World Development 126 (2020) 104713

Contents lists available at ScienceDirect

World Development
journal homepage: www.elsevier.com/locate/worlddev

Humanitarian food aid and civil conflict

Sébastien Mary a,c,⇑, Ashok K. Mishra b
a
The University of Kansas, Department of Economics, Lawrence, KS 66045, United States
b
Arizona State University, W.P. Carey School of Business, Morrison School of Agribusiness, Mesa, AZ 85212, United States
c
DePaul University, Department of Economics, Chicago, IL 60604, United States

a r t i c l e i n f o a b s t r a c t

Article history: Humanitarian food aid has long been considered to be an effective tool towards conflict mitigation among
Accepted 11 October 2019 donors and policymakers. Within the Sustainable Development Goals that have the objectives of ending
hunger before 2030 (SDG#2) and bringing peace and justice (SDG#16), humanitarian food assistance may
play a critical role in delivering progress in developing countries. However, there have been growing con-
JEL codes: cerns that it may actually have counter-intended effects by aggravating civil conflicts in recipient countries.
F35 We estimate the effect of humanitarian food aid on civil conflict using a sample of 79 recipient countries
Q18
between 2002 and 2017. Our analysis exploits cross-sectional and time variation in between-country
D74
humanitarian food aid displacements. Our baseline instrumental variables estimates imply that a 10 percent
Keywords: increase in humanitarian food aid per capita decreases the incidence of civil conflict by about 0.2 percentage
Civil conflict point (or by about 0.9 per cent at the mean conflict incidence). Humanitarian food aid also decreases the
Humanitarian food aid incidence of small-scale and large-scale civil conflicts, and the onset and duration of civil conflicts.
Sustainable development goals Ó 2019 Elsevier Ltd. All rights reserved.
Africa

1. Introduction towards hunger reduction and conflict prevention and mitigation

Does humanitarian food aid save lives in the developing world?
The question has received increased scientific (e.g. Nunn & Qian,
2014), media (e.g. The Economist, 2014), and policy attention
(e.g. CIA, 2014). The recent worsening of food insecurity in sub-
Saharan Africa and South Eastern and Western Asia caused by a
resurgence of civil conflicts (FAO, Ifad, UNICEF, WFP, & WHO,
2017), despite having received large amounts of humanitarian food
assistance, reinforces the legitimacy of the question (and by impli-
cation, the need to provide an answer).
Humanitarian food aid1 is typically provided in emergency and
conflict situations, and includes the provision and distribution of
food, cash and vouchers for food purchases, as well as non-medical
nutritional interventions for the benefit of crisis-affected people
(OECD, 2019)2. It has long been considered to be an effective tool
⇑ Corresponding author.
E-mail addresses: smary@depaul.edu (S. Mary), ashok.k.mishra@asu.edu (A.K. Mishra).
1
Humanitarian or emergency food aid (coded 72,040 under the Credit Reporting
System) is different from food aid (coded 52,010 under the Credit Reporting System).
2 (Polman, 2010; Magone, Neuman, & Weissman, 2012), though it is
Humanitarian aid includes humanitarian food aid, as well as several other aid
types: (1) material relief and assistance and services related to shelter, water,
sanitation and health services, supply of medicines; (2) relief coordination, protection
and support services; (3) reconstruction relief and rehabilitation, especially for pre-
existing infrastructures (e.g. roads, bridges, water and sanitation); (4) disaster
prevention and preparedness, including disaster risk reduction activities, early
warning systems, emergency contingency stocks; (5) administrative costs and other.
among policymakers (e.g. Open letter to US congress, 2017).
There are indeed several mechanisms through which humani-
tarian food aid may mitigate civil conflict. If civil conflicts arise
because of competition for scarce resources crucial to survival
(i.e., food), then the delivery of humanitarian food aid increases
the opportunity cost of joining a rebellion (Collier and Hoeffler,
1998, 2004) and reduces the incentive to fight by reducing scarcity.
Also, to the extent that the distribution of humanitarian food aid
reduces the actual grievances of population mostly at risk of fight-
ing, the conflict-reducing effect may be increased (e.g. Messer &
Cohen, 2004). In addition, the provision of humanitarian food assis-
tance in collaboration with domestic governments may win over
the ‘‘hearts and minds” of local populations and/or be seen as a
reward for sharing critical information that could result in a
reduced probability of conflict or conflict duration (Zurcher, 2017).
However, others have argued that humanitarian food aid
instead promotes conflict. Food assistance can be hijacked by
rebels allowing them to fight longer (Messer, Cohen, &
Marchione, 2000; Hendrix & Brinkman, 2013; Nunn & Qian,
2014). There is indeed anecdotal evidence supporting this channel
not clear how much of humanitarian food aid is sensitive to theft.
In addition, the information sharing resulting from the delivery of
humanitarian food aid can result in insurgents targeting local pop-
ulations to deter them from further engaging with the government,
and thus in additional deaths and injuries (Zurcher, 2017). Also, if

https://doi.org/10.1016/j.worlddev.2019.104713
0305-750X/Ó 2019 Elsevier Ltd. All rights reserved.
2 S. Mary, A.K. Mishra / World Development 126 (2020) 104713
humanitarian food assistance is politicized in the recipient country,
this may add to local resentment based on pre-existing perceptions
on unfairness and inequality (Messer, Cohen, & D’Costa, 1998).
Given the existence of multiple but opposite channels, it is analyt-
ically impossible to provide a clear conclusion on the net impact of
humanitarian food aid on civil conflict. It is also important to high-
light that these mechanisms are specific to each aid type, and that
they likely play out differently for other aid types (e.g. food aid and
medical humanitarian aid inflows).
More fundamentally, these conceptual mechanisms are also
dependent on the allocation mode of humanitarian food aid. Food
assistance is typically provided through two ways: in-kind aid
(shipping commodities produced in donors’ countries to countries
in need) and cash-based assistance. The latter includes local and
regional purchases, direct cash transfers, and food vouchers. His-
torically, most donors have largely, if not exclusively, relied on
in-kind aid, but in the mid/late 2000s, international food assistance
has shifted toward a combination of in-kind and cash-based
assistance3.
While both aid modalities, in-kind and cash-based, have advan-
tages and challenges4, studies comparing the effectiveness of food
aid modalities (in-kind versus cash-based transfers) have found that
all modalities support food security outcomes, but more interest-
ingly, that their effectiveness is on average similar (World Bank,
2016; GAO, 2016). However, most of these studies have evaluated
the impact of various aid modalities in non-emergency settings. A
notable exception is Tranchant et al. (2019) that provides an impact
evaluation of food assistance in conflict-affected areas of Mali. Poli-
cymakers cannot reform humanitarian food assistance on the basis
of anecdotal evidence without a more robust evidence base (nor,
candidly, should they believe that humanitarian food aid prevents
or mitigates conflict either).
The evidence base has remained scarce in the face of uncertain
and complex conceptual links. Recently, a much-publicized paper
by Nunn and Qian (2014) – NQ after that – has stirred controversy
because the authors find that an increase in US humanitarian food
aid increases the incidence and duration of civil conflicts in recip-
ient countries using a sample of 125 developing countries between
1971 and 2006. These results have been disputed by USAID (2014)
that shows that their results are sensitive to the exclusion of unre-
liable food aid data and the inclusion of additional controls. More
fundamentally, Christian and Barrett (2017) argue that the identi-
fication strategy in Nunn and Qian (2014) may be susceptible to
spurious trends that further cast doubt on NQ’s findings.
In a related literature on development aid, a systematic litera-
ture review by Zurcher (2017) suggests that overall humanitarian
aid increases violence. Yet, a careful re-examination of the studies
sampled in this literature review may not support such conclusion.
For example, the interpretation of the econometric results is inva-
lid in Narang (2014, 2015) leading the author to misconclude about
the negative effects of humanitarian aid (see Mary, 2019). In par-
ticular, the author fails to interpret the interaction term as a ratio
of hazard ratios. In addition, Mary (2019) shows that conclusions
3
For instance, about 45 per cent of humanitarian food assistance from the
European Commission was provided in cash or vouchers in 2018. In a more dramatic
fashion, since 2008, Canada exclusively relies on cash-based assistance for food aid.
However, it is noteworthy that the US, the largest donor of food aid, still relies
primarily on in-kind aid due to US law requirements (Casey, 2018).
4 show that when our baseline model does not include country-
In-kind aid is often criticized for its costs, delays, and potential disruptions to
local markets. Local and regional purchases are much quicker and arguably create
fewer side effects to local markets (Lentz et al., 2013), but are dependent on the
calorific and nutritional quality of the available food supply in the region or country.
Furthermore, in-kind aid and LRP may also be difficult to deliver when it is unsafe to
operate in conflict zones. On the contrary, cash transfers and food vouchers are
appropriate when there are security concerns about transporting in-kind aid, but they
can also be stolen or used by recipients to purchase non-food items in poorly
controlled settings (e.g. World Bank, 2016; GAO, 2016).
in Wood and Molfino (2015) and Wood and Sullivan (2014) may
be affected by the inclusion of year dummies that capture common
shocks. Finally, a few studies (e.g. De Ree & Nillesen, 2009; Nielsen,
2011) have examined the effect of total aid on conflict, but the use
of total aid, rather than sector aid, makes it difficult to isolate the
causal transmission channels (Zurcher, 2017).
Despite the political importance of this topic, little is known of
the effect of humanitarian food aid on civil conflict. This is partic-
ularly relevant given the Sustainable Development Goals (SDG) #2
‘‘end hunger, achieve food security and improved nutrition” and
#16 [promote] ‘‘peace, justice, and strong institutions.” If humani-
tarian food aid indeed aggravates conflict, food assistance policies
must be reformed to target only the environments meeting the
conditions where we can ensure the efficacy of humanitarian food
aid must be scaled down or ended (e.g. Lentz & Barrett, 2008).
Therefore, the objective of this study is to examine the causal
effect of humanitarian food aid on civil conflicts. To accomplish
this objective we use data from 2002 to 2017 for a sample of 79
recipient countries, from the Credit Reporting System (CRS) of
the Organization for Economic Development and Cooperation
(OECD), the Peace Research Institute Oslo (PRIO) and Uppsala Con-
flict Data Program (UCDP), and the World Development Indicators
(WDI) of the World Bank. Our identification strategy exploits
between-country humanitarian food aid displacements within
the sample (between the domestic country and the rest of the sam-
ple) and uses the latter to instrument domestic food aid per capita
using a two-stage least squares (2SLS) instrumental variables (IV)
estimation. These displacements result from the emergence of a
loud (i.e., attracting media and donors’ attention) food crisis hap-
pening in the rest of the world (Barrett, 2002). The validity of the
strategy relies on donors’ responses to loud emergencies and food
aid budget constraints (e.g. Barrett, 2002; Neumayer, 2005; World
Food Programme, 2005; United Nations, 2010; Kuhlgatz & Abdulai,
2012; Reuters, 2014; World Hunger, 2018). Our modeling specifi-
cation explicitly accounts for time-invariant omitted variables,
time fixed effects, and country-specific time trends. We also run
a battery of robustness analyses accounting for, among others,
country-specific linear and non-linear time trends and time-
varying omitted variables affecting civil conflict and humanitarian
food aid.
Our baseline 2SLS-IV estimates suggest that a 10 percent
increase in humanitarian food aid per capita results in a 0.2 per-
centage point decrease in conflict incidence. In other words,
humanitarian food aid saves lives in the developing world. Also,
we find that humanitarian food aid decreases the onset and dura-
tion of civil conflicts. Our results are important for policymakers in
that they confirm the use of humanitarian food aid as an effective
tool in promoting security in developing countries. To the extent
the pass-through between conflicts and food insecurity is strong
(FAO, Ifad, UNICEF, WFP and WHO, 2017), increasing humanitarian
food aid inflows would likely result in much-improved food secu-
rity, especially considering the direct effect of humanitarian food
aid on nutrition (e.g. Mary, Saravia-Matus, & Gomez y Paloma,
2018). Implicitly, our results point out the need to reallocate for-
eign sector aid towards humanitarian food aid if donors want to
achieve SDG#2 and SDG#16.
In addition, our results raise important concerns with respect to
the modeling of civil conflict in previous studies. In particular, we
specific time trends, we find that humanitarian food aid does not
affect civil conflict incidence. Researchers studying the links
between humanitarian food aid and conflict should, at the least,
test the robustness of their results to the inclusion of linear and
non-linear time trends as the latter seem to be a driving force
behind recent results (i.e., Christian & Barrett, 2017). Lastly, our
paper provides preliminary insights about the existence of
 S. Mary, A.K. Mishra / World Development 126 (2020) 104713
between-country humanitarian food aid displacements within
recipient countries and extends the empirical literature about
health aid displacements to a different aid type (e.g. Lordan, Ki
Tang, & Carmignani, 2011). In particular, we find that a 1 basis per-
centage point increase in the average share of humanitarian food
aid in the rest of the world (excluding the domestic country)
decreases humanitarian food aid per capita in the domestic coun-
try by about 0.7 USD.
The rest of the paper is structured as follows. Section 2
describes the empirical strategy. Section 3 presents the data and
descriptive statistics. Section 4 analyses the main results. Section 5
examines heterogeneous effects. Section 6 concludes.
2. Empirical strategy
To examine the impact of humanitarian food aid on conflict the
empirical model builds upon the linear framework that has been
used by Nunn and Qian (2014) and can be expressed as:
C it ¼ bAIDit þ cit X it þht þ di þ eit ð1Þ
AIDit ¼ aSjt þ lit X it þ pt þ qi þ
it ð2Þ
Eq. (1) is the second stage of our 2SLS system and Eq. (2) is the
first stage. The use of 2SLS estimation is justified by the existence
of reverse causality and simultaneity between conflict and human-
itarian food aid5. Note that we do not lag humanitarian food aid in
our model. Lagging a suspected endogenous explanatory variable is
common practice in applied research to avoid potential endogeneity
biases. However, Reed (2015) and Bellemare, Maski, and Pepinsky
(2017) find that this practice may result in inconsistent estimates
and misleading inference. More fundamentally, the use of contem-
poraneous aid is in line with the belief that humanitarian food aid
will have a within-year effect (e.g. Lentz, Passarelli, & Barrett, 2013)6.
Throughout Eqs. (1) and (2), countries are indexed by i and year
is indexed by t; C it is the existence of a civil conflict in country i in
year t; AIDit is humanitarian food aid per capita in US dollars; di (qi Þ
are country-specific (time-constant unobserved) fixed effects con-
trolling for time invariant heterogeneity. ht ðpt ) are time fixed
effects controlling for common shocks that affect both conflict
and humanitarian food aid, such as global or region-wide business
cycle, food price or natural resource price shocks. The instrument
Sjt is the share of humanitarian food aid out of total aid averaged
across all sampled countries other than country i. The share is cal-
culated on the monetary amount of humanitarian food aid, rather
than on its quantity.
Before we discuss the identification strategy, it is important to
highlight that NQ (2014) has come under scrutiny because their
identification strategy relying on a continuous difference-in-
difference estimation with Bartik-shift instruments may be suscep-
tible to spurious nonlinear trends (Christian & Barrett, 2017). To
prevent trend effects to bias our estimates, we add country-
specific time trends in Eqs. (1) and (2):
C it ¼ bAIDit þ cit X it þht þ uit þ di þ eit ð3Þ
AIDit ¼ aSjt þ lit X it þ pt þ xit þ qi þ
it ð4Þ
where uit and xit are country-specific time trends.
5
In this paper, we cannot address questions with respect to the modality
composition of food aid over time, and thus implicitly assume that the effectiveness
of a 1$ of humanitarian food aid per capita is constant throughout the study period
and is not affected by its modality structure.
6 7
We run additional estimations replacing current aid with one-period lagged aid
and find no significant effect, which supports the argument that the effect of
humanitarian food aid does not last past a year in our sample (see column 4 Table A3).
3
The importance of controlling for country-specific time trends
(as well as time fixed effects) has already been highlighted in the
climate-conflict literature (see Burke et al., 2009; Buhaug, 2010;
Couttenier & Soubeyran, 2014; Hsiang & Meng, 2014). Country-
specific time trends7 have been used to control for unobserved
time-varying heterogeneity, that is, variables that could be evolving
over time (such as economic performance or political institutions)
and altering conflict risk (e.g. Burke et al., 2009). In contrast, a com-
mon, rather than country-specific, time trend requires the assump-
tion that all countries are comparable over time in the factors that
influence conflict risk, e.g. geopolitics, natural resources, colonial his-
tory, international trade patterns, and geographic constraints
(Hsiang & Meng, 2014). Intuitively, there is no reason to believe that
these factors would be the same across countries. Hsiang and Meng
(2014) suggest using a F-statistic test that jointly tests whether
trends in conflict are statistically different from zero, and thus coun-
tries have the same trends in conflict.
Furthermore, in line with Christian and Barrett (2017), we check
for the presence of nonlinear country-specific trends in conflict.
The visual inspection of conflict and aid variables, the Akaike Infor-
mation Criterion (AIC), and a F-statistic test can help inform model
selection, i.e. which trend specification best fits the data, and all
suggest that country-specific cubic time trends should be included
in our baseline specification. This is in line with the conflict litera-
ture that has shown the importance of modeling time dependence
in binary (conflict) data to the cubic degree (e.g. Carter & Signorino,
2010).
X it is a vector of independent variables that includes food aid
that is not humanitarian food aid, the logarithm of non-food for-
eign aid (defined as total aid minus humanitarian food aid and
other food aid), the logarithm of GDP per capita (based on Purchas-
ing Power Parity), the logarithm of inflation rate, and a set of cli-
mate variables controlling for the average temperature and
precipitation in each month of the year. These monthly variables
capture the fact that different countries have different crop-
specific growing seasons and different sensitivities to weather
variations (Nunn & Qian, 2014). We also include two weighted
averages, respectively, of humanitarian food aid and conflict inci-
dence in neighboring countries to account for potential spillover
effects.
The choice of entering specific variables in levels or logarithms
is driven our intuition behind what the relationship is between,
let’s say, non-food aid and conflict risk. In particular, we believe
that non-food aid has marginally increasing/decreasing effects on
conflict risk. Such modeling for non-food aid has been used in
the literature (e.g. Dube & Naidu, 2015). We also test whether log-
ging the aid variables or using the levels of other non-food aid
affects the results in robustness analyses and we confirm the base-
line results using these alternative models (see Table A3).
We also include an indicator of the level of ethnic tensions and
an indicator of the political regime. The inclusion of these variables
is common in the conflict literature (e.g. Burke et al., 2009). One
concern about the inclusion of the vector X it relates to the issue
of bad controls caused by the inclusion of potentially endogenous
and proxy regressors (Angrist & Pischke, 2008), so we estimate par-
simonious models in robustness analyses to examine the extent to
which bad controls potentially affect inference.
The identification strategy exploits cross-sectional and time
variation in between-country humanitarian food aid displace-
ments between country i and all other recipient countries included
in the sample. We borrow from the health economics literature to
design our IV strategy (e.g. Lordan et al., 2011). In our context,
In practice, country-specific time trends are constructed by interacting the time
variable (‘‘year”) with each country (via the country-specific fixed effect). This allows
each country to have their own time trend.
4 S. Mary, A.K. Mishra / World Development 126 (2020) 104713
these displacements result from the emergence of a (relatively)
loud(er) food crisis happening in the rest of the world (Barrett,
2002). The validity of the strategy relies on two main elements:
donors’ responses to loud emergencies and food aid budget con-
straints. First, there is evidence that donors respond to media cov-
erage (Kuhlgatz & Abdulai, 2012). As the media coverage of a food
crisis is amplified in a particular country, donors will redirect and
refocus their efforts towards this area, while humanitarian food aid
for the country in a ‘silent’ emergency or in a situation that is rel-
atively less strategic in terms of the donors’ interests might receive
less humanitarian food aid (Barrett, 2002; Neumayer, 2005). Sec-
ond, donors’ budgets for food aid may not be increased during
the fiscal or calendar year, therefore putting more stress on the
inter-country competitive use and (re-)allocation of resources,
especially in a context where donors are slow to pledge and deliver
humanitarian food aid (Kuhlgatz & Abdulai, 2012). There are
indeed numerous and recurrent policy and media reports about
the World Food Programme had to reduce or cut food aid in devel-
oping countries during the year because of a lack of funds (e.g.
World Food Programme, 2005; Reuters, 2014; United Nations,
2010; The Washington Post, 2018; World Hunger, 2018). In addi-
tion, the coordination between multiple donors of humanitarian
food aid responses may also exacerbate such between-country dis-
placements (Kuhlgatz, Abdulai, & Barrett, 2010). Also, Non-
Governmental Organizations and aid delivery organizations may
contribute to this phenomenon by redirecting aid funds to partic-
ularly distressed regions.
Conditional on country-specific time trends, country fixed
effects, time fixed effects, and the vector of controls X it , the
instrument Sjt that is defined as the humanitarian food aid
funding received by all countries other than i, (AIDt  AIDit ), to total
aid funding received by all countries other than i,
(TOTAIDt  TOTAIDit ), averaged across all sampled countries other
than i, represents the donors’ response to the occurrence of a (rela-
tively) loud(er) food emergency in the rest of the world. a in Eq. (4)
captures the between-country displacement effect, or in other
words, translate the redirection of humanitarian food aid resources
towards the rest of the world, but away from country i. Thus we
expect a to be negative, especially in a context of limited budget
constraints, donors’ fatigue, or long funding authorization and allo-
cation procedures.
What about b? Nunn and Qian (2014) suggest that increased
humanitarian food aid may cause increased civil conflict risks
because humanitarian food aid is vulnerable to theft allowing
rebels to fight longer. However, Blattman and Miguel (2010) high-
light exogenous sources of income growth are generally associated
with lower conflict risks, thus making this channel unlikely. In
addition, while some humanitarian food aid may be captured by
rebels, it is equally likely that most humanitarian food aid ends
up being delivered and arguably reduces conflict risks (through
reduced starvation), so we could now expect b to be negative in
such context. On the whole, it is not analytically clear what to
expect for b.
Our strategy to causal identification depends on the credibility
of the exclusion restriction behind the IV strategy, which is that the
instrument, conditional on the set of fixed effects, time trends, and
controls, included in Eqs. (3) and (4), is correlated with conflict in
country i only through the domestic humanitarian food aid chan-
nel. A potential threat to our identification strategy exists if the
instrument influences domestic conflict through other channels
that are not controlled for. For example, the potential redirection
of food aid resources away from the domestic country may also
affect other types of aid that are related to conflict risks (e.g.
Gupta et al., 2018). Equally, the redirection of aid may affect eco-
nomic growth and/or governance that have been both linked to
conflict. In a related manner, humanitarian food aid in neighboring
countries could have spillover effects in the domestic country via
cross-border trade, while civil conflicts in neighboring countries
could have spillover effects via migration. To control for these
channels, we include a number of covariates, including other
non-food aid, other food aid, GDP per capita, an indicator of the
political regime, as well as two weighted averages, respectively,
of conflict incidence8 and humanitarian food aid in neighboring
countries (based on the relative border’s length of a recipient coun-
try with its neighbors).
Furthermore, the instrument may lead to changes in food prices
that may contribute to social unrest and conflict risks (Bellemare,
2015). However, we include time fixed effects controlling for global
or region-wide economic shocks affecting all countries in the
region, lessening these concerns. We also include the inflation rate
to control for the fact that these price shocks may be domestic, as
well as region-wide. Moreover, it is possible that the instrument
affects the country’s reliance on humanitarian food aid, but again
this is likely to be mostly controlled via the country fixed effects
and the time trends. Still, to be sure of the validity of the identifi-
cation strategy, we provide an additional analysis in which we
include lagged humanitarian food aid to account for time-varying
heterogeneity and aid inertia.
A more serious problem to our strategy may come from the fact
that conflict in a large country may create substantial migratory
flows that result in a loud emergency crisis in other countries. In
other words, there could be reverse causality between our depen-
dent variable in Eq. (3) and the instrument in Eq. (4). To double-
check the validity of our approach, we drop the five largest popu-
lated countries of our sample, i.e., China, India, Indonesia, Nigeria,
Pakistan, in sensitivity analyses, and confirm this does not affect
our baseline estimates. Another threat to our identification strat-
egy might come from potential reverse causality between the
dependent variable AIDit and the instrument Sjt in Eq. (4). However,
an endogeneity test fails to reject the null that the instrument can
be treated as exogenous in Eq. (4), thus indicating this is not a
problem here.
The approach also relies on the fact that humanitarian food aid
increases in the rest of world due to a food crisis, but not because
food aid directed at the country i has decreased following changes
in conflict risks in this country. In other words, conflict in country i
should not affect the instrument. We run additional estimations
and indeed show that conflict in country i does not affect the
instrument, thus providing supporting evidence for this assumption.
Moreover, if donors increase overall aid as a response to con-
flict, the IV estimate of humanitarian food aid on conflict will be
biased upwards. This is not a serious concern though, because it
means our baseline estimate may be viewed as a lower bound to
the positive effect of food aid. Nonetheless, we can examine this
problem through a robustness analysis where data, when total
aid per capita changes by more than 10 per cent (decrease and
increase), is excluded.
Additionally, we implement a test in support of the validity of
the exclusion restriction following Brückner (2013). We can intro-
duce the instrument Sjt as a right hand side variable in Eq. (3) and
use six-period lagged humanitarian food aid as a new instrument
in Eq. (4). If the coefficient associated to Sjt in the second stage is
statistically significant, this may indicate the violation of the exclu-
sion restriction. We also perform several sensitivity analyses while
8
The weighted average is linked to the dependent variable. For example, if we use
the number of battle-related deaths as dependent variable in the structural model, we
include the weighted average of the number of battle-related deaths in neighboring
countries.
 S. Mary, A.K. Mishra / World Development 126 (2020) 104713
relaxing the exclusion restriction using the Union of Confidence
Intervals approach designed by Conley, Hansen, and Rossi (2012).
Another concern potentially affecting our estimates of Eq. (3) is
that the latter do not fully account for time-varying heterogeneity.
Arguably, country-specific time trends capture some of this hetero-
geneity (Burke et al., 2009), but it is likely some may be left out. To
account for this problem, we estimate a model with lagged conflict.
The use of a lagged dependent variable explicitly controls for time-
varying omitted variable biases (Mary et al., 2018). If these results
remain in line with our static estimates of Eq. (3), time-varying
omitted variables biases are likely unimportant in driving our
baseline results.
Through the estimations, the strength of instruments is
examined by reporting the Kleibergen-Paap rk Wald F statistic
(or F-statistic) that we compare with the critical values from
Stock and Yogo (2005) for testing weak instruments. We test the
null hypothesis that the maximum size distortion is greater than
10, 15 or 20 percent and (if needed) the null that the relative bias
is larger than 10, 15, or 20 percent. We report inference robust to
weak instruments, based on the Stock-Wright S Lagrange
Multiplier (LM) statistic (Pflueger & Wang, 2015). Following
Young (2018), we test the inference relative to our main results
to bootstrapping and proceed to leave-one (cluster)-out tests to
examine the robustness of our panel IV regressions.
Last, it is important to note that 2SLS estimators typically apply
when both the dependent variables in the structural model (Eq.
(3)) and the reduced form for the endogenous regressor (Eq. (4))
are continuously distributed. This is neither the case here because
conflict incidence is binary and humanitarian food aid cannot be
negative. Ideally, one would like to extend the 2SLS reasoning to
nonlinear models. Unfortunately, there is no commonly accepted
and easily adaptable approach that allows doing so, especially
where the first stage would be a tobit model and the second-
stage a probit model. Trying to extend such reasoning often results
in (infamous) forbidden regressions. Thus, the most often applied
approach has been to use linear 2SLS estimation. This choice is sup-
ported by evidence that show little differences between partial
effects from more plausibly correct nonlinear models to partial
effects from linear models (e.g. Papke & Wooldridge, 2008).
More fundamentally, whether the dependent variable in the
second stage is binary, non-negative, or continuously distributed,
the 2SLS-IV approach captures the local average treatment effect,
that is, in this context, the effect of humanitarian food aid for coun-
tries receiving such aid, which is the main question of interest in
our paper (Angrist & Pischke, 2008). There are, of course, more
complex alternatives to estimating binary models with endoge-
nous regressors such as control function (CF) methods or maxi-
mum likelihood (ML) estimators. The former is typically used
when the endogenous variable is continuously distributed while
the latter is used when the endogenous regressor is binary. How-
ever, neither CF methods nor ML estimators can accommodate lim-
ited regressors. Both CF and ML approaches also come at the cost of
additional assumptions with respect to the joint distribution of
errors in the structural model and the reduced form for the
endogenous regressor. Moreover, it is noteworthy that, unlike CF
or ML estimators9 that become generally inconsistent if an appropri-
ate instrument is not included, the 2SLS estimator only loses
efficiency (Baum, Dong, Lewbel, & Yang, 2012).
Given these difficulties, one might be tempted to change the
dependent variable and/or the independent variable of interest
9
Furthermore, these alternatives rarely allow for the inclusion of fixed effects. A
potential exception is the approach developed by Papke and Wooldridge (2008), but
the approach only fits the case where the endogenous regressor is continuous. Also,
these approaches do not converge under more advanced model specifications, i.e.
when we include country-specific nonlinear time trends.
5
(humanitarian food aid) to make the estimation more fitting to
the 2SLS framework. For instance, we could use the number of
battle-related deaths or civilian casualties for the dependent vari-
able (in levels or logs), and/or use the logged value of aid. Note,
though, both changes would represent a change in the functional
form of the structural model, and the choice of the new baseline
model would then have to be informed by AIC and prior beliefs.
More importantly, while this may lessen some of the issues for
the estimations discussed above, this would also bring about
another new set of issues, related to the zero values in both the
dependent and the aid variables. If we log variables whose values
often take 0, we risk losing a lot of information. The traditional
solution to this problem has been to use a transformation such
as lnðAID þ xÞ with x being a small number. As a way to test the
robustness our results, we also provide these alternative estima-
tions in robustness analyses (see Table 3 and A3).
3. Data and descriptive statistics
The sample is composed of 79 humanitarian food aid recipient
countries. Data are collected for the period 2002–2017. The avail-
ability and quality of data restricts the selection of countries and
time coverage. Especially, data on sector foreign aid before 2002
is not suitable for empirical analysis (Mary et al., 2018)10. The full
list of countries can be found in the Appendix while descriptive
statistics can be found in Table 1.
The indicator of civil conflict incidence is from the PRIO/UCDP
database (version 19.1) (Gleditsch, Wallensteen, Eriksson,
Sollenberg, & Strand, 2002; Pettersson et al., 2019), where civil
conflict (or state-based armed conflict) is defined as ‘‘a contested
incompatibility that concerns government and/or territory where
the use of armed force between two parties, of which at least
one is the government of a state, results in at least 25 battle-
related deaths in a calendar year.” (Pettersson, 2019). In practice,
civil conflict is captured by a binary variable taking the value 1 if
the conflict has resulted in more than 25 battle-related deaths in
year t, 0 otherwise. We also use several alternative conflict vari-
ables in robustness analyses, especially a binary variable for the
onset11 of civil conflicts. We also test the conflict incidence for minor
and major civil conflicts. A minor conflict is defined by the number of
battle-related deaths being between 25 and 999. If a conflict results
in more than 999 battle-related deaths, the conflict is a major con-
flict. We also run additional estimations where the dependent vari-
able is the number of battle-related deaths and the number of
civilian casualties12. Battle-related deaths refer to those deaths
caused by the warring parties that can be directly related to combat
during an armed conflict, but exclude deaths due to disease and star-
vation, criminality, or attacks deliberately directed against civilians
only. Civilian fatalities result from the use of armed force by the gov-
ernment of a state against civilians which results in at least 25
deaths.
Non-humanitarian food aid is defined as the supply of edible
human food under national or international programmes including
transport costs; cash payments made for food supplies; project
food aid and food aid for market sales (CRS purpose code 52010).
Unlike humanitarian food aid, it typically targets hunger and food
insecurity in the medium run. Other aid is defined as total aid
minus food aid and humanitarian food aid. Aid inflows are gross
disbursements in constant 2017 USD (OECD, 2019).
10
We do not rely on FAO data, unlike Nunn and Qian (2014), because our IV
approach requires total aid which is not available from the FAO dataset. Moreover,
using the same database has the advantage of consistency across aid measurements.
11
The onset (offset) indicator takes the value of 1 in the year where the conflict
started (ended), 0 otherwise.
12
In both cases, we use the best estimates provided by the UCDP databases.
6 S. Mary, A.K. Mishra / World Development 126 (2020) 104713

Table 1
Descriptive statistics of the baseline estimation sample (N = 1158).

Mean Std. deviation Min. Max.

Variables (1) (2) (3) (4)
Civil conflict – more than 25 battle-related deaths 0.194 0.396 0 1
Civil conflict – more than 25 battle-related deaths but less than 1000 0.180 0.385 0 1
Civil conflict – more than 999 battle-related deaths 0.0259 0.159 0 1
Onset of civil conflict – more than 25 battle-related deaths 0.0501 0.218 0 1
Offset of civil conflict – more than 25 battle-related deaths 0.0415 0.199 0 1
Number of battle-related deaths due to conflicts 120.20 577.20 0.00 10,165
Number of battle-related deaths due to conflicts, log 0.778 3.140 2.303 9.227
Number of civilian deaths due to government attacks 15.06 118.50 0.00 2,595
Number of civilian deaths due to government attacks, log 1.797 1.819 2.303 7.861
Weighted average of conflict incidence (more than 25 battle-related deaths) in neighboring countries 0.199 0.259 0 1
Weighted average of conflict incidence (between 25 and 999 battle-related deaths) in neighboring countries 0.176 0.251 0 1
Weighted average of conflict incidence (more than 999 battle-related deaths) in neighboring countries 0.0318 0.116 0 1
Weighted average of number of battle-related deaths in neighboring countries 313.9 2,649 0 57,160
Weighted average of number of civilian fatalities in neighboring countries 19.51 104.6 0 2,553
Humanitarian food aid per capita, constant 2017 USD 0.986 3.117 0 49.37
Other non-food aid per capita, constant 2017 USD 49.58 55.26 0.92 455.90
Other non-food aid per capita, log 3.330 1.198 0.0825 6.122
Food aid per capita, constant 2017 USD 0.833 1.645 0 27.35
Weighted average of per capita humanitarian food aid in neighboring countries 1.166 3.004 0 43.92
Share of humanitarian food aid out of total aid, rest of sample 2.262 0.340 1.520 2.938
GDP per capita, PPP (constant 2011 international $) 7,425 5,856 545.3 29,494
GDP per capita, log 8.540 0.936 6.301 10.29
Inflation, GDP deflator (annual %) 9.317 11.22 0.0195 197.0
Inflation, log 1.792 1.013 3.937 5.283
Polity 2 score 3.606 5.289 9 10
Ethnic tensions 0.379 0.207 0 1
Monthly average precipitation
January 77.55 87.21 0 462.3
February 70.96 78.60 0 400
March 79.19 78.07 0 443.1
April 86.86 79.18 0 419.9
May 109.9 105.8 0 619
June 119.2 121.8 0.100 639.7
July 129.5 134.5 0 849.7
August 135.9 141.5 0 944.6
September 128.0 126.6 0 880.4
October 119.5 118.6 0.100 1,070
November 92.53 94.50 0 528.8
December 85.15 90.13 0 445.6
Monthly average temperature
January 18.74 10.48 25.70 28.70
February 19.83 10.02 24.30 30.60
March 21.56 8.303 11.50 33
April 23.01 6.466 1.600 33.60
May 23.77 5.528 5.100 34.50
June 24.04 5.302 3.100 34.70
July 24.11 5.139 3.600 35.30
August 24.25 4.605 3.600 35
September 23.84 4.505 6 31.70
October 22.90 5.868 0.900 30.60
November 21.01 7.865 12.70 29.30
December 19.25 9.739 21.10 28.60

GDP per capita (in 2011 constant dollars, purchasing power par-
ity), total population and the inflation rate are taken from the WDI
database (WDI, 2019). The index on the level of ethnic tensions
captures the existence of tensions attributable to racial, nationality
or language divisions in a country. We rescale an index taken from
the International Country Risk Guide from the Political Risk Ser-
vices from 0 to 1 (ICRG, 2018), with 0 (1) indicating the lowest
(highest) level of ethnic tensions. Ethnic divides have long been
suspected to be a cause of conflict (e.g. Alesina, Devleeschauwer,
Easterly, Kurlat, & Wacziarg, 2003).
The Polity2 score is taken from the Polity IV database from the
Center for Systemic Peace. The variable captures the type and
quality of political regime and has been previously used in the
literature (e.g. Burke et al., 2009). The monthly values for
average temperature and precipitation are taken from the
Climate Research Unit database of the University of East Anglia
(version TS v 4.01) (CRU, 2019). Temperature data are expressed
in degree Celsius (°C) while precipitation data are expressed in
millimeters (mm).
Fig. 1 shows the humanitarian food aid and civil conflict trends
over the period 2002–2017 for the sample. It is clear that the aver-
age incidence of civil conflict has been overall increasing since
2012, while it had fallen down significantly between 2007 and
2012. Interestingly, the increasing conflict trend seems to coincide
with several years of decreased humanitarian aid inflows (though
with some delay).
 S. Mary, A.K. Mishra / World Development 126 (2020) 104713 7

4. Results Table 2
The effect of humanitarian food aid on civil conflict incidence.

4.1. Baseline estimates OLS-FE 2SLS-IV

First stage Second stage
Table 2 displays the Ordinary Least Squares (OLS) results of Eq. Dependent variable Conflict HFA Conflict
(3) in column 1 and the 2SLS-IV estimation results of the system of (1) (2) (3)
Eqs. (3) and (4) in columns 2 (first stage) and 3 (second stage). 13As Humanitarian food aid 0.008*** 0.018**
explained in Section 2, our baseline modeling specification includes per capita (HFA)
country and time fixed effects, country-specific cubic time trends as (0.003) [0.026]
Other non-food aid per capita, log 0.011 0.634** 0.010
well as a set of monthly weather variables and other covariates. Col- (0.023) (0.281) (0.023)
umn 1 Table 2 suggests that humanitarian food aid is associated with Food aid per capita 0.016# 0.099 0.017#
civil conflict. The coefficient is negative (0.008) and statistically (0.011) (0.169) (0.011)
significant, suggesting that humanitarian food aid reduces civil con- Ethnic tensions 0.662** 0.031 0.664**
(0.290) (1.161) (0.293)
flict incidence. This estimate is, however, likely biased because of
Polity 2 score 0.013# 0.023 0.014#
reverse causality and simultaneity, so we use 2SLS-IV estimation (0.008) (0.043) (0.008)
to control for endogeneity. GDP per capita, log 0.616** 2.054# 0.701***
The first stage of the 2SLS-IV estimation is displayed in col- (0.276) (1.284) (0.268)
umn 2 Table 2. The instrument is clearly correlated with human- Inflation, log 0.014 0.032 0.014
(0.011) (0.076) (0.011)
itarian food aid. The displacement coefficient is largely negative HFA in neighbor countries, 0.003 0.124 0.001
(71.029) and statistically significant. This confirms the exis- weighted average
tence of between-country aid displacements in the context of (0.008) (0.096) (0.009)
humanitarian food aid. In the face of a new food crisis, donors Conflict in neighbor countries, 0.103* 0.223 0.100*
weighted average
reallocate funds away from the domestic country to another
(0.059) (0.289) (0.060)
country (or area). As the reallocation takes place, the share of Displacement (instrument) 71.029***
humanitarian food aid out of total aid averaged across all other (20.359)
recipient countries goes up. These displacements are economi- Observations 1,158 1,158 1,158
cally large as a 1 basis percentage point increase in the average Number of countries 79 79 79
share of humanitarian food aid for the rest of the sample would Country FE YES YES YES
result approximately in a 0.7 USD decrease in domestic human- Year FE YES YES YES
Country-specific time trends CUBIC CUBIC CUBIC
itarian food aid per capita (or about 72 percent of the mean Weather controls YES YES YES
humanitarian food aid per capita). While this is the first study, First-stage, F-stat n.a. n.a. 12.17
to the best of our knowledge, to explicitly model such displace-
Notes: Robust country-clustered standard errors in parentheses ***p < 0.01,
ments in the humanitarian food aid context, there is some liter- **p < 0.05, *p < 0.10, # p < 0.15. Stock-Wright LM p-values in square brackets. n.a.:
ature that has highlighted that foreign aid targeting HIV/AIDS not applicable.
may have displaced aid for other health concerns (e.g. Lordan
et al., 2011).
Column 3 shows the second stage of the 2SLS-IV estimation and
the estimate for humanitarian food aid is negative (-0.018), statis-
tically significant, and larger in absolute values than the OLS esti- aid while the period covered in this paper has seen dramatic
mate. This implies that humanitarian food aid is even more so changes in the modality structure of humanitarian food aid
effective at mitigating civil conflict. A 10 percent increase in (towards cash and vouchers). Moreover, we use humanitarian food
humanitarian food aid per capita would decrease the incidence of aid data from the OECD over a much shorter and recent period. In a
civil conflict by 0.2 percentage point or by 0.9 percent14. Following related literature on humanitarian aid, our results somewhat relate
Young (2018)’s recommendation, note that the estimate remains sig- to Gupta et al. (2018) who find that US health aid contributes to
nificant (p-value 0.09), even when we rely on bootstrapping and security in recipient countries.
check the robustness of the inference by using leave-one (cluster)- Furthermore, our OLS estimate is higher than our 2SLS estimate
out tests. suggesting that the reverse causality is positive. This is in line with
As explained in Section 2, humanitarian food aid may have neg- the stated objective of the donors, since humanitarian food aid is
ative and positive impacts (e.g. sabotage, theft and predation ver- intended to respond to conflict situations. While we acknowledge
sus direct nutrition, reduction in grievances and information- the differences in the study period and design, this is also in con-
sharing), but our findings suggest that on average the positive trast with NQ (2014). Next, we can reject the null hypothesis that
effects dominate the negative ones. the maximum size distortion is greater than 20 percent as the
The effect of humanitarian food aid is economically substantial, Kleibergen-Paap statistic is 12.17, well above the 20 percent
and is, to some extent, in contrast with the findings of NQ (2014) Stock-Yogo critical value (6.66). F-tests fail to reject the (joint
who find that US food aid promotes conflict in recipient countries. and individual) relevance or inclusion of time fixed effects (p-
However, it is important to highlight that our paper is quite differ- value: 0.000) and country-specific cubic time trends (p.value:
ent from theirs. First, our model is different in that we explicitly 0.000). Last, we can briefly discuss other independent variables.
model country-specific time trends. Also, NQ (2014) estimate the Economic growth is associated with decreased conflict incidence.
effect of US food aid using a sample of 125 developing countries, Food aid per capita and conflict in neighboring countries would
and food aid data from the Food and Agriculture Organization over seem to decrease conflict incidence but both coefficients do not
a much longer period. It is important to note that most humanitar- pass Young’s test. The coefficient for ethnic tensions is somewhat
ian food aid during the period covered by NQ (2014) was in-kind surprisingly negative, indicating that increased tensions would
reduce the incidence of civil conflict. Perhaps, this result implies
that further tensions would only dampen the risk for conflict, given
13
Coefficients for weather variables are omitted from all tables for the sake of space. the already high level of tensions, or possibly that the relationship
14
This is calculated as: 0:098  0:018  0:0017. Then: ð0:0025=0:194Þ  100  0:91. is nonlinear.
8 S. Mary, A.K. Mishra / World Development 126 (2020) 104713
4.2. Test of exclusion restriction
The validity in the IV estimation in Table 2 relies on assuming
that the between-country ‘displacement’ instrument affects
domestic civil conflict only through domestic humanitarian food
aid per capita. Table A1 implements an intuitive check for the
validity of the exclusion restriction following Brückner (2013). Col-
umn 1 Table A1 first presents the direct impact of the ‘displace-
ment’ instrument on civil conflict. The OLS coefficient is positive
(1.273) and significant. This suggests that the redirection of food
assistance funds away from (towards) the rest of the world is asso-
ciated with lower (higher) conflict incidence in the domestic coun-
try. This implicitly confirms our previous results and the
dominance of the positive mechanisms with respect to the impact
of humanitarian food aid.
Column 2 introduces the ‘displacement’ instrument as a right-
hand side variable in Eq. (3) and uses instead six-period lagged
humanitarian food aid per capita15 to instrument current humani-
tarian food aid per capita. The ‘displacement’ coefficient is now neg-
ative, in comparison with the direct positive ‘displacement’ effect in
column 1 Table A1, and statistically insignificant. In other words, our
preferred instrument does not have any direct effects on conflict
other than through the humanitarian food aid channel. This intuitive
check, therefore, provides reassuring evidence for the exclusion
restriction. Let’s note that the coefficient for humanitarian food aid
per capita (0.032) is negative, twice the estimate in Table 2, and
statistically significant.
We also perform several sensitivity analyses while relaxing the
exclusion restriction using the Union of Confidence Intervals
approach designed by Conley et al. (2012). We use the support
restrictions ½2db; 2db where d takes the value 0 or 1. b is the
estimate of the effect of food aid on conflict. The 95 per cent
confidence intervals are respectively [0.027;0.007] and
[0.028;0.006] for d ¼ 0 andd ¼ 1. Given that this approach is
considered conservative, it is clear that the relaxation of the exclu-
sion restriction does not invalidate our inference.
4.3. Alternative specifications
Table A2 summarizes the 2SLS-IV estimation results across
several alternative specifications. First, column 1 displays the esti-
mation results of a model in which we exclude country-specific
time trends. The 2SLS coefficient for humanitarian food aid per
capita is now positive and insignificant. This implies that human-
itarian food aid has no effect on conflict incidence in the sample.
When we include country-specific linear time trends in column 2,
the estimate turns negative but is still insignificant. When we
include country-specific quadratic time trends in column 3
Table A2, the estimate is negative and is somewhat imprecise.
This sensitivity analysis somewhat resonates with Christian and
Barrett (2017).
Notably, the AIC clearly point towards the specification with
country-specific cubic time trends being the preferred model over
all specifications, as the AIC for our baseline model in column 3
Table 2 is 846 (against 73, 349, 596, respectively, for the
models without time trends, with linear time trends, and with
quadratic time trends). Overall, these findings illustrate the need
to account for country-specific time trends in modeling conflict,
as had been previously argued in previous studies (e.g. Burke
et al., 2009). It also sends a warning to future researchers in that
the validity of empirical results should be tested to the inclusion
of country-specific nonlinear time trends.
15
We use six-period lagged humanitarian food aid because estimations based on
earlier lags suffer from weak instruments.
Furthermore, one of the issues of our baseline specification is
that our results may be affected by the issue of bad controls
(Angrist & Pischke, 2008). We thus propose a couple of parsimo-
nious models in columns 4 and 5 Table 4. Column 4 excludes all
variables from the model but humanitarian food aid. Column 5
only excludes ethnic tensions and polity2 for which data are often
missing. Overall, these additional results support our baseline
results. For example, the coefficient for humanitarian food aid is
negative in column 4 (and a bit imprecise), though somewhat
lower than in column 3 Table 2. In particular, a 10 percent increase
in humanitarian food aid per capita would decrease the probability
of civil conflict approximately by slightly less than 0.1 percentage
point or by about 0.6 percent. Still, even if this estimate is lower
than our baseline estimate in Table 2, humanitarian food aid
appears as a ‘‘bargain” towards conflict mitigation in terms of
cost-effectiveness. Column 6 Table 4 controls for time-varying
heterogeneity in Eq. (3) by adding a lagged civil conflict variable.
Again, the coefficient for humanitarian food aid remains negative
and is statistically significant (0.013).
Table A3 provides alternative aid specifications. Column 1 logs
all aid variables in the model. Colum 2 uses the transformation
ln(AID + 0.1) to avoid the loss of information related to zero aid
values in the case of humanitarian food aid (and food aid). Column
3 uses the levels of aid variables in the model. Across these alter-
native aid specifications, columns 1–3 all suggest that humanitar-
ian food aid would decrease the incidence of civil conflict in the
sample. When we replace current humanitarian food aid by its
one-period lag, we find no effect of lagged humanitarian food aid
on civil conflict. This suggests that the effect of humanitarian food
aid is short-lived.
Next, column 5 weighs the humanitarian food aid variable by
the country’s relative land area. Conflict is likely to be located away
from the capital (Findley et al., 2011). Humanitarian food aid is tar-
geted at populations in conflict zones. However, humanitarian food
aid may not always and/or fully reach the areas most affected by
conflict (or at least not proportionally to the conflict intensity)
(e.g. Briggs, 2018). This arguably means that our estimated effect
of humanitarian food aid could be underestimating the true
effect16. The estimate in column 5 Table A3 is 0.052, and much lar-
ger than previously estimated in Table 2. This result suggests that
humanitarian food aid is likely to be more effective at mitigating
conflict than what our baseline results indicate.
Finally, column 1 Table A4 drops the five largest countries from
the sample. The estimate is virtually the same than in Table 2 col-
umn 3. Column 2 drops data where total aid responses to conflict
have been larger than 10 per cent. This allows checking whether
the potential endogeneity between overall aid budgets to conflict
affects our IV estimates. The IV estimate in column 2 (0.044) is
negative and much larger than in Table 2. As explained in the mod-
eling section, this is in line with our IV estimate in Table 2 being a
lower bound to the positive effect of humanitarian food aid
towards mitigating civil conflict.
4.4. Persistence in humanitarian food aid
Kuhlgatz and Abdulai (2012) show that humanitarian food aid
is relatively persistent. Our instrumentation strategy does not
explicitly account for aid inertia, though the country-specific time
trends likely account for some time-varying heterogeneity in Eq.
(4). To check whether our instrumentation strategy is affected by
16
To simplify the exposition, let’s assume that 10 per cent of humanitarian food aid
is given to non-conflict areas (rather than 0 per cent), the only way populations in
conflict areas will benefit from such humanitarian food aid is through indirect market
effects. The additional food in non-conflict areas lowers the overall food demand and
puts fewer pressures on food prices in other (conflict) areas.
 S. Mary, A.K. Mishra / World Development 126 (2020) 104713 9

Table 3
The effect of humanitarian food aid on civil conflict: alternative conflict outcomes.

2SLS-IV Weibull-IV
Second stage
Dependent variable Minor Major Ln (Battle-related deaths + 0.1) Ln (Civilian fatalities + 0.1) Onset conflict Offset conflict
(1) (2) (3) (4) (5) (6)
Humanitarian food aid per capita 0.015* 0.005* 0.133* 0.037 0.010# 1.096#
[0.062] [0.099] [0.062] [0.499] [0.102] (0.066)
Other non-food aid per capita, log 0.011 0.003 0.126 0.364* 0.006 0.706*
(0.025) (0.007) (0.176) (0.216) (0.026) (0.127)
Food aid per capita 0.020* 0.001 0.132* 0.032 0.008 1.061
(0.010) (0.003) (0.072) (0.029) (0.007) (0.149)
Ethnic tensions 0.562* 0.200 5.176*** 1.412 1.002*** 10.374**
(0.299) (0.151) (1.952) (1.230) (0.270) (12.31)
Polity 2 score 0.007 0.005 0.105# 0.081* 0.012 0.992
(0.009) (0.006) (0.065) (0.045) (0.009) (0.038)
GDP per capita, log 0.047 0.707*** 6.647** 7.437*** 0.765*** 0.675
(0.162) (0.233) (3.080) (1.351) (0.216) (0.261)
Inflation, log 0.013 0.002 0.090 0.066 0.017# 1.082
(0.012) (0.006) (0.075) (0.073) (0.012) (0.186)
HFA in neighbor countries (w.a.) 0.007 0.000 0.004 0.121 0.009 1.046
(0.010) (0.007) (0.071) (0.106) (0.010) (0.045)
Conflict in neighbor countries (w.a.) 0.072 0.021 0.000 0.000 0.111** 1.274
(0.057) (0.027) (0.000) (0.000) (0.051) (1.178)
Observations 1,158 1,158 1,158 1,158 1,158 934
Number of countries 79 79 79 79 79 79
Country FE YES YES YES YES YES n.a.
Year FE YES YES YES YES YES n.a.
Country-specific time trends CUBIC CUBIC CUBIC CUBIC CUBIC n.a.
Weather controls YES YES YES YES YES YES
First-stage, F-stat 12.25 12.49 12.65 12.36 12.17 n.a.

Notes: Robust country-clustered standard errors in parentheses*** p < 0.01, ** p < 0.05, * p < 0.10, # p < 0.15. The IV in column 6 relies on a control function approach. We
insert the residuals from estimating Eq. (4) into the Weibull model. The estimate in column 6 is a hazard ratio. W.a.: weighted average.

Table 4 slightly somewhat larger to that of column 2 Table 2. The estimate

The effect of humanitarian food aid on civil conflict: regional estimations. is statistically significant and negative (73.75). In addition, the
2SLS-IVSecond stage lagged aid coefficient is positive (0.247) and statistically signifi-
Africa Asia cant, suggesting moderate inertia. This is in line with previous
Conflict (1) (2) findings (e.g. Kuhlgatz & Abdulai, 2012). More importantly, column
Humanitarian food aid per capita (HFA) 0.030** 0.084**
2 Table A5 implies that humanitarian food aid decreases conflict
[0.012] [0.037] incidence, even we account explicitly for humanitarian aid inertia.
Other non-food aid per capita, log 0.023 0.035 The coefficient in column 2 is 0.013 and statistically significant,
(0.036) (0.033) implying relatively lesser effects of humanitarian food aid than
Food aid per capita 0.009 0.013
those estimated in Table 2.
(0.013) (0.011)
Ethnic tensions 0.791 1.219***
(0.557) (0.222)
Polity 2 score 0.006 0.013 5. Heterogeneous effects of humanitarian food aid
(0.016) (0.010)
GDP per capita, log 0.919*** 0.309
5.1. Alternative conflict indicators
(0.230) (0.237)
Inflation, log 0.016 0.034
(0.014) (0.029) Table 3 provides 2SLS-IV estimates using alternative civil con-
HFA in neighbor countries, weighted average 0.005 0.041 flict indicators. First, the coefficients for humanitarian food aid in
(0.010) (0.032) columns 1 and 2 remain negative and statistically significant. This
Conflict in neighbor countries, weighted average 0.221** 0.014
implies that humanitarian food aid reduces the incidence of small-
(0.100) (0.115)
Observations 504 232 scale and large-scale civil conflicts in the sample. Second, if we use
Number of countries 36 15 the logged number of battle related deaths as the dependent vari-
Country FE YES YES able17, we also confirm our baseline results. A 10 per cent increase in
Year FE YES YES
humanitarian food aid per capita would decrease the number of
Country-specific time trends CUBIC CUBIC
Weather controls YES YES
battle-related deaths by about 1.5 deaths.
First-stage, F-stat 90.92 12.60 If we use the logged number of civilian fatalities as dependent
variable in column 4, we find no effect of humanitarian food aid
Notes: Robust country-clustered standard errors in parentheses ***p < 0.01,
**p < 0.05, *p < 0.10, #p < 0.15.
per capita on the number of civilian fatalities, though the estimate
is positive and insignificant. Next, we also examine the impact of
humanitarian food aid on the onset and offset of conflicts. The esti-
aid inertia, we re-estimate the results in columns 2 and 3 Table 2, mate for the effect of humanitarian food aid on the onset of civil
but we marginally modify Eq. (4) by including one-period lagged conflicts is column 5 is negative, implying that humanitarian food
humanitarian food aid. Results are provided in Table A5.
The main pattern of results remains in line with Table 2. The 17
AIC suggest that the log-level specification is best when using the number of
estimated displacement effect in column 1 Table A5 is close but battle-related deaths or civilian casualties as dependent variables.
10 S. Mary, A.K. Mishra / World Development 126 (2020) 104713
aid is effective at preventing new civil conflicts. A 10 per cent
increase in per capita humanitarian food aid would decrease the
onset of civil conflict by 2.3 per cent.
To study the effect of humanitarian food aid on the duration of
conflicts, we follow previous studies that use survival models (e.g.
Narang, 2014). We use the control function approach used by Nunn
and Qian (2014) to account for potential endogeneity using the
‘displacement’ instrument. Column 6 Table 3 provides estimates
for the Weibull model (we find the latter is preferred to a Cox
model). However, the unusual negative AIC value suggests that this
framework may not be applicable to the outcome of interest and
these results should be taken with caution. This caveat aside, the
main message is that humanitarian food aid seems to be associated
with a shorter civil war survival, since the hazard ratio in column 6
is above 1. In other words, humanitarian food aid seems to reduce
the duration of civil conflicts.
5.2. Contextual factors
Table A6 examines whether the effect of humanitarian food aid
is heterogeneous across different contexts. Especially, Table A6
examines whether the impact of food aid on conflict depends on
the level of ethnic tensions in the recipient country (columns 1–2)
as might be implicitly suggested by previous studies (e.g.
Schleussner, Donges, Donner, & Schellnhuber, 2016), on the infla-
tion rate via food prices (columns 3–4) as suggested by
Bellemare (2015), and on the precipitation level (columns 5–6).
High/low is defined with respect to being above/below the mean
sample level of each mitigating variable.
For example, according to column 1 Table 3, when ethnic ten-
sions are higher than the mean sample level, humanitarian food
aid is found to have a statistically significant and large effect on
conflict reduction (coefficient: 0.033). However, when ethnic ten-
sions are lower, the effect of food aid on conflict subsides as the
coefficient is negative but statistically insignificant.
Further, the main message from columns 3–4 is that human-
itarian food aid largely reduces conflict incidence in situations of
low inflation; when inflation is high, food aid is found not to
affect conflict incidence. This is somewhat surprising, but argu-
ably this could be due to the fact that price-driven riots are an
urban phenomenon and might likely happen away from conflict
areas. Columns 5–6 suggest that humanitarian food aid is quite
effective at reducing conflict risks as precipitation levels are espe-
cially low. This is line with the scarcity channel discussed in the
introduction (e.g. Hendrix & Brinkman, 2013). When precipita-
tions are above the mean level (or in situations of relative food
abundance), the coefficient for humanitarian food aid is found
positive though insignificant. This somewhat relates to recent
studies linking food scarcity/abundance to conflict (e.g. Koren,
2018; Koren & Bagozzi, 2017).
5.3. Regional estimations
Last, Table 4 presents regional estimations for Africa and Asia,
two regions that have suffered a great deal because of hunger
and civil conflicts. When we limit the sample to African or Asian
countries, the coefficients for humanitarian food aid seem to be
much larger than estimated in Table 2.
For Africa (sub-Saharan Africa and Northern Africa), a 10 per
cent increase in humanitarian food aid per capita (about
0.15USD) in the region would result in a 2.7 per cent decrease in
conflict incidence. This is more than twice the relative effect esti-
mated for the entire sample. For Asia, a 10 per cent increase in
humanitarian food aid per capita (about 0.15USD) in the region
would result in a 0.3 per cent decrease in conflict incidence. We
are unable to provide estimations for other regions18. Overall,
Table 4 suggests that the humanitarian food aid is particularly effec-
tive in the African context.
6. Conclusions
Humanitarian food aid is designed to save lives and alleviate
human suffering in conflict situations. Within the Sustainable
Development Goals that have the objectives of ending hunger
before 2030 (SDG#2) and bringing peace and justice (SDG#16),
humanitarian food assistance may play a critical role in delivering
progress in developing countries. However, there have been grow-
ing concerns that humanitarian food aid may actually have
counter-intended effects by aggravating conflicts in recipient
countries. Given the importance of humanitarian food assistance
policies, this paper has revisited the effect of humanitarian food
aid on conflict in developing countries.
We find that, on average, an increase in humanitarian food aid is
associated with decreased incidence of civil conflicts in the sample.
Overall, the provision of humanitarian food aid in conflict situa-
tions seems to provide some solace to the populations that may
be more likely to join the rebellion and fight, and weaken the risk
of civil war. The positive effect of humanitarian food aid towards
conflict mitigation is economically large.
Our empirical results are also robust to the use of alternative
conflict indicators, in particular the onset and duration of conflicts.
These additional results convey much political importance and
support the sustained provision of food assistance in high-risk
countries as an effective tool for conflict prevention and mitigation.
We also find that humanitarian food aid is particularly effective
in situations of (likely) weather-related food scarcity and high eth-
nic tensions. We also find that humanitarian food aid is particularly
effective in Africa.
More fundamentally, we raise a major concern in empirical
studies about the modeling of conflict. In particular, we show that
the inclusion of country-specific nonlinear time trends is critical in
identifying the causal impact of humanitarian food aid on conflict.
This resonates with Christian and Barrett (2017) and Burke et al.
(2009) in somewhat related literature. We invite future researchers
to double-check the robustness of their empirical results to the
inclusion of country-specific time trends.
The implication of our findings for policymakers and develop-
ment practitioners is clear. Humanitarian food aid overall saves
lives in recipient countries, and as such, the continuation of
humanitarian food assistance will make the developing world,
especially Africa, safer, and by extension, less food insecure (see
FAO, Ifad, UNICEF, WFP and WHO. , 2017). At a global level, the sta-
bility of developing countries may contribute to security and lesser
terrorist threats worldwide (Piazza, 2008; Steinwand, 2015; Open
letter to US congress, 2017). On a more practical level, our findings
indicate the relatively good ‘value for money’ of humanitarian food
assistance towards conflict prevention and mitigation. Further-
more, the existence of heterogeneous effects offers some guidance
to better target the locations and times where the efficacy of
humanitarian food aid can be improved.
The SDG aims to end hunger and make the world a safer place.
However, our results are at odds with the reality of foreign aid sec-
toral allocation in recent years. Mary et al. (2018) show that the
share of humanitarian food aid has relatively decreased in the last
two decades. According to our findings, this may put at risk the
progress towards achieving the SDG #2 and #16. In addition, in
the US (the major humanitarian food aid donor), the Global Food
18
This is because we do not have enough countries per region or the estimations
suffer from weak instruments.
 S. Mary, A.K. Mishra / World Development 126 (2020) 104713
Security Act of 2016 recommends ‘‘reducing reliance upon emer-
gency food assistance” upon evidence that ‘‘insurgent groups can
exploit international food aid” (Global Food Security, 2016). Our
findings suggest these concerns are not warranted on aggregate.
Given that there are still millions of people in the world, who
suffer from conflicts and its awful consequences, e.g. injuries,
deaths, or extreme hunger, it is a moral imperative to continue
researching the effect of humanitarian aid. Studies adding more
(data) granularity as well as cross-national studies are equivalently
needed to capture the complex relationships between humanitar-
ian food assistance and civil conflict in the developing world and to
further inform policymakers.
Declaration of Competing Interest
The authors declare that they have no known competing finan-
cial interests or personal relationships that could have appeared
to influence the work reported in this paper.
Acknowledgement
The authors gratefully acknowledge the data assistance from
Marissa Sirico.
Appendix A. Supplementary data
Supplementary data to this article can be found online at
https://doi.org/10.1016/j.worlddev.2019.104713.
References
Alesina, A., Devleeschauwer, A., Easterly, W., Kurlat, S., & Wacziarg, R. (2003).
Fractionalization. Journal of Economic Growth, 8(2), 155–194.
Angrist, J. D., & Pischke, J. S. (2008). Mostly harmless econometrics: An empiricist’s
companion. Princeton, NJ: Princeton Univ. Press.
Barrett, C. (2002). Food aid effectiveness: It’s the targeting, stupid. Cornell University
Applied Economics and Management Working Paper No. 2002-43.
Baum, C. F., Dong, Y., Lewbel, A., & Yang, T. (2012). Binary choice models with
endogenous regressors. State Conference, 2012, San Diego http://fmwww.bc.edu/
EC-C/S2013/823/baum.san2012.pdf.
Bellemare, M. F. (2015). Rising food prices, food price volatility, and social unrest.
American Journal of Agricultural Economics, 97(1), 1–21.
Bellemare, M. F., Maski, T., & Pepinsky, T. B. (2017). Lagged explanatory variables
and the estimation of causal effect. Journal of Politics, 79(3), 949–963.
Blattman, C., & Miguel, E. (2010). Civil war. Journal of Econic Litwerature, 48(1), 3–57.
Briggs, R. C. (2018). Poor targeting: A gridded spatial analysis of the degree to which
aid reaches the poor in Africa. World Development, 103, 133–148.
Brückner, M. (2013). On the simultaneity problem in the aid and growth debate.
Journal of Applied Econometrics, 28(1), 126–150.
Buhaug, H. (2010). Climate not to blame for African civil wars. PNAS, 107,
16477–16482.
Burke, M., Miguel, E., Satyanath, S., Dykema, J. A., & Lobell, D. B. (2009). Warming
increases the risk of civil war in Africa. PNAS, 106(49), 20670–20674. https://
doi.org/10.1073/pnas.0907998106.
Carter, D. B., & Signorino, C. S. (2010). Back to the future: Modeling time dependence
in binary data. Political Analysis, 13(3), 271–292.
Casey, A., 2018. US International Food Assistance: An Overview, 2018.
Congressional Research Service Report R45422, December 6.
Center for Systemic Peace, Accessed 25.01.2019 https://www.systemicpeace.org/.
Christian, P., Barrett, C. (2017). Revisiting the effect of food aid on conflict: a
methodological caution, Policy Research Working Paper 8171, World Bank,
August. https://sites.tufts.edu/neudc2017/files/2017/10/paper_42.pdf.
CIA (2014). World Threat Assessment of the US Intelligence Community, Senate
Select Committee on Intelligence, statement for the record, James Clapper.
https://www.dni.gov/files/documents/Intelligence%20Reports/2014%20WWTA
%20%20SFR_SSCI_29_Jan.pdf.
Climate Research Unit (2019). University of East Anglia, Accessed 6.28.2019. http://
www.cru.uea.ac.uk/data.
Collier, P., & Hoeffler, A. (2004). Greed and grievance in civil war. In Oxford Economic
Papers 56 (pp. 563–595).
Collier, P., & Hoeffler, A. (1998). On economic causes of civil war 50, No. 4 (Oct.,
1998). In Oxford Economic Papers (pp. 563–573).
Conley, T., Hansen, C. B., & Rossi, P. E. (2012). Plausibly exogenous. The Review of
Economics and Statistics, 94(1), 260–272.
Couttenier, M., & Soubeyran, R. (2014). Drought and civil war in sub-Saharan Africa.
The Economic Journal, 124, 201–244.
11
De Ree, J., & Nillesen, E. (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, 301–313.
Dube, O., & Naidu, S. (2015). Bases, bullets, and ballots: The effect of US military aid
on political Conflict in Colombia. The Journal of Politics, 77(1), 249–267.
FAO, IFAD, UNICEF, WFP, WHO (2017). The State of Food Security and Nutrition in
the World 2018. Building resilience for peace and food security. Rome, FAO.
http://www.fao.org/state-of-food-security-nutrition/en/.
Findley, G. M., Powell, J., Strandow, D., & Tanner, J. (2011). The Localized Geography
of Foreign Aid: A New Dataset and Application to Violent Armed Conflict. World
Development, 39(11), 1995–2009.
GAO (2016). International cash-based food assistance, Report to the Chairman,
committee on foreign relations, US Senate, United States Government
Accountabiity Office, GAO-16-819, September.
Gleditsch, N. P., Wallensteen, P., Eriksson, M., Sollenberg, M., & Strand, H. (2002).
Armed conflict 1946–2001: A new dataset. Journal of Peace Research, 39(5),
615–637.
Global Food Security Act of 2016 (2016). US Congress. Public Law 114-195, July 20,
2016. https://www.congress.gov/114/plaws/publ195/PLAW-114publ195.pdf.
Gupta, V., Mason-Sharma, A., Lyon, Z. M., Orav, E. J., Jha, A. K., & Kerry, V. B. (2018).
Has development assistance for health facilitated the rise of more peaceful
societies in sub-Saharan Africa? Glob. Public Health. https://doi.org/10.1080/
17441692.2018.1449232.
Hendrix, C., & Brinkman, H. (2013). Food insecurity and conflict dynamics: Causal
linkages and complex feedbacks stability. International Journal of Security &
Development, 2(2), 1–18.
Hsiang, S. M., & Meng, K. C. (2014). Reconciling disagreement over climate–conflict
results in Africa. PNAS, 111, 2100–2103.
International Country Risk Guide, the Political Risk Services Group. https://www.
prsgroup.com/.
Koren, O., & Bagozzi, B. E. (2017). Living off the land: The connection between
cropland, food security, and violence against civilians. Journal of Peace Research,
54(3), 351–364.
Koren, O. (2018). Food abundance and violent conflict in Africa. American Journal of
Agricultural Economics.
Kuhlgatz, C., & Abdulai, A. (2012). Food aid and malnutrition in developing
countries: Evidence from global food aid allocation. Journal of Development
Studies, 48(12), 1765–1783.
Kuhlgatz, C., Abdulai, A., & Barrett, C. (2010). Food aid allocation policies:
Coordination and responsiveness to recipient country needs. Agricultural
Economics, 41(3–4), 319–327.
Lentz, E. C., Passarelli, S., & Barrett, C. B. (2013). The timeliness and cost-
effectiveness of the local and regional procurement of food aid. World
Development, 49(C), 9–18.
Lentz, E., & Barrett, C. (2008). Improving food aid: What reforms would yield the
highest payoffs? World Development, 36(7), 1152–1172.
Lordan, G., Ki Tang, K., & Carmignani, F. (2011). Has HIV/AIDS displaced other health
funding priorities? Evidence from a new dataset of development aid for health.
Social Science & Medicine, 73, 351–355.
Magone, C., Neuman, M., Weissman, F. (2012). Humanitarian Negotiations
Revealed: the MSF experience, 17 February 2012, https://www.msf-crash.org/
en/publications/humanitarian-negotiations-revealed-msf-experience.
Mary, S. (2019). Humanitarian aid does not create conflict. Working Paper, available
at https://sites.google.com/view/sebastienjmary/research/working-papers?
authuser=0.
Mary, S., Saravia-Matus, S., & Gomez y Paloma, S. (2018). Does nutrition-sensitive
aid reduce the prevalence of undernourishment? Food Policy, 74, 100–116.
Messer, E., Cohen, M.J. (2004). Breaking the links between conflict and hunger in
Africa, 2020 Africa Conference Brief 10, IFPRI Vision 2020.
Messer, E., Cohen, M.J., Marchione, T. (2000). ‘‘Conflict: A Cause and Effect of
Hunger.” Environmental Change and Security Report 7.
Messer, Ellen, Marc J. Cohen, and Jashinta D’Costa (1998). Food from Peace:
Breaking the Links Between Conflict and Hunger. Food, Agriculture, and the
Environment Discussion Paper No. 24. Washington: IFPRI.
Narang, N. (2014). Humanitarian assistance and the duration of peace after civil
war. The Journal of Politics, 76(2), 446–460.
Narang, N. (2015). Assisting uncertainty: How humanitarian aid can inadvertently
prolong civil war. International Studies Quarterly, 59, 184–195.
Neumayer, E. (2005). Is the allocation of food aid free from donor interest bias?
Journal of Development Studies, Taylor & Francis Journals, 41(3), 394–411.
Nielsen, R. A. (2011). Foreign aid shocks as a cause of violent armed conflict.
American Journal of Political Science, 55(2), 219–232.
Nunn, N., & Qian, N. (2014). US food aid and civil conflict. American Economic Review,
104(6), 1630–1666.
OECD (2019). Credit Reporting System, Accessed 6.25.2019. http://www.oecd.
org/dac/stats/crsguide.htm.
OECD (2019). List of Purpose CRS Codes, Updated April 2019.
Open letter to US congress (2017). http://www.usglc.org/downloads/2017/02/
FY18_International_Affairs_Budget_House_Senate.pdf.
Papke, L., & Wooldridge, J. (2008). Panel data methods for fractional response
variables with an application to test pass rates. Journal of Econometrics, 145,
121–133.
Pettersson, T. (2019). UCDP/PRIO Armed Conflict Dataset Codebook v 19.1 https://
ucdp.uu.se/downloads/.
Pettersson, T., Högbladh, S., & Öberg, M. (2019). Organized violence, 1989–2018 and
peace agreements. Journal of Peace Research, 56(4).
12 S. Mary, A.K. Mishra / World Development 126 (2020) 104713
Pflueger, C. E., & Wang, S. (2015). A robust test for weak instrument in Stata. The
Stata Journal, 15: 1, 216–225.
Piazza, J. A. (2008). Incubators of terror: Do failed and failing states promote
transnational terrorism?. International Studies Quarterly, 52(3), 469–488.
Polman, L. (2010). The Crisis Caravan: What’s Wrong with Humanitarian Aid?. New
York: Henry Holt and Co.
Reed, W. R. (2015). On the practice of lagging variables to avoid simultaneity. Oxford
Bulletin of Economics and Statistics, 77(66), 0305–9049.
Reuters (2014). U.N. seeks $2 billion for Sahel, fears donor fatigue. https://www.
reuters.com/article/us-sahel-un-appeal-idUSBREA120P720140203.
Schleussner, C.-F., Donges, J. F., Donner, R. V., & Schellnhuber, H. J. (2016). Armed-
conflict risks enhanced by climate-related disasters in ethnically fractionalized
countries. PNAS, 113(33), 9216–9221.
Steinwand, M. C. (2015). Foreign aid and political stability. Conflict Management and
Peace Science, 32(4), 395–424.
Stock, J. H., & Yogo, M. (2005). Testing for weak instruments in Linear IV regression.
In J. H. Stock & D. W. K. Andrews (Eds.), Identification and inference for
econometric models: Essays in Honor of Thomas J. Rothenberg. New York:
Cambridge University Press.
The Economist (2014). Cash for conflicts. https://www.economist.com/free-
exchange/2014/08/15/cash-for-conflicts.
The Washington Post (2018). A widening budget gap is forcing the U.N. to slash
food aid to refugees. https://www.washingtonpost.com/world/africa/a-
widening-budget-gap-is-forcing-the-un-to-slash-food-aid-to-refugees/2017/
12/27/b34cfd40-e5b1-11e7-927a-e72eac1e73b6_story.html?utm_term=.
f425fb73aa11.
Tranchant, J. P., Gelli, A., Bliznashka, L., Diallo, A. S., Sacko, M., Assima, A., & Masset,
E. (2019). The impact of food assistance on food insecure populations during
conflict: Evidence from a quasi-experiment in Mali. World Development, 119,
185–202.
United Nations (2010). Lack of funds may force UN agency to cut food aid in Central
African Republic. https://news.un.org/en/story/2010/08/348532.
Uppsala Conflict Data Program, Peace Research Institute Oslo, Accessed 6.28.2019
https://www.prio.org/Data/Armed-Conflict/UCDP-PRIO/.
USAID (2014). (Re)Assessing The Relationship Between Food Aid and Armed
Conflict. USAID Technical Brief. October.
Wood, R. M., & Sullivan, C. (2014). Doing harm by doing good? The negative
externalities of humanitarian aid provision of civil conflict. The Journal of
Politics, 77(3), 736–748.
Wood, R. M., & Molfino, E. (2015). Aiding victims, abetting violence: The influence of
humanitarian aid on violence patterns during civil conflict. Journal of Global
Security Studies, 1(3), 186–203.
World Bank (2016). Humanitarian cash and in-kind transfers across sectors:
Section, performance, and research priorities, Background paper for the Inter-
Agency Standing Committee (IASC), The World Bank Group, Evidence Paper,
Appendix A.
World Development Indicators (2019). World Bank. Accessed 6/26/2019.
http://databank.worldbank.org/data/source/world-development-indicators.
World Food Programme (2005). WFP forced to cut food aid in Burundi at critical
stage of peace process. https://www.wfp.org/news/news-release/wfp-forced-
cut-food-aid-burundi-critical-stage-peace-process.
World Hunger (2018). UN agency forced to cut food to 229,000 Syrian refugees.
https://www.worldhunger.org/un-agency-forced-to-cut-food-aid-to-229000-
syrian-refugees/.
Young, A. (2018). Consistency without inference: Instrumental variables in practical
application. Unpublished manuscript, London: London School of Economics
and Political Science. https://personal.lse.ac.uk/YoungA/ConsistencyWithout
Inference.pdf.
Zurcher, C. (2017). What do we (not) know about development aid and violence? A
systematic review. World Development, 98, 506–522.