std.error)
Model10
Model11
Model12
Model13(std.error)
Model14
Model15
Model16
WeightedService
Provision
3.27***(0.98)
3.03**(1.19)
4.20***(1.08)
5.09**(2.23)
RankedService
Provision
1.76***(0.60)
1.28*(0.70)
2.09***(0.63)
1.86(1.18)
ReligiousGroup
-1.33***(0.50)
-1.40*(0.82)
-1.23**(0.48)
-1.35*(0.79)
Group Age 0.01(0.01)
0.02(0.01)
0.01(0.01)
0.01(0.01)
NumberOther
Rebels
-0.02(0.07)
-0.10(0.13)
-0.003(0.06)
-0.43(0.12)
PoliticalParty
1.47***(0.41)
2.08**(0.84)
1.52***(0.41)
2.23***(0.83)
MultipleBases
-0.01(0.39)
-2.35**(1.03)
-0.04(0.39)
-2.49**(1.03)
Democracy 0.69*(0.39)
0.68*(0.39)
TotalPopulation
1.15***(0.29)
1.09***(0.29)
CINC -1.44***(0.33)
-1.38***(0.32)
CountryFixed
Effects
No No No Yes No No No Yes
Constant -2.55(0.23)
-2.68(0.31)
-22.80(5.00)
-22.35(1.60)
-3.08(0.40)
-3.00(0.42)
-22.40(4.93)
-22.24(2.37)
Log-likelihood
-130.16 -115.52 -114.46 -52.39 -130.94 -116.76 -116.06 -54.27
N 392 381 381 168 392 381 381 168
Pseudo X² 0.04 0.13 0.13 0.38 0.03 0.13 0.12 0.35
Table 4: Testing Negotiation Stability using logit models. ***significant at 0.01, **significant at 0.05,*significant at 0.10
36
Variable Model 17(Educate)
Model 18(Health)
Model 19(Security)
Model 20(Finance)
Model 21(Disaster)
Model 22(Pub Serv)
Model 23(Religion)
RankedService
Provision
2.69**(0.94)
0.15(0.80)
0.15(0.80)
0.56(0.88)
2.49***(0.89)
1.73**(0.87)
0.73(0.82)
GroupControls
Included?
Yes Yes Yes Yes Yes Yes Yes
CountryFixed
EffectsIncluded ?
Yes Yes Yes Yes Yes Yes Yes
Constant -25.50(1.88)
-24.81(2.72)
-24.812.72)
-24.80(1.75)
-24.87(3.01)
-25.12(1.89)
-24.76(1.82)
Log-likelihood
-82.06 -86.49 -86.49 -86.30 -82.35 -84.43 -86.10
N 213 213 213 213 213 213 213Pseudo X² 0.37 0.33 0.33 0.33 0.36 0.35 0.34
Table 5: Testing Negotiation Occurrence (“talks”) using logit models. ***significant at 0.01, **significant at0.05, *significant at 0.10
Variable Model 24(Educate)
Model 25(Health)
Model 26(Security)
Model 27(Finance)
Model 28(Disaster)
Model 29(Pub Serv)
Model 30(Religion)
RankedService
Provision
2.04**(1.19)
0.31(1.13)
0.31(1.13)
0.46(1.19)
2.15*(1.15)
2.06*(1.12)
0.84(1.13)
GroupControls
Included?
Yes Yes Yes Yes Yes Yes Yes
CountryFixed
EffectsIncluded?
Yes Yes Yes Yes Yes Yes Yes
Constant -22.37(2.29)
-21.55(1.74)
-21.55(1.74)
-21.61(2.20)
-22.18(1.76)
-22.40(1.77)
-21.62(1.72)
Log-likelihood
-54.00 -55.51 -55.51 -55.47 -53.70 -53.77 -55.27
N 168 168 168 168 168 168 168Pseudo X² 0.36 0.34 0.34 0.34 0.36 0.36 0.34
Table 6: Testing Negotiation Stability (“talksstable”) using logit models. ***significant at 0.01, **significant at0.05, *significant at 0.10
37
APPENDIX
Table 1A: Service Terms and Sub-Categories
Education/Youth/Recreationamusement
Amusement park
child
childcare
clinics
education
educational
festival
football
fotball
futball
literacy
Primary education
school
secondary
Secondary education
soccer
sport
training
tuition
vocational
Vocational training
youthYouth camp
Health/Emergency
ambulance
ambulance_mobile
antibiotic
blanket
blood
Blood bank
care
clinic
crescent
elderly
Elderly care
healthcare
hospital
immunization
Mobile clinics
rehabilitation
soup
Soup kitchen
Security/Justice
adjudication
court
Court mediation
defense
dispute
Dispute adjudication
Dispute resolution
justice
legal
Legal fees
mediation
militia
police
protection
reprisal
resolution
safety
Jobs/Welfare/Security
bank
benefits
cash
charity
crop
Crop assistance
Development assistance
fertilizer
harambee
insurance
Kitchen
Martyr’s fundSocial insurance
Social welfare
subsidy
unemployment
Unemployment benefits
welfare
widow
zakat
Natural Disaster
disaster
Disaster relief
Disaster services
earthquake
relief
tsunami
Tsunami relief
Public Services
center
collection
Community center
Community centre
Community services
development
food
Food bank
foodbank
garbage
generator
haram
infrastructure
library
light
Mail service
media
38
minister
news
orphanage
park
political
Political wing
postal
Postal service
Public goods
Public safety
Public works
radio
rebuild
reconstruction
Reconstruction subsidy
removal
representation
rubbish
Rubbish collection
sanitation
Sanitation protection
septic
service
services
sewage
Sewage removal
shadow
Social mission
Social service
Social services
street
Street light
taxi
Taxi service
trash
visa
waste
Waste removal
Water
Water access
Water sanitation
Religiouschristian
church
cleric
hadith
islamic
Islamic institutions
mosque
religious
Religious institution
seminaries
madrassa
madrassas
39
Variable Observations Mean Std. Dev Min MaxTotal Services 396 0.11 0.14 0 1
Education,Youth,Recreation
396 0.03 0.08 0 1
Health,Emergency
396 0.04 0.09 0 1
Security,Justice
396 0.06 0.12 0 1
Financial,Jobs, Welfare,Subsidies
396 0.10 0.16 0 1
NaturalDisaster
396 0.04 0.11 0 1
PublicServices
396 0.14 0.19 0 1
Religious 396 0.04 0.09 0 1
Table 2A. Summary statistics of Weighted Service indicator across groups
Moving from…changes the probability of Y by
X%
Model A6: Y=Talks
(CI Bounds)
Model A14: Y=StableTalks
100th ranked service providerto 300th ranked
9.9%**(0.01, 0.18)
5.7%*(-0.005, 0.13)
From Non-Religious Group toReligious Group
-10.9%***(-0.16, -0.04)
-6.5%**(-0.11, 0.01)
From 4 years to 20 years (25th
to 75th percentile)6.7%***
(0.02, 0.11)2.1%
(-0.005, 0.05)From 0 groups to 2 other
groups (25th to 75th percentile)0.06%
(-0.01, 0.03)-0.08%
(-0.02, 0.02)
From no political party topolitical party
27.2%***(0.09, 0.46)
24.4%***(0.09, 0.42)
From 1 base of operations tomultiple bases
-0.3%(-0.08, 0.09)
-0.1%(-0.06, 0.07)
Table 3A: Predicted Probabilities, models 6 and 14, holding all variables at their median values. ***significant at0.01, **significant at 0.05, *significant at 0.10