Article

Journal of Conflict Resolution

2020, Vol. 64(5) 817-843
ª The Author(s) 2019
Refugees, Mobilization, Article reuse guidelines:
sagepub.com/journals-permissions
and Humanitarian Aid: DOI: 10.1177/0022002719885176
journals.sagepub.com/home/jcr

Evidence from
the Syrian Refugee
Crisis in Lebanon
Daniel Masterson1 and M. Christian Lehmann2

Abstract
This article examines whether refugees are prime candidates for recruitment into
armed groups and whether humanitarian aid to refugees impacts their choice to
join armed groups. First, our original survey data of 1,358 Syrian households in
Lebanon provide evidence that mobilization among the refugee population is low
at baseline—the first empirical estimates of the magnitude of the rate of Syrian
refugees returning home to fight. Second, leveraging as-if random assignment
around a strict altitude cutoff for a United Nations cash transfer program for
Syrian refugees, we find little evidence that the aid program had a large effect on
mobilization. If anything, our estimates indicate a small decrease in mobilization.
Our results stand in contrast to published literature arguing that refugees are
prime candidates to join armed groups and humanitarian aid to refugees may
support armed groups and fuel recruitment.

Keywords
civil wars, foreign aid, rebellion, political economy

1
Immigration Policy Lab, Stanford University, CA, USA
2
Department of Economics, University of Brasilia, Brazil

Corresponding Author:
Daniel Masterson, Immigration Policy Lab, Stanford University, CA, USA.
Email: [email protected]
818 Journal of Conflict Resolution 64(5)

Refugees in particular exit the state because of a direct experience of persecution or

political violence and therefore have strong reasons to oppose the regime from which
they have fled. Although refugees are of course victims of violence, they are also prime
candidates for recruitment involvement in rebel factions.
Salehyan (2009, 40)

In this article, we examine the Syrian refugee crisis in Lebanon to test whether
Syrian refugees were prime candidates for recruitment into armed groups and
whether humanitarian aid delivered to the refugees had the perverse effect of fueling
mobilization. The Syrian refugee crisis in Lebanon exhibits a vast majority of the
risk factors for refugee mobilization and aid worsening conflict presented by pub-
lished literature. In Lebanon, we find many of the country’s political and paramili-
tary groups directly or indirectly involved in Syria’s war, often split along
preexisting politicized ethnic divisions; resentment among the host population about
the refugee population; a central state with little capacity to limit mobilization; and
armed groups that regularly crossed porous borders between Lebanon and Syria,
operated near refugee populations, and could have captured humanitarian aid and
recruited refugees. Despite these risk factors, we do not find evidence of refugee
mobilization or that aid affected mobilization.
When and why refugees join armed groups is not simply a question of academic
concern. Over the past one hundred years, two world wars, numerous civil wars,
ethnic cleansing, and genocidal violence have forced millions of people to leave
their homes and communities in order to escape violence. As of June 2018, the
number of refugees worldwide totaled 25.4 million. If we include people forcibly
displaced within the borders of their home country, the number of displaced people
worldwide rises to 68.5 million. Understanding when refugees fight in armed groups
can shed light on the origins and evolution of conflict. It can also help in evaluating
strategies for responding to refugee crises, a central question for the international
community. The United Nations (UN) spends billions of dollars every year on
responses to refugee crises and the amount is increasing. Whether refugees are at
significant risk of joining armed groups, and whether humanitarian aid will make
conflict worse, has important implications for how policy makers should design
responses to refugee crises.
Existing academic and journalistic work highlights cases of refugee crises
where refugees joined armed groups and where transnational insurgent groups
captured aid intended for refugees. Such capture increases groups’ operating bud-
get and capacity to recruit fighters. Furthermore, transnational insurgent groups
may see the potential for profit from recruiting refugees who are well positioned to
help the group secure aid rents. If this is the case, then humanitarian aid that is
meant to alleviate the suffering of refugees may in fact increase refugee mobiliza-
tion into armed groups, strengthening those groups, prolonging refugee crises,
increasing violence in the home country, and increasing the risk of conflict spil-
lover into the refugee hosting country.
Masterson and Lehmann 819

This article makes two key contributions: one is descriptive and the other is
causal. The article’s descriptive contribution studies original household-level data
that record the migration choices of 1,358 Syrian refugee families in Lebanon, and
our analysis finds that the refugees in our sample are unlikely to return home to join
armed groups in Syria. This result stands in contrast to the dominant findings from
published literature (reviewed in the next section) arguing that refugees are prime
candidates to join insurgencies.
The article’s causal contribution provides quasi-experimental evidence suggest-
ing that humanitarian aid to the Syrian refugees in our sample did not impact
mobilization of aid recipients into armed groups. This stands in contrast to a body
of literature arguing that aid to refugees exacerbates conflict. The aid program we
study is a UN unconditional cash transfer program for Syrian households in Leba-
non, implemented in 2013 and 2014, coinciding with a period of intense violence in
Syria and the highest levels of refugee movement from Syria into Lebanon.1 To
identify the causal effect of the aid program, we exploit the fact that only low-
income refugees residing above 500 meters altitude were eligible to receive cash
transfers; hence, our strategy is to compare migration choices among low-income
refugees slightly above and below this arbitrary eligibility cutoff (using a regression-
discontinuity design).
We use original household data with a high level of demographic disaggrega-
tion and a sample of 1,358 refugee households. We are able to provide causal
estimates of the effect of humanitarian aid to refugees because households were
“as-if” randomly assigned to aid based on their altitude of residence. We use proxy
variables to measure mobilization into armed groups, studying a set of variables
that would be affected by armed group mobilization. Each outcome captures
variation in mobilization and other activities as well, which provides respondents
with the safety to answer honestly, avoiding many of the threats to measurement
that arise in survey research on sensitive topics. For example, one metric captures
the change in the number of adult men in the household. If men returned to Syria to
fight, ceteris paribus, we would observe a decrease in the number of men in
households. Our measurement strategy relies on the fact that a proportion of
refugees return not to fight but for other reasons such as to earn an income in
conventional occupations. Because this strategy, by design, adds noise around our
outcome of interest, we draw conclusions from the set of metrics in aggregate and
not any single metric. By examining a large number of outcomes, we can test for
the existence and scale of mobilization. We discuss the measurement strategy in
detail in the Measuring Refugee Mobilization section.
A sizable body of existing work addresses this article’s two research questions of
whether refugees are prime candidates for recruitment into armed groups and
whether humanitarian aid exacerbates mobilization. Although existing research on
refugees and conflict is theoretically rich, offering numerous testable predictions,
many of the research designs are ill-suited to provide dispositive conclusions. Much
theory building for both of the article’s research questions is based on case studies of
820 Journal of Conflict Resolution 64(5)

a small number of exceptional and extreme instances of refugee mobilization and

may exaggerate the risk of refugee recruitment and militarization. Existing theory
testing is often based on observational analysis of cross-national data on refugees
and conflict and consistently finds a positive correlation between refugees and
conflict. This correlational relationship is not surprising as a majority of refugees
flee their home countries because of conflict, and despite a battery of control vari-
ables, the specter of endogeneity—that is, either conflict causing refugees (and
hence aid to refugees) through reverse causation or omitted variables causing both
conflict and refugees (and hence aid)—remains more plausible than the assertion
that refugees (or aid) cause conflict.
One caveat of our quasi-experimental design is external validity. We only sur-
veyed poor refugee families, and of those only the ones living around the elevation
cutoff that determines eligibility for the aid program we study (more specifically,
poor refugee families residing between 450 and 550 meters altitude); hence, it is not
clear whether our results can be extrapolated to Syrian refugees in Lebanon more
broadly. We further discuss external validity in the Online Appendix.

Theory and Existing Evidence on Humanitarian Aid,

Refugees, and Conflict
The majority of existing studies on refugees, aid, and conflict make ominous pre-
dictions, arguing that both refugees and aid to refugees exacerbate conflict.
Although effect sizes are generally not quantified, no clear set of necessary or
sufficient conditions is offered, and counterfactual causal claims are not always
clear, the literature often makes strong statements about the risks of refugee mobi-
lization and of aid to refugees fueling insurgencies.

Are Refugees Prime Candidates to Join Armed Groups?

The existing literature on refugees and conflict argues that refugees may be prime
candidates for recruitment into armed groups and that numerous factors exacerbate
this risk, including refugee camps, lootable aid, a weak host state, preexisting ethnic
rivalries that map onto the cleavages driving the civil war in the country of origin,
transnational insurgent groups operating among refugee populations and camps, and
negative attitudes among refugees toward their home government. All of these
factors are theorized to increase the probability that refugees will join armed groups.
Refugee crises may facilitate the transnational spread of arms, combatants, and
ideologies conducive to conflict (Lischer 2006; Salehyan 2007). In such volatile
settings, refugees have strong incentives to join rebel groups in order to oppose the
regime from which they fled (Salehyan 2009, 40). Refugee populations are not
entirely composed of victims escaping persecution but also political activists who
mobilize while in the host country (Zolberg, Suhrke, and Aguayo 1992). Further-
more, material hardship and legal marginalization, both widespread experiences for
Masterson and Lehmann 821

refugees worldwide, make the financial rewards and sense of purpose from fighting
in an insurgency relatively more attractive (Salehyan 2009, 40-41).
Refugee camps are cited as a leading risk factor for the militarization of refugees.
Salehyan and Gleditsch (2006, 324) write that refugee camps “often provide sanc-
tuary to rebel organizations, a base of operations, and fertile recruitment grounds.”
Zolberg, Suhrke, and Aguayo (1992) describe refugee camps as potential military
bases for “refugee warriors” to continue opposition activities. Weak states hosting
refugees may be incapable of carrying out effective actions to prevent the mobiliza-
tion of refugees, particularly if the state lacks capable and well-equipped security
forces (Lischer 2002, 2006; Salehyan 2007, 2009). A precarious ethnic balance and
preexisting politicized ethnic rivalries in the host country may also increase the risk
of mobilization (Loescher 1992; Whitaker 2003).
All these arguments suggest or state that refugees are viable candidates to join
armed groups but contradict the intuitive fact that most refugees flee their home
country to escape conflict and may be unlikely to return to fight. A sparse but
growing body of work argues that refugee populations are unlikely to incite conflict
because they generally comprise civilian noncombatants (e.g., Leenders 2009;
Onoma 2013; Zhou and Shaver 2019). If refugees are indeed, in general, prime
candidates to join armed groups, then a conducive context may be sufficient to
prompt widespread refugee mobilization. If a conducive environment for mobiliza-
tion, however, is not sufficient, then we will observe refugee crises in weak states
with politicized ethnic divisions where refugees do not mobilize on any substan-
tively important scale.

Does Humanitarian Aid Make Refugees More Likely to Join Armed Groups?
We find divergent predictions in theoretical and empirical work that focuses on how
aid to refugees affects conflict. One line of work, focusing specifically on aid to
refugees, argues that distributing aid near conflict zones may inadvertently support
armed groups and refugee warriors, increasing their capacity and incentives to
recruit new fighters from refugee populations. In contrast, more general work on
crime and insurgency would predict that aid to refugees would reduce mobilization
since it increases the opportunity cost of fighting.
Humanitarian aid to refugees may increase mobilization when it is easily diverted
into the hands of armed groups, such as through taxation, looting, and theft. Such
capture increases the operating budget of transnational insurgent groups (see, e.g.,
Choi and Salehyan 2013, 57), thereby augmenting groups’ capacity to recruit and
pay fighters. Armed groups may increase the intensity and frequency of their activ-
ities in order to access aid rents, to increase the inflow of aid into an area, or to
sustain the conflict and therefore the aid economy that they benefit from. If refugee
populations receive aid, then transnational insurgent groups may see the potential for
profit from recruiting more refugees who are well positioned to help the group
secure aid rents. Aid may be particularly likely to support recruitment efforts if
822 Journal of Conflict Resolution 64(5)

neighbors are unwilling or unable to police refugee communities and prevent the
intrusion of transnational insurgents (Salehyan 2009).2
Existing work argues that refugee camps are particularly conducive to insurgent
recruitment because they create a unique opportunity for fighters to capture huma-
nitarian aid and for fighters themselves to receive aid by blending into refugee
populations (Adelman 1998). Even beyond camp settings, refugees may be partic-
ularly likely to join armed groups due to the dire circumstances of oppression,
poverty, and abuse, and as armed groups capture more aid, the expected returns for
refugees from joining an insurgent group increase.
Furthermore, aid may affect mobilization through its impact on refugee commu-
nity economies. If aid is given in the form of in-kind transfers, it decreases recipi-
ents’ demand for goods and services in their community, which may adversely affect
wages and employment, hence refugee incomes, and reduce the opportunity cost of
returning home to fight. The opposite reasoning applies if aid is given in the form of
cash because it increases recipients’ demand for goods and services, hence income in
the refugee economy.
On the other hand, humanitarian aid to refugees increases the opportunity cost of
fighting, and economic theories of insurrection predict that incentives to rebel grow
smaller as household income and economic opportunities from non-rebel activities
rise (Grossman 1991; Collier and Hoeffler 1998). The theories imply that an increase
in the income of the population raises the opportunity cost of participating in conflict
and thus, all else equal, will reduce refugee mobilization. Despite the widespread use
of the opportunity-cost framework to consider the economic incentives of partici-
pation in insurgencies (Collier and Hoeffler 1998; Fearon and Laitin 2007; Hum-
phreys and Weinstein 2008), scholarship about whether aid in refugee crises will
exacerbate conflict rarely employs it in developing predictions.

The Syrian Refugee Crisis in Lebanon

Alarmingly, the Syrian refugee crisis in Lebanon manifests nearly all the risk
factors forwarded by existing work on refugee mobilization and on the link
between aid to refugees and conflict. Given the presence of many risk factors for
refugee mobilization and for aid exacerbating conflict, the crisis appears to be what
the case-study literature would call a “most likely case” for observing refugee
mobilization and aid to refugees increasing mobilization. Approximately 15 per-
cent to 20 percent of Syrians in Lebanon live in thousands of camps across the
country’s east near the Syrian border.3 Syrian insurgent groups could have found a
rear guard for continuing the war in Syria in Lebanon’s Syrian refugee camps and
areas of the country with dense refugee populations. Syrian and Lebanese armed
groups could have used the areas as recruiting sites for fighting in Syria or for
bolstering their strength in Lebanon.
Many Syrian refugees hold negative views about the Syrian government. As
Corstange (2019) reports, the majority of Syrian refugees (53 percent) support a
Masterson and Lehmann 823

faction of the opposition compared to 39 percent who sympathize with the govern-
ment. Furthermore, Corstange shows that the government draws its popular support
from a base of wealthier Syrians, meaning that the population eligible for humani-
tarian aid will likely exhibit an even higher level of opposition to the government
than in Corstange’s representative sample.
Syrian refugees in Lebanon cannot work legally and have few legal rights;
because Lebanon has not ratified the 1951 Refugee Convention, Syrian refugees
lack many of the rights normally afforded to refugees under the convention. The
Lebanese government crafted its policy response to the refugee crisis to leave
Syrians in a position of precarity, forced to either leave the country or accept
exploitation, vulnerability, and material hardship (Janmyr 2016).
Lebanon is often labeled a “weak state” and state institutions have limited capac-
ity to police refugee populations. After fifteen years of civil war from 1975 to 1990,
Lebanon’s central government never established strong control of all its territory.
Today, the central government fails to deliver basic public services like water,
electricity, and waste collection to most of the country. The country’s system of
sectarian power sharing hinders the state’s ability to provide universal access to
public goods and social services for its citizens, since non-state actors seek to
maintain their own communally based institutions, even going as far as blocking
the development of effective state responses to social problems (Cammett 2014).
Public service provision is further hindered by widespread corruption at high levels
of state bureaucracy (Leenders 2012). Large parts of the country’s south and east are
controlled by the non-state political and military group Hezbollah. Parts of the
country’s northeast lie outside of effective state control, and the Lebanese police
and army rarely enter.
During our study period of 2013 to 2014, Syrian armed groups operated in
Lebanon, traversing the porous border between the two countries. If refugees had
wanted to join a group, they could have done so in Lebanon and then traveled to
Syria. Armed groups fighting in Syria’s war had members living among or near
refugee populations and controlled parts of Lebanese territory in the mountainous
border region with little state presence (Walsh 2017). Throughout 2013 and 2014,
fears among Lebanese policy makers rose sharply that insurgents had “sleeper cells”
among the refugee population (Dionigi 2016, 15). ISIS and Jubhat al-Nusra had a
presence in Syria in the border region adjacent to the central Beqaa valley, along
Lebanon’s eastern border with Syria, and clashed with the Lebanese army and
Hezbollah throughout 2014 (Lebanon World Report 2015). Lebanon’s mountainous
border areas offered a natural pathway for transnational insurgent groups between
contested areas of Syria and Lebanon.
The Lebanese military and security forces lack the capacity to adequately police
mobilization and transnational insurgent movements. Many things move across the
borders, including people, materiel, and medical supplies. The Military Balance
(International Institute for Strategic Studies 2014) states that “Lebanon’s over-
stretched and unevenly capable military and security services struggled to deal with
824 Journal of Conflict Resolution 64(5)

the pressure” (p. 299) of refugee flows, cross-border trade, and movement of jihadi
fighters and government agents, and that the security forces are generally “afflicted
by severe logistical, operational and political problems” (p. 330). In addition to the
general challenges of policing a long and mountainous frontier, many border towns
have close ties with villages just over the border in Syria and long histories of
unfettered cross-border travel.4 Although official numbers are not available, formal
border crossings probably only account for a minority of human movement between
the two countries. In October 2014, an article in the pro-Syrian Lebanese Arabic-
language newspaper, As-Safir, argued that the “and” in “Syria and Lebanon” had
disappeared.5 In April 2015, a member of Lebanon’s Border Control Committee
stated that effective Lebanese border management was impossible while the Syrian
war continued (The Daily Star 2015).
The cleavages of Syria’s war in 2013 and 2014 also drove conflict in Lebanon,
where the divide between the country’s two political factions flows from the coun-
try’s civil war and the question of Syria’s influence in Lebanon. The March 14
political coalition formed in 2005 in opposition to the Syrian occupation of Lebanon
(1976–2005) and today is largely bound together by its anti-Syrian regime stance.
Lebanon’s pro-Syrian regime coalition, March 8, includes Hezbollah as its most
powerful member. Hezbollah is actively fighting on the government’s side in Syria’s
civil war, and the organization receives financial, political, and military assistance
from Iran, a leading patron of the Syrian government in the civil war. Even the
Lebanese Armed Forces suffer from politicized sectarian divisions, reflecting rather
than transcending the country’s internal fault lines (International Institute for Stra-
tegic Studies 2014). Given the clear divisions in Lebanon, and their close relation to
Syria’s civil war, the country’s ethno-sectarian groups could have recruited and
mobilized refugees to form fighting forces.
The foremost driver of Lebanon’s limited capacity to police the refugee popula-
tion was the enormous scale of the population influx, which was larger than nearly
any country would be able to adequately monitor. Lebanon witnessed a 25 percent
population increase from 2011 to 2014, arguably one of the largest relative popula-
tion increases that any modern state has experienced. Despite the limitations on the
Lebanese security apparatus, Lebanon (along with other states in the region)
increased electronic surveillance and bolstered border security to manage and mon-
itor the movement of goods, refugees, fighters, and government agents. Lebanese
security forces also demonstrated capability to disrupt terrorist threats within Leba-
non, some related to the war in Syria.6
Finally, one must not ignore that the decision of armed groups to recruit from
refugee populations, and refugees’ decision to participate, depends on the constraints
imposed by the Lebanese government. We view the observed choices or refugees,
armed groups, and the Lebanese state as an equilibrium with those three actors
making choices to maximize the expected benefits from, respectively, for the actors,
mobilization, recruitment, and monitoring, given the constraints they each face.
Without knowledge of the mobilization rate, the Syrian refugee crisis in Lebanon
Masterson and Lehmann 825

could have been in a high–refugee mobilization equilibrium. Alternatively, the crisis

might have been in an equilibrium where refugee mobilization was low and not a
meaningful threat (and Lebanese security capacity therefore does not matter very
much). Last, the crisis could have been in an equilibrium where refugee mobilization
would have been a threat but for the actions and capacity of the Lebanese security
apparatus. In the descriptive contribution of this article, we provide evidence that
refugee mobilization was low, arguing that the crisis was, at the very least, in one of
the equilibria with a low level of realized mobilization rather than one where refugee
mobilization was high. Our research design is not able to provide evidence as to what
would have happened if Lebanon’s security capacity had been higher or lower.

Experimental Design
The UN Refugee Agency (UNHCR) and partners ran an unconditional cash trans-
fer (UCT) program for Syrian refugees in Lebanon giving US$575 (US$993 in
purchasing power parity (PPP) terms) over six months via ATM cards to 87,700
families.7 US$95 per month is a significant amount for the beneficiary population,
equal to about two-thirds of control-group household monthly income (US$149)
and about one-third of the value of control-group household food consumption,
which is largely supplied by humanitarian aid.8 Our identification strategy relies
on the fact that due to funding restrictions and a desire to target people living in
colder climates, the UN delivered cash transfers to refugee households living
above 500 meters altitude. For a detailed discussion of the cash-grant program,
refer to the Online Appendix.

Sharp Regression-discontinuity Design

We use the 500-meter altitude eligibility cutoff to estimate a (sharp) regression
discontinuity (RD), using household- and community-level data we collected for
all refugees classified as “poor” and residing between 450 and 550 meters altitude.
We chose this particular bandwidth based on available funding for the survey and
bandwidths chosen by the existing literature: +50 meters—that is, +10 percent of
the 500-meter altitude cut point—is similar to bandwidths used in the existing
literature (e.g., Ludwig and Miller 2007). Our bandwidth is small relative to the
entire range of the forcing variable, which ranges from sea level to 2,209 meters
altitude, the highest altitude at which a registered Syrian refugee lives.
Following Hahn, Todd, and der Klaauw (2001) and Imbens and Lemieux
(2008), our estimation approach is local regression around the 500-meter altitude
eligibility cutoff.
Yi ¼ a þ b1ðAi  500Þ þ f ðAi Þ þ ZXi þ ei ; 8450  Ai  550 ð1Þ
where Ai is the altitude of household i at the time of treatment assignment in
October 2013, before the program began in November. a is a constant, b measures
826 Journal of Conflict Resolution 64(5)

the causal effect of the intervention on the outcome Yi at Ai ¼ 500, 1 is the indicator
function, Xi is a vector of covariates (and Z is the corresponding vector of coefficient
estimates), and ei is a mean-zero disturbance term. If Ai  500, the unit received the
treatment and otherwise it did not. We refer to households that lived between 450
and 499 meters altitude at the time of treatment assignment as our control group and
households that lived between 500 and 550 meters altitude as the treatment group.
The term f ðAi Þ is a polynomial function of Ai. Following the suggestion of Lee and
Lemieux (2010), we use linear and quadratic functional forms for f ðAi Þ, allowing for
different slopes of the regression function on both sides of the cutoff:
f ðAi Þ ¼ g1 ðAi  500Þ þ g2 1ðAi  500Þ  ðAi  500Þ ð1aÞ

f ðAi Þ ¼ y1 ðAi  500Þ þ y2 1ðAi  500Þ  ðAi  500Þ þ y3 ðAi  500Þ2

ð1bÞ
þ y4 1ðAi  500Þ  ðAi  500Þ2
In the main article, we present results for equation (1a), and in the Online
Appendix, we present results for equation (1b) and results without covariates, show-
ing that neither changes the article’s overall conclusions.

Study Population
Our population comprises all 1,851 refugee households that, at the time of treatment
assignment, resided between 450 and 550 meters altitude and qualified as “poor”
(1,000 households between 450 and 499 meters, and 851 households between 500
and 550 meters). Eligibility was determined by the 500-meter geographic criterion to
target refugees living at high altitudes exposed to cold weather as well as demo-
graphic criteria to target vulnerable refugees. The UN determined vulnerability
according to a weighted mean of household demographic variables, further dis-
cussed in the Online Appendix. We discuss the ways in which this population differs
from the broader Syrian refugee population in the external validity section in the
Online Appendix.

Data Collection
We surveyed households immediately after the program ended and reached 1,358 of
1,851 households. The main reasons for attrition were that contact information from
the UN was incorrect, people refused to be interviewed, or families had moved back
to Syria. Attrition was balanced across treatment and control groups (74.1 vs. 72.7
percent, Pearson’s w2 test p value ¼ 0.52).9 To facilitate measurement symmetry
between treatment and control groups, we used the same survey technique, instru-
ment, and enumerators to collect data for both treatment and control units, and we
collected data for both groups at the same time and under similar conditions. We also
use baseline data from UNHCR to conduct randomization checks and to construct
Masterson and Lehmann 827

estimates of household demographic changes. A full range of descriptive statistics

about the population is discussed in the Online Appendix.10

Internal Validity
As with all regression-discontinuity designs, our regression estimates reflect the
local average treatment effect at the cutoff, in our case the difference of the values
of the regression functions at the cutoff for each group. This allows us to determine,
all else equal, the effect of being assigned to benefit from this humanitarian aid
program by calculating the magnitude of the discontinuity at the theoretical limit.
The UN’s altitude measure is defined at the town-level, using the highest natural (not
man-made) point. Therefore, a household could be located at 400 meters altitude in a
town with a hilltop above 500 meters and receive aid, whereas a household located at
another town’s highest point at 499 meters would not receive aid.
To violate the as-if random nature of the decision rule, certain types of house-
holds would need to self-select into the treatment or control groups, whether
deliberately or not, before the program began. In order to avoid the possibility
of families sorting into the program, the UN determined eligibility using location
and vulnerability data from October 2013, a full month before the program began,
and we define our treatment and control groups using the same data the UN used.
Refugees received no information about the aid program in advance and no infor-
mation about targeting criteria, mitigating the threat that Syrians self-select into
the treatment or control groups. We show in the Online Appendix that we find no
evidence of sorting or manipulation of the forcing variable, which is consistent
with reports by the UN that it did not publicly announce the program or the
selection criteria. The Online Appendix also presents results for other tests of
identifying assumptions of the RDD, including as-if randomness, contamination
of the control group, and statistical power.

Sample Descriptives
Figure 1 shows the distribution of refugees in Lebanon across altitudes. The upper
graph plots a histogram of all 2,736 towns and villages in Lebanon. The middle
graph shows the altitude distribution of all the 158,129 refugee households who
registered with the UN between March 2011 and October 2013. Of those, 89,597
households were classified as “poor” by the UN Refugee Agency, and the lower
graph plots the altitude distribution of these poor refugee households. The figure
reveals two clusters where refugees mainly settled: first, in Lebanon’s western,
coastal, metropolitan area (at sea level); and second, in the eastern mountainous
area around 1,000 meters altitude, close to the border with Syria. The remainder of
the refugees are spread out between sea level and the mountains. When comparing
refugees classified as poor versus nonpoor, we see that poor refugees are less likely
828 Journal of Conflict Resolution 64(5)

Figure 1. (Top) Nationwide population density of towns, (middle) density of registered

refugees, and (bottom) density of eligible “poor” refugee households.

to live in the metropolitan area and more likely to live at higher altitudes, where the
cost of living is lower.
The geographic distribution of Syrian households registered with UNHCR living
between 450 and 550 meters covers nearly the entire country, running from the north
in Akkar to the south in Bint Jbeil. Figure 2 illustrates the location of all towns where
survey respondents lived at the time of treatment assignment. Respondents who
researchers could contact were surveyed wherever they were living at the time of
survey conduct. In November 2013, when the program began, survey respondents
lived in sixteen of Lebanon’s twenty-five districts (Arabic: aqdia).11 Visual inspec-
tion of treatment and control communities suggests no systematic distribution of
treated and untreated towns across the country.
In our sample, the mean household size at baseline is 5.5 individuals and is
balanced between treatment arms (5.53 individuals among control households and
5.47 among treatment households). We define a household as a group of people
who spend most nights under the same roof and share financial activities like
income and spending.
Twenty-nine percent of our respondents live in a rented house and 59 percent in a
rented apartment. The remainder lives in tents or other improvised shelters; 53
percent have a fridge and 21 percent have a freezer; 2 percent have a car and 4
Masterson and Lehmann 829

Figure 2. Locations of survey respondents at the time of treatment assignment, treatment

communities in blue triangles, and control communities in black squares.

percent have a motorbike; 97 percent of households have a cell phone; 34 percent of

household heads have no schooling or have not finished primary school, 30 percent
have completed primary school, 27 percent have completed middle school, and 9
percent have completed secondary school or higher. The average age of the house-
hold head is 38, and he or she has been in Lebanon for eighteen months. The average
value of household food consumption is about US$300 per month (which is less than
US$2 per day per capita).
Average household labor income of respondents is US$149 per month. This may
understate income since some economic activities may not require the exchange of
currency, such as work for housing; 77.4 percent of adult beneficiaries report zero
working days during the past four weeks, which shows how scarce employment
opportunities are for refugees in Lebanon. Among the 23 percent of adults who
worked at least one day during the past four weeks, the average number of days
worked was 11.7 days. Many households take on debt to make ends meet, with the
average value of currently outstanding cash loans being about US$570 per house-
hold. The average community where our respondents live has 264 Lebanese house-
holds and 96 refugee households.
More than 85 percent of respondents come from five of Syria’s fourteen gover-
norates (Homs: 22.4 percent, Deraa: 20.9 percent, Idlib: 18.6 percent, Aleppo: 15.9
830 Journal of Conflict Resolution 64(5)

percent, and Hama: 8.7 percent). The remaining respondents came from Syria’s
other governorates, except Suweida, where no respondents came from.

Measuring Refugee Mobilization

The challenges of survey research on sensitive topics go beyond the threat of mea-
surement error for the sensitive question itself. In some contexts, explicitly asking
about a sensitive topic may end an interview, introduce measurement error for all
subsequent questions, or put the data collectors or respondents at physical risk.
Existing methods for sensitive topics, like randomization techniques and list experi-
ments, require an explicit statement of the sensitive topic, which was infeasible with
our population and topic.
Our outcome of interest is whether refugees join armed groups, and we use a set
of proxy variables as evidence for that empirical question. To measure refugee
mobilization without explicitly mentioning the topic, we measure outcomes that are
consequences of refugee mobilization. By examining a large number of these out-
comes, we can test for the existence of evidence of an effect of aid on mobilization
and its direction, although we cannot estimate a point estimate of the magnitude of
the effect. We measure thirteen outcomes that would be affected by refugee mobi-
lization. Each outcome captures variation in mobilization and other activities as
well, thereby providing respondents with the safety to answer honestly. Measuring
consequences of our outcome of interest is similar to existing sensitive topic meth-
odologies in that it adds noise around the true estimate of interest, creating a situ-
ation with reduced incentives for systematic misreporting, and we seek to learn
something through the noise.
We draw conclusions from the set of metrics in aggregate. If we estimate a zero
treatment effect for a large number of outcomes that mobilization affects, then the
results serve as evidence that the aid intervention did not affect mobilization.
Although a zero treatment effect for a single metric could suggest an absence of
evidence rather than evidence of a null effect, the larger the number of metrics for
which we estimate no treatment effect, the more confident we can be that we
demonstrate evidence of the absence of an effect.
We measure demographic and behavioral metrics of consequences of mobiliza-
tion. For example, mobilization affects the demographic profile of refugee families,
and if treatment caused more men to return home to fight, we would see fewer adult
men in those households compared to control-group households. We ask how many
individuals moved and how many moved to Syria, and we ask about the age and
gender of all movers. Families with more individuals who returned from Lebanon to
Syria to fight would have, all else equal, more family members who moved and more
who moved to Syria. We also ask whether someone in the family is living in parts of
Syria which are under siege or where there is fighting since if family members are
actively fighting, there is a reasonable chance that they would be in an embattled or
besieged area. We ask if anyone in the household moved to earn money and
Masterson and Lehmann 831

specifically whether anyone moved to Syria to earn money. Because some fighters
in Syria send money to their families living as refugees in neighboring countries,
we also ask households how much money they received from people not living in
the household in general and specifically from individuals living in Syria. We
would expect that, all else equal, families with more members fighting would
receive more transfers in general and more from Syria. We ask whether anyone
in the household undertook physically dangerous activities to earn money, which
could include fighting, and we would expect that households with more individuals
fighting would more frequently report that a family member is engaged in physi-
cally dangerous activities.
The relevant questions were spread throughout the survey and were always
surrounded by other questions on similar topics. Questions about earnings and
movement were placed in a broader section about earnings and employment. We
asked demographic questions about the age and gender of all household members,
instead of focusing on young adult men alone. A question about whether individuals
are under siege was asked in a battery of questions about protection issues. The
survey’s primary purpose was an impact evaluation, and the vast majority of ques-
tions were clearly intended to evaluate straightforward humanitarian concerns.
We present the results for five key variables below and for eight other outcomes
in the Online Appendix. The five outcomes that we present below are the change in
the number of men aged 18 to 50 in each household over the course of the program;
whether someone in the household returned to Syria during the program; the inter-
action of the change in the number of men aged 18 to 50 and whether a household
member returned to Syria during the program; whether a family member is currently
living in Syria in an active war zone, defined as an area in Syria that is under siege or
where there is active fighting; and the interaction of whether a family member is in
an active war zone in Syria and whether a household member returned to Syria
during the program.
The interacted outcomes maintain the protection of individual responses, while
increasing the precision of overall measurements. The interactions focus our anal-
ysis on the substantively important variation in the consequence of the outcome. The
first interaction (the interaction of the change in the number of men aged 18 to 50
and whether a household member returned to Syria during the program) forces a 0
value for the metric of the change in the number of men if the household had no one
return to Syria. The second interaction (the interaction of whether a family member
is in an active war zone in Syria and whether a household member returned to Syria
during the program) forces a 0 value for whether someone returned to Syria if no one
in the family is living in a siege zone. Tests on the two interacted outcomes should be
interpreted, respectively, as (i) whether there is a treatment effect on the change in
the number of men who returned to Syria and (ii) whether there is a treatment effect
on whether a household had someone return to Syria and has a family member living
under siege.
832 Journal of Conflict Resolution 64(5)

Although estimating treatment effects for numerous outcomes and interacted

outcomes would normally raise concerns about fishing or p-hacking, we do not
adjust p values for multiple comparisons because we are arguing that there is no
effect of treatment on our outcomes, meaning that the unadjusted p values provide a
conservative measure. Adjusting the p values for multiple comparisons would only
increase our p values, thus bolstering our argument.

Results
Refugees and Mobilization
Based on the dominant predictions from the existing literature, we would expect that
the Syrian refugee crisis in Lebanon, in the context of a weak state, ethnic divisions,
and porous borders, would pose a significant risk of refugee mobilization. In con-
trast, we present evidence that the share of refugees who went back to fight is
probably very small. That is, at the time of the study, November 2013 to March
2014, a very small number of Syrian refugees returned home to fight.
We begin by estimating the magnitude of the number of Syrian refugees who
returned to Syria, regardless of the reason. The survey asked respondents how many
members of their household returned to Syria during the study period, which allows
us to estimate our sample’s baseline rate of return to Syria from our control group.
We then extrapolate from our control group to the broader population to obtain an
estimate of the number of returnees to Syria in Lebanon’s refugee population during
the study period. We multiply the average number of returnees per household in our
control group, denoted rC, by the total number of Syrian households in Lebanon k.
As we discuss more below, our sample is not representative of the full Syrian
population, and we therefore cautiously interpret our estimates as evidence of the
magnitude of return but not as point estimates of the return rate. The magnitude of
rC  k denotes an estimate of the magnitude of the number of Syrian refugees who
returned during the study period. Next, to estimate the magnitude of return to join
armed groups, we estimate the number of respondents in our sample who returned to
Syria to join armed groups, denoted m, by offering a hypothetical value of the share
of refugees who returned to Syria to join armed groups, denoted s. Our estimate of
the number of Syrians who returned to Syria to join armed groups is calculated as:
m ¼ rC  k  s ð2Þ
We observe rC from the survey data, draw k from UNHCR registration data,
and input conservative (i.e., high) estimates of s. By calculating m for a range of
values of s, for example, ranging from 1 percent to 5 percent, we can imagine
the plausible magnitude of return to join armed groups given the return rate we
observe in our control group.
To be clear, this simulation does not provide evidence of the mobilization rate
among refugee returnees, and we do not suggest that s could be as high as 5 percent.
Masterson and Lehmann 833

We use s ¼ 5 percent to illustrate that even if one assumes a high mobilization rate,
the overall magnitude of mobilization would be small given the low return rates in
our data. Five percent would be a very high mobilization rate if applied to the general
population, given the collective action problem of rebellion (see Lichbach 1995, the
5 percent rule).
We find evidence that the magnitude of refugee mobilization was likely substan-
tively small, suggesting that Syrian refugees in Lebanon were not prime candidates
for recruitment into armed groups in Syria’s civil war. Of the 727 households in the
control group, 49 had at least one person return to Syria during the study period, only
6.7 percent of households, and among those households, the average number of
returnees was 1.6 (rC ¼ 0.067 1.6). In October 2013, the point at which program
eligibility was determined, 691,709 registered Syrians lived in Lebanon in 158,129
households (k ¼ 158,129). If 6.7 percent of the households had an average of 1.6
people return, then 16,867 Syrians returned to Syria from November 2013 to April
2014. If we were to assume that s ¼ 0:01, that is, 1 percent of Syrians who returned
to Syria went to join armed groups, our best guess is that a very small number of
Syrian refugees returned to fight—roughly 170 people from 691,709 refugees. Even
if we were to assume that a staggering 5 percent of returnees went to join armed
groups, our best guess is that the Syrian refugee population in Lebanon contributed
roughly 850 people from 691,709 refugees to the ranks of Syrian armed groups, at a
time when there were between 75,000 and 110,000 members in insurgent groups in
Syria and approximately 178,000 soldiers in the Syrian Army.12
Our estimate should be understood as evidence of the magnitude of mobilization
and should not be taken as a point estimate of the number of Syrian refugees
returning to fight in Syria. When the estimate is interpreted as evidence of the
magnitude of mobilization, the number is clearly substantively small. Although our
sample is not representative of the Syrian refugee population in Lebanon, the
broader population’s behavior would need to be dramatically different from our
sample’s to suggest the possibility of a meaningful level of refugee mobilization
among Syrian refugees in Lebanon. To imply that even one percent of refugees
joined armed groups, one would need to argue that the rate of mobilization was
approximately forty times higher than in our subsample of poor Syrians living
around 500 meters altitude. Overall, the evidence presented here suggests that the
Syrian refugee crisis in Lebanon does not justify a claim like Salehyan’s (2009) that
refugees are prime candidates for recruitment into rebel factions.
Although our estimate is almost certainly biased by the fact that our sample is not
representative of the broader Syrian refugee population, this is the best data available
on the phenomenon since the Lebanese government does not share border-crossing
data. Furthermore, the estimates may be upwardly biased because our sample is a
poor subpopulation. Due to the economic geography of the conflict, poor and lower-
middle-class Syrians may be more likely to fight. And they have few economic
opportunities in Lebanon, which existing research would predict puts them at higher
risk of mobilization (Humphreys and Weinstein 2008). Therefore, the bias may run
834 Journal of Conflict Resolution 64(5)

Table 1. Results: Linear Fit with Covariates (Equations [1] and [1a]).

3. Change in
Number of Men 5. Household
1. Change in 2. Household Aged 18 to 50 4. Family Member
Number of Member when Someone Member in Returned to
Men Aged 18 Returned Returned an Active an Active
to 50 to Syria to Syria War Zone War Zone

^
b .002 .03 .001 .1188 .064
(.069) (.033) (.018) (.1194) (.028)
Control-group mean .173 .067 .011 .238 .043
p Value .978 .366 .946 .32 .021
^
Standardized b .003 .12 .007 .2787 .317
[.107] [.132] [.122] [.279] [.138]

Note: N ¼ 1,358. Because the forcing variable, and thus treatment assignment, was determined at the
village level, we use Eicker–Huber–White robust standard errors, clustered at the village where
respondents lived at the time of treatment assignment, reported in parentheses. The standardized
standard error is presented in square brackets. The bandwidth in all regressions is h ¼ 50 meters. The
control-group mean shows the regression model’s prediction of the dependent variable for refugees
residing at 499 meters altitude. All variables refer to the time period from November 2013 to April
2014. Covariates include baseline household demographics (number of children, adults, elderly),
education, age, and Syria origin of household head.

counter to our argument that refugee mobilization among Syrians in Lebanon in

2013 and 2014 was low.
Syrians may return from Lebanon to fight in the Syrian Army rather than in
insurgent groups, and the evidence above pertains to both forms of mobilization.
Many Syrians fled violence perpetrated by insurgent groups, rather than the gov-
ernment, and Salehyan’s (2009) grievance-based argument about why refugees
return to fight could be understood to explain why Syrians would want to join the
national army. We know from literature on recruitment that fighters often join the
group that is present or strongest in the area in which they live (e.g., Arjona and
Kalyvas 2012). Our data indicate that the magnitude of return in 2013 to 2014 was
very small, and the magnitude of return to fight in armed groups including insurgent
groups and the army must therefore have been small.

Aid and Mobilization

Although published evidence suggests that aid delivered to refugees often exacer-
bates conflict, we find no strong evidence that this occurred in the Syrian refugee
crisis in Lebanon. The majority of metrics and models in Table 1 and the Online
Appendix have point estimates that are close to zero and most standardized effect
sizes are small, providing little evidence that the program had a large effect on
mobilization. Furthermore, when estimates are not well-identified zeros, the signs
of point estimates suggest a small decrease in mobilization due to aid rather than an
Masterson and Lehmann 835

increase. This suggests that if aid had a nonnegligible impact on mobilization, it was
likely a decrease, not an increase, possibly because aid increases the opportunity
costs of participation in insurgency.
Table 1 presents results of the linear model with covariates (equations [1] and
[1a]). Figure 3 offers a visual presentation of the same results. Because the
forcing variable, and thus treatment assignment, was determined at the village
level, we use Eicker–Huber–White robust standard errors clustered at the village
level according to where respondents lived at the time of treatment assignment.
Results in the article are robust to model specification and different outcome
variables. The Online Appendix presents results for linear and quadratic models
with and without covariates and a number of additional outcomes including
dimension-reduced outcomes calculated using principal component analysis
(PCA).
In Table 1, b ^ denotes the estimated treatment effect at Ai ¼ 500. The control-
group mean shows the regression model’s prediction of the dependent variable for
refugees residing at 499 meters altitude. Because the outcomes are measured in
different units, we also derive standardized effect-size estimates by dividing b ^ by
the control-group standard deviation, giving us the estimated treatment effect at
Ai ¼ 500 in terms of each outcome’s standard deviation, which is analogous to
Cohen’s d effect size. The standardized standard error, presented in square brackets,
indicates our uncertainty about the treatment effect estimate in terms of each out-
come’s standard deviation. Cohen (1988) offers a rule of thumb that d ¼ 0:2 can be
considered a small, but not necessarily trivial, effect, and d ¼ 0:5 can be considered
a medium effect.
The results show a lack of strong evidence of an effect of treatment on mobiliza-
tion. The evidence points toward a null effect and if anything a negative effect of aid
on mobilization. Across all the metrics, we find essentially no evidence pointing
toward an increase in mobilization. The treatment effect estimates for outcomes (1)
and (3) are well-identified zeros—small treatment effect estimates with narrow
confidence intervals. The treatment effect estimate for outcome (2) is less precise
but also small. These three null results provide evidence that the cash program did
not affect Syrians’ choices to return to Syria and did not affect the return patterns of
Syrian refugee men aged 18 to 50.
Looking at outcomes (4) and (5), we see that estimated effect sizes are not
small, but the sign of the point estimates suggests that aid decreased mobilization,
rather than increasing it. Although we cannot confidently assert that there was no
change in whether a family member is living in an active war zone or whether a
household member returned to one, the findings at most suggest that humanitarian
aid decreased the rates of these outcomes. If humanitarian aid had increased
refugee mobilization, we would be unlikely to observe the estimated negative
relationship between treatment and living in an active war zone. A decrease in
mobilization aligns with what we would expect from the opportunity-cost
836
3

0.5

4
0.4
2

2
●

0.3
1

0
0.2
●

0
● ●

Var. 1 by household
Var. 3 by household

0.1
●

Var. 2 mean by altitude

−2
−1

0.0
−40 −20 0 20 40 −40 −20 0 20 40 −40 −20 0 20 40
Altitude Altitude Altitude

Var. 1: Change in number

1.0
of men ages 18−50

0.30

0.8
Var. 2: A household member
returned to Syria (0/1)

0.6
0.20
Var. 3: Change in number of men 18−50
when someone returned to Syria

0.4
Var. 4: A family member in

0.10
●
an active war zone (0/1)

0.2
●

Var. 4 mean by altitude

Var. 5 mean by altitude
Var. 5: Household member
returned to an active war zone (0/1)

0.0
0.00
−40 −20 0 20 40 −40 −20 0 20 40
Altitude Altitude

Figure 3. Linear fit with covariates (equations [1] and [1a]).

Masterson and Lehmann 837

literature but contradicts what we would expect from the literature on refugees,
humanitarian aid, and conflict.
Looking at the full set of 40 regressions presented in the main body of the
article and the Online Appendix, we find the same overall trends: the evidence
points toward a null effect and if anything a decrease in mobilization due to aid.
We find essentially no evidence pointing toward an increase in mobilization.
Looking at the sign of our point estimates, twenty-four of forty treatment effect
estimates have a sign pointing toward a decrease in mobilization. Among these
twenty-four, five estimates have p values less than 0.05 and ten have standar-
dized coefficient sizes greater than 0.2. These five estimates are entirely for two
outcome variables: whether a household member undertook dangerous work in
the past six months and the interaction of whether a household member returned
and whether a family member is living in a siege zone.
Among the sixteen of forty treatment effect estimates that have a sign point-
ing toward an increase in mobilization, almost all are well-identified zeros.
Among these sixteen, only two estimates have a p value less than 0.05, and all
sixteen have standardized coefficient sizes less than 0.2. That is, among the 40
percent of treatment effect estimates where the sign points toward an increase in
mobilization, they are all substantively small and only two are statistically
distinguishable from zero.
Looking at p values of our point estimates, only seven of forty treatment effect
estimates have p values less than 0.05, and five of these seven estimates point toward
a negative treatment effect on mobilization. That is, among the 17.5 percent of
treatment effect estimates that are statistically distinguishable from zero, all but two
point toward a decrease in mobilization.
Looking at standardized coefficient size, ten of forty treatment effect estimates
have a standardized treatment effect estimate larger than 0.2, and among these ten
regressions, all estimates point in the direction of a decrease in mobilization. That
is, among the 25 percent of treatment effect estimates that are not substantively
small, all have signs pointing toward a decrease in mobilization due to aid. In
addition to the forty regressions discussed above, the Online Appendix presents
results from twelve PCA regressions, which provide no strong evidence of a
treatment effect on the metrics studied. It is important to note again that while
such a large number of regressions would often raise concerns about fishing or
p-hacking, it should not in our study because we are arguing that there is no effect
of treatment on our outcomes.
Similar to the results from the descriptive analysis of the magnitude of Syrian
refugee return, if aid did affect Syrian refugees’ choices about whether to return
to Syria to join an armed group, we would not be able to distinguish joining an
insurgent group from joining the national army. Nonetheless, the extremely low
levels of return, and the lack of evidence of any effect of aid on return rates,
stands as a lack of evidence of return to join the insurgency or to join the
national army.
838 Journal of Conflict Resolution 64(5)

Conclusion
Given the prevailing conclusion of existing literature that refugees and aid often
exacerbate conflict, we might question the wisdom of providing humanitarian
aid to refugees. This article is the first to provide direct quasi-experimental
evidence to the relevant policy questions: when refugee crises occur, are refu-
gees likely candidates to join armed groups, and what are the impacts of huma-
nitarian aid on mobilization? Our main results provide evidence that contradicts
the dominant findings in the published literature that refugees and aid cause
conflict. First, we present evidence that Syrian refugees were not prime candi-
dates for recruitment into armed groups. Second, we show that aid did not have
a large effect on refugee mobilization in this study. Among estimates that have
broad confidence intervals, the signs of point estimates suggest a decrease in
mobilization due to aid rather than an increase, likely because aid increases the
opportunity costs of participation in fighting.
This study uses a natural experiment and in-depth knowledge of the context to
establish specific causal facts for a well-defined subpopulation rather than a general
“universe-of-cases” regression, and this should not be viewed as a limitation (Samii
2016). As Aronow and Samii (2016) show, the trade-off between internal and
external validity is illusory; simply because a regression model includes observa-
tions for many countries does not mean it is representative of that population.
Questions about whether some feature of the context may limit external validity
are at their core questions about heterogeneous treatment effects or conditional
average causal effects. Future research should explore whether effects of humani-
tarian aid vary by features of the context such as the brutality of the civil war and
whether different types of humanitarian aid may fuel armed groups differently.
While the majority of published literature would predict refugee mobilization as
well as a high risk of aid to refugees worsening conflict in the context under study,
we find little evidence of either. Our results suggest that existing theories arguing
that refugees are prime candidates for mobilization, and that aid will often exacer-
bate conflict, have limited predictive power. Our estimates provide a baseline from
which other studies can progress in accumulating evidence addressing the question
of whether refugees and aid cause conflict. Our findings also highlight the need for
more case-oriented research in the domain of civil war, which may cumulatively
shape our understanding of when and why humanitarian aid and refugees will
exacerbate, alleviate, or—importantly—have no effect on civil conflict.

Authors’ Note
We benefited from conversations with Peter Aronow, Vivek Ashok, Adam Baczko, Natalia
Bueno, Alex Coppock, Allan Dafoe, Thad Dunning, Ellen Lust, Lama Mourad, Lilla Orr,
Amelia Reese Masterson, Niloufer Siddiqui, Korinayo Thompson, Guadalupe Tuñón, Andres
Vargas, Elisabeth Wood, Hikaru Yamagishi, Remco Zwetsloot, and participants at the Aro-
now lab and the Yale Order, Conflict, and Violence workshop.
Masterson and Lehmann 839

Acknowledgments
We thank the International Rescue Committee and the United Nations Refugee Agency for
collaboration and Chris Blattman and Juliette Seban for their support in the early stages of
this project. We are grateful for funding through DFID grant agreement number 204007-
111. We have no relevant or material financial interests that relate to the research described
in this article. Replication materials are available at http://jcr.sagepub.com/ and Master-
son’s Dataverse.

Declaration of Conflicting Interests

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

Funding
The author(s) disclosed receipt of the following financial support for the research, authorship,
and/or publication of this article: This work was supported by the Department for Interna-
tional Development grant agreement number 204007-111.

Supplemental Material
Supplemental material for this article is available online.

Notes
1. Source for levels of violence in Syria: Price, Gohdes, and Ball (2014). Source for levels of
refugee movement from Syria into Lebanon: UNHCR Data Portal. Accessed September
19, 2018. https://data2.unhcr.org/en/situations/syria/location/71.
2. The literature about how aid to conflict zones (i.e., nonrefugee populations) affects
conflict, specifically on the perverse effects of aid, highlights the effects of aid on the
budget constraints of rebel groups and state actors (Besley and Persson 2011; Nunn and
Qian 2014; Collier and Hoeffler 2002; De Ree and Nillesen 2009; Ahmed and Werker
2015) as well as the potential incentive it provides for actors to increase violence
against civilians in expectation of obtaining aid (e.g., Anderson 1999; De Waal
1997; Polman 2010).
3. The United Nations (UN) has not established official refugee camps in Lebanon. People
conversant in nongovernment organization (NGO)/UN legalese may be familiar with the
term “informal settlement” (IS) used to describe refugee camps in Lebanon. The term is
meant to emphasize the fact that the camps are not run by the UN Refugee Agency. We
maintain that the difference is more bureaucratic than useful, and we deliberately use the
term “camp” rather than IS. There is no reason that the existing literature would predict
that the camp being formally run by the UN would be a necessary condition for
mobilization.
4. For example, Karha and Kaniset Akkar in Wadi Khaled, the village of Deir El Aachayer
in the south of Beqaa Governorate, and the area along the western Qalamun Mountains
near Aarsal. For more discussion, see Vignal (2017).
840 Journal of Conflict Resolution 64(5)

5. www.assafir.com/Article/1/377949, quoting the mid-century Lebanese writer Sayyid

Taqi al-Din, the “and” in Syria and Lebanon has become an “unfaithful and,” waw
ka fira.
6. During a spate of terrorist attacks in 2014, a number of attacks were carried out prema-
turely when Lebanese security forces stopped suspects’ vehicles at checkpoints, which
security services often put up around the country. Lebanese intelligence was actively
trying to prevent terrorist attacks. On June 20, 2014, for example, when 102 people were
arrested in Hamra hotels and 17 were detained as terrorist suspects at the International
Security Force Intelligence Bureau.
7. “Unconditional” refers to requirements on the beneficiaries’ actions, like children’s
school attendance or regular medical checkups. “Unconditional” does not mean that
selection is not conditional on eligibility criteria, like poverty or altitude. Although UN
Refugee Agency (UNHCR) determined eligibility for the program, managed the funding
and the partnership with the bank that managed the ATM cards, and scheduled distribu-
tions and electronic transfers to the ATM cards, Lebanese and international nongovern-
mental organizations with local capacity in certain parts of the country were contracted to
implement the physical distribution of ATM cards at the beginning of the program.
UNHCR distributed more than half of the cards. Other organizations that managed local
card distribution included ACTED, AMURT, AVSI, CARE, Caritas, CISP, DRC, Handi-
cap International, Humedica, IOCC, IOM, Makhzoumi, MEDAIR, Mercy Corps, NRC,
Oxfam, Save the Children, SHEILD, SIF, Solidar Suisse, and World Vision.
8. Both income and household consumption statistics come from our survey data.
9. Enumerators informally collected information on reasons for attrition from neighbors and
shopkeepers. Beyond demonstrating that attrition rates were balanced across groups, we
cannot test for balance in the reasons for attrition.
10. The complete survey questionnaire is available at http://tinyurl.com/pvyub87.
11. At the time of treatment assignment, respondents were living in the following districts: El
Batroun, Chouf, Kesrwane, El Nabatieh, Marjaayoun, Aley, Akkar, El Minieh-Dennie,
Sour, Jbeil, El Koura, El Meten, Bent Jbeil, Baabda, Jezzine, and Hasbaya.
12. Insurgent Groups: January 29, 2014. Congressional testimony by US Director of National
Intelligence James Clapper. Transcript available via The Washington Post. http://wapo.st/
1lmFlmR?tid¼ss_tw. Accessed March 17, 2017. Syrian Army: Aram Nerguizian. 2015.
“The Military Balance in a Shattered Levant.” Center for Strategic and International
Studies, p. 22.

References
Adelman, Howard. 1998. “Why Refugee Warriors Are Threats.” Journal of Conflict Studies
18 (1): 49-69.
Ahmed, Faisal Z., and Eric Werker. 2015. “Aid and the Rise and Fall of Conflict in the
Muslim World.” Quarterly Journal of Political Science 10 (2): 155-86.
Anderson, Mary B. 1999. Do No Harm: How Aid Can Support Peace—Or War. Boulder, CO:
Lynne Rienner.
Masterson and Lehmann 841

Arjona, Ana M., and Stathis N. Kalyvas. 2012. “Recruitment into Armed Groups in Colombia:
A Survey of Demobilized Fighters.” In Understanding Collective Political Violence,
edited by Y. Guichaoua, 143-71. London, UK: Palgrave Macmillan.
Aronow, Peter M., and Cyrus Samii. 2016. “Does Regression Produce Representative Esti-
mates of Causal Effects?” American Journal of Political Science 60 (1): 250-67.
Besley, T., and T. Persson. 2011. “The Logic of Political Violence.” Quarterly Journal of
Economics 126 (3): 1411-45.
Cammett, Melani. 2014. Compassionate Communalism: Welfare and Sectarianism in Leba-
non. Ithaca, NY: Cornell University Press.
Choi, Seung-Whan, and Idean Salehyan. 2013. “No Good Deed Goes Unpunished: Refugees,
Humanitarian Aid, and Terrorism.” Conflict Management and Peace Science 30 (1):
53-75.
Cohen, Jacob. 1988. Statistical Power Analysis for the Behavioral Sciences. Hillsdale, NJ:
Lawrence Erlbaum Associates.
Collier, P., and A. Hoeffler. 1998. “On Economic Causes of Civil War.” Oxford Economic
Papers 50:563-73.
Collier, P., and A. Hoeffler. 2002. “Aid, Policy and Peace: Reducing the Risks of Civil
Conflict.” Defence and Peace Economics 13 (6): 435-50.
Corstange, Daniel. 2019. “The Syrian Conflict and Public Opinion among Syrians in
Lebanon.” British Journal of Middle Eastern Studies 46 (1): 178-200.
De Ree, Joppe, and Eleonora Nillesen. 2009. “Aiding Violence or Peace? The Impact of
Foreign Aid on the Risk of Civil Conflict in Sub-Saharan Africa.” Journal of Development
Economics 88 (2): 301-13.
De Waal, Alexander. 1997. Famine Crimes: Politics & the Disaster Relief Industry in Africa.
Bloomington: Indiana University Press.
Dionigi, Filippo. 2016. “The Syrian Refugee Crisis in Lebanon: State Fragility and Social
Resilience.” LSE Middle East Centre Paper Series 15. London, UK: Middle East Centre,
London School of Economics.
Fearon, James D., and David D. Laitin. 2007. “Civil War Termination.” Stanford University.
Unpublished Manuscript.
Grossman, Herschell I. 1991. “A General Equilibrium Model of Insurrections.” The American
Economic Review 81 (4): 912-21.
Hahn, Jinyong, Petra Todd, and Wilbert der Klaauw. 2001. “Identification and Estimation of
Treatment Effects with a Regression-discontinuity Design.” Econometrica 69 (1): 201-09.
Humphreys, Macartan, and Jeremy M. Weinstein. 2008. “Who Fights? The Determinants of
Participation in Civil War.” American Journal of Political Science 52 (2): 436-55.
Imbens, Guido W., and Thomas Lemieux. 2008. “Regression Discontinuity Designs: A Guide
to Practice.” Journal of Econometrics 142 (2): 615-35.
International Institute for Strategic Studies. 2014. “The Military Balance.” 114 (1).
Janmyr, Maja. 2016. “Precarity in Exile: The Legal Status of Syrian Refugees in Lebanon.”
Refugee Survey Quarterly 35 (4): 58-78.
Lebanon World Report 2015: Lebanon. Events of 2014. Human Rights Watch. Accessed
February 25, 2018. https://www.hrw.org/world-report/2015/country-chapters/lebanon.
842 Journal of Conflict Resolution 64(5)

Lee, David S., and Thomas Lemieux. 2010. “Regression Discontinuity Designs in Economics.”
Journal of Economic Literature 48 (2): 281-355.
Leenders, Reinoud. 2009. “Refugee Warriors or War Refugees? Iraqi Refugees’ Predicament
in Syria, Jordan and Lebanon.” Mediterranean Politics 14 (3): 343-63.
Leenders, Reinoud. 2012. Spoils of Truce: Corruption and State-building in Postwar Leba-
non. Ithaca, NY: Cornell University Press.
Lichbach, Mark I. 1995. “The 5% Rule.” Rationality and Society 7 (1): 126-28.
Lischer, Sarah Kenyon. 2002. “Catalysts of Conflict: How Refugee Crises Lead to the Spread
of Civil War.” PhD Thesis, Massachusetts Institute of Technology, Cambridge.
Lischer, Sarah Kenyon. 2006. Dangerous Sanctuaries: Refugee Camps, Civil War, and the
Dilemmas of Humanitarian Aid. Ithaca, NY: Cornell University Press.
Loescher, Gil. 1992. Refugee Movements and International Security. London, UK: Brassey’s
for the International Institute for Strategic Studies, number 268.
Ludwig, Jens, and Douglas L. Miller. 2007. “Does Head Start Improve Children’s Life
Chances? Evidence from a Regression Discontinuity Design.” The Quarterly Journal of
Economics 122 (1): 159-208.
Nunn, Nathan, and Nancy Qian. 2014. “US Food Aid and Civil Conflict.” American Eco-
nomic Review 104 (6): 1630-66.
Onoma, Ato Kwamena. 2013. Anti-refugee Violence and African Politics. Cambridge, MA:
Cambridge University Press.
Polman, Linda. 2010. The Crisis Caravan: What’s Wrong with Humanitarian Aid? New
York: Palgrave Macmillan.
Price, Megan, Anita Gohdes, and Patrick Ball. 2014. Updated Statistical Analysis of Docu-
mentation of Killings in the Syrian Arab Republic. Geneva, Switzerland: Human Rights
Data Analysis Group.
Salehyan, Idean. 2007. “Transnational Rebels: Neighboring States as Sanctuary for Rebel
Groups.” World Politics 59 (02): 217-42.
Salehyan, Idean. 2009. Rebels without Borders: Transnational Insurgencies in World Politics.
Ithaca, NY: Cornell University Press.
Salehyan, Idean, and Kristian Skrede Gleditsch. 2006. “Refugees and the Spread of Civil
War.” International Organization 60 (02): 335-66.
Samii, Cyrus. 2016. “Causal Empiricism in Quantitative Research.” The Journal of Politics
78 (3): 941-55.
The Daily Star. 2015, April 29. Accessed February 25, 2018. http://www.dailystar.com.lb/
News/Lebanon-News/2015/Apr-29/296141-effective-border-management-a-pipedream.
ashx.
Vignal, Leila. 2017. “The Changing Borders and Borderlands of Syria in a Time of Conflict.”
International Affairs 93 (4): 809-27.
Walsh, Declan. 2017. “Hezbollah and Syrian Army Attack Islamists on Lebanon Border.”
New York Times. Accessed February 25, 2018. https://www.nytimes.com/2017/07/21/
world/middleeast/hezbollah-syrian-army-lebanon.html.
Whitaker, Beth Elise. 2003. “Refugees and the Spread of Conflict: Contrasting Cases in
Central Africa.” Journal of Asian and African Studies 38 (2-3): 211-31.
Masterson and Lehmann 843

Zhou, Yang-Yang, and Andrew Shaver. 2019. “Reexamining the Effect of Refugees on Civil
Conflict: A Global Subnational Analysis.” Working Paper. Princeton, NJ: Princeton
University.
Zolberg, Aristide R., Astri Suhrke, and Sergio Aguayo. 1992. Escape from Violence: Conflict
and the Refugee Crisis in the Developing World. Oxford, UK: Oxford University Press on
Demand.