IFPRI Discussion Paper 01374 

September 2014 

Do Girls Pay the Price of Civil War? 

Violence and Infant Mortality in Congo 

Olivier Dagnelie 

Giacomo De Luca 

Jean-François Maystadt 

Development Strategy and Governance Division 



INTERNATIONAL FOOD POLICY RESEARCH INSTITUTE 

The International Food Policy Research Institute (IFPRI), established in 1975, provides evidence-based 
policy solutions to sustainably end hunger and malnutrition and reduce poverty. The institute conducts 
research, communicates results, optimizes partnerships, and builds capacity to ensure sustainable food 
production, promote healthy food systems, improve markets and trade, transform agriculture, build 
resilience, and strengthen institutions and governance. Gender is considered in all of the institute’s work. 
IFPRI collaborates with partners around the world, including development implementers, public 
institutions, the private sector, and farmers’ organizations, to ensure that local, national, regional, and 
global food policies are based on evidence. IFPRI is a member of the CGIAR Consortium. 

AUTHORS 
Olivier Dagnelie (olivier.dagnelie@unamur.be) is an invited lecturer in the Adult Program in 
Economics and Management, Centre for Research in the Economics of Development at the University of 
Namur, Belgium. 

Giacomo De Luca (giacomo.deluca@york.ac.uk) is a lecturer in the Department of Economics at the 
University of York, UK. 

Jean-François Maystadt (j.maystadt@lancaster.ac.uk) was an associate research fellow in the 
Development Strategy and Governance Division of the International Food Policy Research Institute, 
Washington, DC, when he wrote this work. He is currently a senior lecturer in the Department of 
Economics at Lancaster University Management School, Lancaster, UK.  

Notices 
1. IFPRI Discussion Papers contain preliminary material and research results and are circulated in order to stimulate discussion and
critical comment. They have not been subject to a formal external review via IFPRI’s Publications Review Committee. Any opinions 
stated herein are those of the author(s) and are not necessarily representative of or endorsed by the International Food Policy 
Research Institute. 
2. The boundaries and names shown and the designations used on the map(s) herein do not imply official endorsement or
acceptance by the International Food Policy Research Institute (IFPRI) or its partners and contributors. 

Copyright 2014 International Food Policy Research Institute. All rights reserved. Sections of this material may be reproduced for 
personal and not-for-profit use without the express written permission of but with acknowledgment to IFPRI. To reproduce the 
material contained herein for profit or commercial use requires express written permission. To obtain permission, contact the 
Communications Division at ifpri-copyright@cgiar.org.

mailto:j.maystadt@lancaster.ac.uk


Contents 

Abstract v 

Acknowledgments vi 

1. Introduction 1 

2. Historical Background 3 

3. Data Sources and Sample Construction 4 

4. Empirical Strategy 8 

5. Results 11 

6. Behavioral versus Biological Factors 20 

7. Conclusions 24 

25 Appendix: Supplementary Tables and Figure

References 31 

iii 



Tables 

3.1 Descriptive statistics (1997–2004) 6 

4.1 Placebo test of exclusion restriction 9 

5.1 Panel regressions on the impact of mineral prices on conflict 12 

5.2 First-step specifications 14 

5.3 Regressions of intensity of conflict on child mortality 15 

5.4 Regressions of conflict on girl mortality: Alternative specifications 16 

5.5 Regressions of conflict on infant mortality: Mother fixed effects 17 

5.6 Alternative specifications with mother fixed effects 18 

5.7 Panel regressions on the impact of conflict on child mortality 19 

5.8 Regressions of conflict on girl mortality: Alternative panel specifications 19 

6.1 Regressions of conflict on girl mortality: Behavioral factors 21 

6.2 Regressions of conflict on girl mortality: Biological factors 23 

A.1 Regressions of conflict on girl mortality: Interaction with female-headed HH 25 

A.2 Regressions of conflict on girl mortality: Interaction with mother’s being a widow 26 

A.3 Regressions of conflict on girl mortality: Interaction with number of brothers 27 

A.4 Regressions of conflict on girl mortality: Interaction with number of sisters 28 

A.5 Regressions of conflict on girl mortality: Interaction with a proxy for son preference 29 

Figures 

2.1 Map of conflict events in Democratic Republic of Congo, 1997–2004 3 

3.1 Map of mineral deposits in Democratic Republic of Congo 5 

5.1 First-stage regressions on extended periods 13 

5.2 Lowess estimation of first stage 13 

A.1 Placebo test on girls sample 30 

iv 



ABSTRACT 

Civil wars inflict considerable development costs. Understanding the relative fragility of certain segments 
of the population is a necessary condition to build resilience to ongoing and future violence outbreaks. 
This paper documents the impact of the violent civil war affecting the Democratic Republic of Congo in 
the period 1997–2004 on infant mortality. It adopts an instrumental variable approach to correct for the 
nonrandom timing and location of conflict events using mineral price index variations by district, taking 
account of the mineral locations and prices, as instrument. Strong and robust evidence, including mother 
fixed effects regressions comparing siblings, shows that conflict significantly increases girl mortality. The 
paper also examines the mechanisms explaining this phenomenon, with a focus on disentangling the 
behavioral from the biological factors. The analysis suggests that gender imbalances in infant mortality 
are driven by the selection induced by a higher vulnerability of boys in utero rather than by gender 
discrimination. 

Keywords:  civil war, infant mortality, gender discrimination 
 

v 



ACKNOWLEDGMENTS 

We would like to thank Harold Alderman, Olivier Ecker, Kalle Hirvonen, Hannes Mueller, Petros 
Sekeris, and all participants in conferences and seminars where previous versions of this paper were 
presented. All errors and opinions expressed remain our own. 

vi 



1. INTRODUCTION

The impact of violence on the demography of a society can substantially increase the overall costs of a 
conflict, and heavily affect the time and the nature of the recovery process (Ghobarah, Huth, and Russett 
2003; Chen, Loayza, and Reynal-Querol 2008). A solid knowledge of the effects of war on the most 
fragile section of the population therefore represents a necessary condition for devising the proper 
responses to protect the next generations in conflict-prone environments, given the persistence over the 
life cycle of the detrimental impact of shocks experienced in early life (Aguero and Deolalikar 2012; 
Akresh et al. 2012; Dominguez and Barre 2013). 

A variety of factors may have a negative impact on infants’ health. Malnutrition, resulting from 
the contraction of the internal supply of food and a partial collapse of trade in the regions in which 
violence unravels, for instance, worsens the general health status of the affected population (Alderman, 
Hoddinott, and Kinsey 2006; Jenkins, Scanlan, and Peterson 2007; Bundervoet, Verwimp, and Akresh 
2009; Akresh and Edmonds 2011). Areas affected by violence are characterized by losses of or poor 
access to health infrastructures, due to lack of equipment and human resources. Timely interventions in 
cases of illness are crucial to guarantee full recovery. Limited access to health centers seriously affects 
antenatal care, professional birth attendance, and postnatal care, thereby endangering the survival of 
infants. Amid violence, health programs (such as prevention through vaccination and health education) 
are usually interrupted or implemented discontinuously, increasing the spread of vector-borne diseases. 
When displacement of large shares of a population occurs, lack of clean water and hygiene leads to a 
higher risk of diarrhea, one of the major causes of child morbidity and mortality. Finally, infants’ health 
may deteriorate as a consequence of conflict-related shocks experienced in utero (Camacho 2008; 
Almond and Currie 2011; Akresh, Lucchetti, and Thirumurthy 2012; Mansoor and Rees 2012; Minoiu 
and Shemyakina 2012). 

In line with these intuitive considerations, the literature regularly reports increased infant 
mortality rates in areas affected by civil war. The detrimental effect is consistent across humanitarian 
organization reports and the medical literature (Toole, Galson, and Brady 1993; Goma Epidemiology 
Group 1995; Danish Epidemiology Science Centre 1999; Kiros and Hogan 2001; Médecins Sans 
Frontières 2003; Coghlan et al. 2006). For instance, Coghlan et al. (2006) reported higher infant mortality 
rates in the eastern side of the Democratic Republic of Congo (DRC), the region of the country most 
heavily affected by the recent civil war. 

Contributions in the demography and economics literature, closer to this work in scope and 
methodologies, also consistently find higher child mortality rates in conflict-ridden areas (Guha-Sapir and 
van Panhuis 2004; Guha-Sapir et al. 2005; Guha-Sapir and D’Aoust 2010). Davis and Kuritsky (2002) 
showed that severe military conflicts in Africa south of the Sahara (SSA) raised infant mortality by 12 per 
thousand. Studying Khmer Rouge Cambodia, de Walque (2005) showed that infant and under-five 
mortality were very high for children born from 1970 to 1979. In particular, a child born between 1975 
and 1979 had roughly a 15 percent risk of dying within the first year of life, with no significant difference 
in mortality across gender. No gender differences were reported by Singh et al. (2005), who studied child 
mortality among a displaced population in Sudan and Uganda. Interestingly, they also found no difference 
in child mortality between refugees and the resident population, hinting that camps for internally 
displaced people might have an ambiguous effect on mortality. Finally, Verwimp and Van Bavel (2005) 
found that girls born during the refugee crisis in Rwanda displayed a particularly higher mortality rate as 
compared with those in the nonrefugee population. 

None of the previous studies, however, explicitly controlled for the potential endogeneity of 
conflict location. In particular, this paper argues that conflict is typically not randomly located and that 
failing to properly account for this may lead to a bias in the estimated results. The fact that violence has 
been reported to target wealthier households in neighboring countries like Burundi (Bundervoet 2010), 
Rwanda (Verpoorten 2009), and Uganda (Blattman and Annan 2010) suggests a bias that may push the 
estimated response of socioeconomic outcomes to conflict toward zero. Besides the likely nonrandomness 

1 



of conflict, microlevel data on conflict events (based on news reports) may suffer from measurement 
error. Conflict events in more remote and less connected locations will typically be underreported in the 
news and consequently in the data (Restrepo, Spagat, and Vargas 2006; Verpoorten 2012). Controlling for 
the endogeneity solves these two issues at least partially. 

This brings us to the main contributions of the present article. First, it studies the impact of recent 
violence in DRC on infant mortality rates, explicitly addressing the potential endogeneity of conflict 
location and timing. It instruments for conflict intensity using a mineral price index. In other words, it 
exploits the exogenous variation in the potential value of mineral sites generated by changes in world 
mineral prices to predict the geographic distribution of the conflict. The resulting estimates confirm the 
concerns expressed above. Simple ordinary least squares (OLS) results predict (for some specifications of 
the model) a decrease in mortality rates during the conflict in the districts most heavily affected by the 
violence, perhaps because he violence targets wealthier households likely to suffer lower infant mortality 
rates. An instrumental variables analysis, however, yields the more intuitive result that conflict increases 
infant mortality rates, but interestingly the detrimental effect seems to concentrate only among girls. This 
finding is robust to many different specifications, including a very demanding one that controls for 
mother fixed effects. The magnitude of the effects is substantial. According to the specification 
controlling for mother fixed effects (along with a long list of individual and climatic variables), an 
increase in conflict of 1 standard deviation would translate into a 9 percent increase in the likelihood that 
a girl will die before she turns one year old. 

The second contribution of the current study is to shed light on the gender-specific impact it 
uncovers. It investigates whether the imbalance is driven by biological or behavioral factors. More 
specifically, it identifies in the literature potential factors that may foster gender discrimination against 
girls in households facing difficult times. The analysis finds no evidence for the existence of gender 
discrimination in the context of DRC conflict. In line with medical evidence (Shettles 1961; Mizuno 
2000; Kraemer 2000; Catalano et al. 2006), an alternative explanation relates to the biological 
vulnerability of boys in utero that would generate gender selection at birth (Valente, forthcoming). The 
present study assesses whether girls’ higher mortality rate may be partially explained by a selection in 
utero against boys due to conflict exposure, and it finds tentative evidence for this mechanism. Overall, 
the analysis suggests that gender imbalance in child mortality is largely explained by selection in utero 
against boys rather than resulting from gender discrimination in times of scarce resources. In other words, 
the survival of the fittest boys in utero would explain gender imbalances during the first 12 months of life. 

In the next section we provide the relevant background information on the armed conflict in the 
DRC. Section 3 presents the data. Section 4 lays out the empirical strategy, and results are presented in 
Section 5. Section 6 discusses the relative importance of behavioral and biological factors in explaining 
the different impact of violence on male and female infants. The last section concludes. 

2 



2. HISTORICAL BACKGROUND

The DRC has experienced two of the most violent wars in recent history. The first Congolese war, which 
started at the end of 1996, is usually interpreted as a fight by the coalition of the Congolese rebellion led 
by Laurent-Desire Kabila with the foreign governments of Rwanda and Uganda not only to overthrow 
Mobutu but also to eradicate the presence of Rwandan Hutu refugees in eastern DRC, where they had 
escaped in the aftermath of the 1994 genocide (Vlassenroot and Raeymaekers 2004; Prunier 2009). 

The second Congolese war unraveled between 1998 and 2004, with an astonishing estimated 
death toll of more than 3.8 million people (International Rescue Committee 2011) and an estimated 1.7 
million internally displaced people (Internal Displacement Monitoring Center 2011). This magnitude of 
death and displacement is likely to have impinged upon the health of infants in affected areas. 
Interestingly, there is extensive anecdotal evidence of the role of minerals in shaping the dynamics of the 
conflicts, particularly during the second war (Congdon Fors and Olsson 2004; Turner 2007; Gambino 
2011; Stearns 2011; Sanchez de la Sierra 2013). This link will constitute the rationale of our instrumental 
variable strategy. 

The provinces most heavily affected by the violence were Orientale and North and South Kivu, 
the areas in which the concentration of local and foreign armed groups was highest. Conflict was also 
concentrated in the territory of Pweto (Haut-Katanga district) in the Katanga province as well as in 
Kinshasa (Figure 2.1). 

Figure 2.1 Map of conflict events in Democratic Republic of Congo, 1997–2004 

Source:  Authors. 

3 



3. DATA SOURCES AND SAMPLE CONSTRUCTION

To assess the impact of the conflict in terms of infant mortality—defined as child mortality at 12 
months—we make use of the Demographic and Health Survey (DHS) on DRC carried out in 2007. Since 
we take advantage of the timing and location variations of conflict events, our main sample excludes 
children for whom we do not know exactly where their household was living at the time of their birth.1 
We select those children born between 1997 and 2004, encompassing the two Congolese wars. DHS 
surveys are meant to be nationally representative and collect individual information on women aged 15 to 
49 on education, demographic, and health issues as well as some information on the location of the 
interview, among which are GPS coordinates.2 Thanks to the inclusion of each woman’s maternal history 
in the dataset, we have recovered detailed information such as when her children were born; whether they 
are still alive and if not, when they died; and whether they were part of a multiple birth. This enables us to 
create variables counting the number of brothers and sisters alive at the time of a child’s birth. 

We also take advantage of the geographical information linked to each DHS cluster to create 
three climatic variables. Given the emerging evidence on the links between weather shocks and violence 
(Hsiang, Burke, and Miguel 2013), introducing these variables could potentially reduce the risk of 
confounding factors. The first two variables are expressed in standard deviations from a long-term 
average (that is, of the previous 25 years) and are the relative sum of, respectively, monthly rainfall and 
temperature observations during the first 12 months of life of each child. The data used to construct the 
measure of precipitation and temperature come from Terrestrial Precipitation: 1900–2008 Gridded 
Monthly Time Series, Version 2.01, interpolated and documented by Matsuura and Willmott (2009). 
This dataset is a compilation of updated sources and provides monthly precipitation (and mean 
temperature) interpolated to a latitude/longitude grid of 0.5 degree by 0.5 degree from an average of 20 
weather stations. 

We also create a third variable combining rainfall with daily temperature obtained from the 
Prediction of Worldwide Energy Resource (POWER) database of the US National Aeronautics and Space 
Administration (NASA): the number of months of potential malaria exposure in the first 12 months of 
life. To build such an index we apply the approach proposed by Kudamatsu, Persson, and Stromberg 
(2012). Four conditions have to be simultaneously satisfied for a month to be considered as malaria 
prone: The malaria index Mdm for district d and month m is set to 1 if and only if 

1. average monthly rainfall during the past 3 months is at least 60 mm,
2. rainfall in at least 1 of the past 3 months is at least 80 mm,
3. no day in the past 12 months has an average temperature below 5◦C, and
4. the average temperature in the past 3 months exceeds 19.5◦C plus the standard deviation

of the monthly temperature in the past 12 months.
For the first stage of our instrumental variables estimates, we investigate the relationship between 

conflict events and mineral prices. To this end, we construct a panel dataset of conflict events and a price 
index. We filter the data from the Armed Conflict Location and Event Data Project (ACLED) dataset on 
the DRC and keep events from January 1, 1997, to December 12, 2004, that are not described as riots 
(Raleigh et al. 2010). A conflict event is defined as a single altercation wherein force is used by one or 
more groups for a political end (Raleigh et al. 2010). Thanks to the availability in the dataset of GPS 
coordinates for each conflict event, we assign each conflict to its respective district and time period, using 
the shapefiles on DRC from the Global Administrative Areas Database. For each month in the period 

1 DHS datasets provide the number of years a household has been living in the village where the interview took place. 
Information about previous location is, however, not available, which means that we do not know whether households who 
migrated remained in the same district. In Section 5 we show that our results are robust to the inclusion of children belonging to 
nonresident households. 

2 Because only surviving women are interviewed, we are likely to underestimate the impact of conflict on child mortality. 
Nonetheless, there is no reason to believe that this source of sample selection will affect gender imbalances in child mortality. 

4 



considered, we create a district-level measure of conflict by summing all events taking place in a 
given district (Conflict Event). 

The location of ore for various minerals, obtained from the mineral occurrences map of DRC, is 
also assigned to one of the 38 districts.3 We therefore know the mineral potential of each district of DRC, 
which we use to compute a price index. 

One can observe, in Figures 2.1 and 3.1, the geographical repartition by district, respectively, of 
conflict events from 1997 to 2004 and mineral exploitation sites in DRC. It is rather striking that the 
eastern part of the country, richer in minerals as confirmed by Figure 3.1, experienced more conflict 
events. This correlation, however, does not account for time variations, which we intend to exploit by 
interacting the mineral potential of each district with the monthly price of the corresponding resource. 

Figure 3.1 Map of mineral deposits in Democratic Republic of Congo 

Source:  Authors. 

We turn to the United Nations Conference on Trade and Development (UNCTAD) dataset to get 
the monthly price series of 12 minerals (aluminum, copper, gold, iron, lead, manganese, nickel, oil, 
phosphate, tin, wolframite, and zinc) and compile information from The Economics of Tantalum (Roskill 
Information Services 2009), Metal Pages, and the US Geological Survey to build our price series for 
tantalum. The number of extraction sites is interacted with the monthly mineral prices to obtain a time-
varying measure of relative mineral value by district (Price Index). For each time period and district, we 
compute a price index taking account of the number of extraction sites and the price for the set of 13 
minerals, as follows: 

(1) 

3 The geological service of the African Museum of Tervuren, Belgium, provided the mineral occurrences map of DRC. 

5 

Price Indexit =
∑
r

ωriprt,



                                                                                                                                  
 r in district i with respect to other districts and prt is the price of mineral r at period t at a price 
normalized to 100 for the first period (January 1997). Since we have no information as to the realized or 
potential extraction of ores of each mineral location, we decided to weight each monthly ore price by a 
ratio of the number of mineral deposits in the district over the total number of deposits of this particular 
mineral in the country. This way, we intend to proxy the extraction potential by ore of each district. We 
sum this potential over all minerals in a price index in order to reflect the district resource endowment 
value at each point in time. In other words, we capture how the monthly change in mineral prices alters 
the relative potential value of the mining sector across districts. This strategy is similar to the one 
adopted by Bruckner and Ciccone (2010) in their study on conflict in SSA. 

All the variables used in our analysis are described in Table 3.1. In order to obtain a nationally 
representative dataset, we resort to sampling weights provided in the DHS. These are needed to render the 
estimates independent from the sampling design. 

Table 3.1 Descriptive statistics (1997–2004) 

Variable N Mean Standard 
deviation

Minimum Maximum 

10,397 0.107 0.310 0 1 
5,242 0.117 0.321 0 1 
5,155 0.098 0.297 0 1 
10,397 7.896 14.467 0 123 
10,397 2.276 2.820 0 12 
10,397 492.04 624.06 0 3,002.6 
10,397 0.017 0.131 0 1 
10,397 0.195 0.396 0 1 
10,397 1.209 1.457 0 8 
10,397 1.175 1.430 0 8 
10,397 7.4 2.991 0 11 
10,397 -0.239 0.990 -3.477 5.149 
10,397 0.737 1.083 -1.546 5.397 
10,397 0.251 0.434 0 1 
10,397 0.424 0.494 0 1 
10,397 0.315 0.464 0 1 
10,397 -1,655.5 97,422.5 -107,521 341,565 
10,397 0.139 0.346 0 1 

Panel A: Cross-section dataset  
Infant mortality
Boy mortality 
Girl mortality 
Number of conflict events 
Exposure to conflict (# months) 
Price index 12 month 
Twin 
1st child 
# brothers alive 
# sisters alive 
Exposure to malaria 
Rain 25 year (dev) 
Temp 25 year (dev) 
Mother —no education 
Mother —primary 
Mother —secondary 
Wealth index 
Female-headed household 
Household size 10,397 7.336 3.063 1 28 

2,876 0.106 0.221 0 1 
2,876 0.090 0.234 0 1 
2,876 0.076 0.218 0 1 
2,876 5.639 12.327 0 123 
2,876 1.769 2.538 0 12 
2,876 417.351 585.228 0 3,002.567 
2,876 7.699 2.666 0 11 
2,876 -0.213 0.944 -3.239 5.149 
2,876 0.826 1.096 -1.523 5.259 
2,598 0.511 0.350 0 1 
2,598 4.412 10.079 0 103 
2,598 1.353 1.951 0 9 

Panel B: Panel dataset (district-month) 
Infant mortality rate (12 month) 
Boy mortality rate (12 month) 
Girl mortality rate (12 month) 
Number of conflict events 
Exposure to conflict 
Price index 12 month 
Months of malaria exposure (average) 
Rain 25 year (average deviation) 
Temp 25 year (average deviation) 
Sex ratio at birth 
Number of conflict in utero 
Exposure to conflict in utero 
Price index 9 month in utero 2,598 296.009 412.647 0 1,924.32 

Source:  Authors. 
Notes:    dev = deviation. Correction for sampling weights. 

6 

where ωri = mineralsri/
∑

j mineralsrj is a weight measuring the relative importance of mineral



Table 3.1 shows a few noteworthy things. First, infant mortality is relatively high during the 
period 1997–2004 as 1 child in 10 failing to reach the age of one year. As is commonly observed 
worldwide, boy mortality is higher than girl mortality (Wilson 1975; Waldron 1998; Garenne 2003). 
Interestingly, the average ratio of male over female mortality stands at about 1.2, which lies toward the 
bottom of the range of 1.15–1.30 identified by Garenne (2003) in other countries in normal times. This 
may already suggest a detrimental impact of warfare on girls’ mortality at the aggregated level. Our 
empirical analysis will put forward compelling evidence that, once the potential endogeneity of conflict 
location is properly addressed, conflict significantly increases girl mortality.

7 



4. EMPIRICAL STRATEGY

The ultimate goal is to estimate the impact of conflict on infant mortality. Since the timing and location of 
conflict events is likely to be nonrandom, we turn to instrumental variable analysis. We first discuss in 
detail the first-stage relationship, wherein we predict conflict distribution based on the change in mineral 
prices. We then use this instrument to assess the impact of conflict on infant mortality. 

Conflict 
In the context of conflict in the DRC, we expect resources to shape the distribution of the violence. We 
therefore use the previously described price index to predict the intensity of conflict by district and over 
time. Formally, the first-stage specification is as follows: 

(2) 

where Conflictdt is one of the measures of conflict. We control for district fixed effects, αd, time fixed 
effects, βt, and district-specific linear time trends, δdt. We run several specifications of the model using 
Conflict Events as dependent variable to check the robustness of the first-stage relationship. Next, given 
the interest of this study, we run the first stage to predict, for each month in the sample, the conflict 
distribution over the preceding 12 months, denoted by Conflict Events 12. We also adopt an alternative 
measure of conflict, which records for each month the number of months featuring violence in the 
districts over the preceding 12 months, denoted by Conflict Exposure 12. In those cases, we change the 
price index used accordingly (Price Index 12). 

Infant Mortality 
The second stage of the analysis focuses on infant mortality, defined as the mortality of children by 12 
months of age. In order to exploit the richness of the dataset, we first run cross-sectional regressions with 
several types of fixed effects, including a within-household comparison through the use of mother fixed 
effects. As an alternative specification, we next run panel regressions to estimate mortality rates at 12 
months by district. 

Cross-section Regressions 
In this context, the unit of observation is a child, i, born at month m, in district d. We check 12 months 
after her birth whether she is still alive and assign value 1 to our binary variable Mortalityimd if childi died 
during the first 12 months of her life. We denote by Conflict Events 12dm the sum of conflict events that 
took place in district d during the first 12 months of life of child i born 12 months before month m, that is

   Formally, we estimate the following model: 

(3) 

where Xi is a vector of control variables including whether the child was part of a multiple birth, whether 
she was the first child, her number of siblings alive, a malaria index (summing the number of months of 
exposure to malaria), and rainfall and temperature anomalies (with respect to a long-term average of 25 
years). A further set of variables, Xhh, controls for household characteristics including the education 
level of the mother and a measure of the wealth of the household. 

To deal with the potential endogeneity of conflict distribution, the conflict intensity measure is 
instrumented by the mineral price index described above. Formally, the following system of two 
equations is estimated by limited information maximum likelihood:

8 

Conflict Events 12dm =
∑

j
12
=1 ConflictEventsdm−j .

Mortalityidm = αd +βm + δ
′
Xi +λ

′
Xhh + γConflict Events 12dm + εidm,

Conflictdt = αd + βt + δdt + γPrice Indexdt + εdt,



        (4) 

 (5) 

The idea behind this empirical strategy is to exploit the timing and location variations of conflict 
events and to compare children born in the same month in districts affected differently by conflict. 
Standard errors are clustered at the village level, and sampling weights are used to render the estimates 
independent of the sampling design. As mentioned before, the regressions are run on the sample of 
children known to have been born in their mother’s interview district. Enlarging the sample to include 
those children who migrated does not qualitatively change the results. 

The most demanding and most convincing specification of this empirical exercise is to include in 
the model mother fixed effects, enabling us to compare along the dimension of exposure to conflict 
children born to the same mother. Controlling for mother fixed effects allows us to take account of both 
environmental differences between children and each mother’s genetic features. In order to ease the 
estimation process, we partial out each variable with the series of all the fixed effects. By the Frisch-
Waugh-Lovell theorem, we know this method to keep the sign and magnitude of the coefficients 
unchanged. In order to ensure that the initial inference was correct, we turn to 999 replications of wild 
bootstrap (percentile-t method), known to resist to heteroskedasticity, to produce confidence intervals.4 

Beyond the strength of the instruments, this empirical strategy is based on a key identifying 
assumption. If mineral prices influence infant mortality through another channel than the occurrence of 
violence, this would violate our exclusion restriction. In particular, one could claim that wealth effects 
coming from mineral price variations could invalidate our identification strategy. A first indication that it 
should not be the case is that our point estimates are virtually unchanged when controlling for household 
characteristics such as the education of the mother or household wealth. Controlling for wealth effects, 
the exclusion restriction is likely to hold. To test further the plausibility of our exclusion restriction, we 
proceed to a placebo test. Regressions of the same price index on infant mortality over the period 1980–
1996 showed that the exclusion does not seem to be at risk.5 In particular, we fix the length of the sample 
to 96 months, as in the main analysis. Next, we run regressions starting on the first month of the enlarged 
sample (January 1980). Moving each time by one month our sliding window of 96 months from the start 
of the sample until the end, we test the reduced-form relationship in 95 regressions for the preconflict 
period. When we include no district-specific linear trends, we obtain only 9 out of 95 times a (weakly) 
significant reduced-form relationship. Including linear district-specific trends yields no significant 
coefficient out of the 95 regressions. The results of this placebo exercise are summarized in Table 4.1. 
Such a result is difficult to conciliate with price-induced wealth effects. 

Table 4.1 Placebo test of exclusion restriction
p-value less than: 0.1 0.05 0.01 # regressions 

No trends 9 0 0 95 
Linear trends 0 0 0 95 

Source:  Authors.  
Notes:   The table reports the number of times we found a significant reduced-form relationship between our price index and 

infant mortality for the prewar period 1980–1996 using a sliding window of 96 months for each regression. 

4 Traditional robust two-stage least squares standard errors produced qualitatively similar results. 
5 Infant mortality for this period is reconstructed retrospectively from the DHS. 

9 

Mortalityimd = αd + βm + δ
′
Xi + λ

′
Xhh + γConflict Events 12dm + εidm 

ConflictEvents 12dm = δd + ρm + η
′
Xi + θ

′
Xhh + νP rice Index 12dm + υidm



Panel Regressions 
As an alternative specification, we aggregate the data at the district level (at the cost of not controlling for 
changes in the composition of the district populations). The unit of observation is then the district, d, for 
which we compute a mortality rate on all the children born there 12 months before, Mortalitydm. 

We run the following fixed-effect panel regression (with or without district-specific linear trends): 

(6) 

As for the cross-section analysis, we correct for endogeneity with two-stage least squares, instrumenting conflict by 
mineral prices, and estimate the following system of equations: 

 (7) 

 (8) 

10 

Mortalitydm = αd + βm + γConflictEvents 12dm + εdm.

Mortalitydm = αd + βm + γConflict Events 12dm + εdm

 Conflict Events 12dm = δd + ρm + ηP rice Index 12dm + εdm.



5. RESULTS

Conflict 
Table 5.1 exhibits that at the district level, the relationship between mineral prices and conflict events is 
highly significant and negative (first column and row of Panel A). Adding linear district-specific time 
trends confirms the significance of the price index coefficient (first column and row of Panel B). We run a 
set of further robustness checks on the first stage. In the second row of column 1, we exclude tantalum 
from the price index, given the alleged relevance of this mineral in the conflict dynamics as stressed in the 
international press. Conversely, in the third row we compute the price index only on tantalum. 
Interestingly, removing one mineral at a time from the analysis does not alter the results. The next three 
rows replicate the first three but predict conflict in a 12-month period based on the sum of the price 
variations over a 12-month period. 

In column 2, we restrict the analysis to the 19 districts (half of DRC districts) hosting most of the 
violence in the period considered. Here also, removing one district at a time does not qualitatively change 
the findings. In column 3 we restrict the analysis to the 19 districts richest in minerals (based on the price 
index). 

In column 4–6 we focus on the role of tantalum in the relationship between mineral prices and 
conflict. In column 4 the analysis is conducted only on districts with tantalum mines. Column 5 restricts 
the sample to the 5 districts most influenced by the tantalum price changes, and the last column looks at 
the districts excluded from column 5. All specifications confirm the negative relationship. 

Alternative specifications, displayed in Figure 5.1 and in which we extend the period considered, 
confirm the robustness of this relationship, which still holds after 2004, the end of the second Congolese 
war but with a lesser magnitude. The longer the period included in the analysis after 2004, the lower the 
coefficient of interest. 

Finally, we test the first stage using a nonparametric smoother (a locally weighted smoother, 
specifically the lowess estimator) on the demeaned and detrended versions of the conflict and mineral 
price measures. The result of this exercise, reported in Figure 5.2, confirms that the linear specification 
constitutes a good approximation of a potentially more complex relationship.6 The negative relationship is 
again confirmed using a nonparametric approach. 

6 Reducing the bandwidth of this estimator from 0.80 to 0.20, and hence producing less smoothing by focusing on closer 
points in the local regressions, leaves the general trend un- changed. 

11 



Table 5.1 Panel regressions on the impact of mineral prices on conflict 
Dependent variable: 

(1) 
All 

(2) 
19 highest C 

Conflict events 
(3) (4) 

19 highest P TA districts 
(5) 

5 districts 
(6) 

Drop (5) districts 
Panel A: No trend 

Price index -.00922*** -.0107*** -.0102*** -.0125*** -.0226*** -.00498* 
(.00224) (.00271) (.00297) (.00234) (.0033) (.00262) 

Price index No TA -.00892*** -.0106*** -.00902*** -.0225* -.0467* -.00582** 
(.00281) (.00341) (.00246) (.0118) (.0205) (.00245) 

TA price index -.0106** -.012** -.0129** -.0114** -.0145 .151 
(.0044) (.00509) (.00493) (.00471) (.00931) (.119) 

Price index 12 -.012*** -.0154*** -.0151*** -.0218*** -.0415** -.00406 
(.00391) (.0048) (.00496) (.00393) (.0108) (.00251) 

Price index 12 No TA -.00618** -.00874** -.00896** -.00755 -.039* -.00441* 
(.00256) (.00343) (.00312) (.0107) (.0166) (.00251) 

TA price index 12 -.0298*** -.0303*** -.0322*** -.0309*** -.0445** .144 
(.00554) (.00619) (.00596) (.00591) (.0124) (.118) 

Panel B: Linear trend 

Price index -.0103*** -.0116*** -.0105*** -.0125*** -.0171** -.0073** 
(.00198) (.00233) (.00283) (.0022) (.00543) (.00331) 

Price index No TA -.0133*** -.0154*** -.011*** -.0328** -.0362 -.00866*** 
(.00373) (.0043) (.00333) (.0143) (.0332) (.0026) 

TA price index -.00906** -.0103** -.0113** -.00983** -.0115 .151 
(.00406) (.00476) (.0046) (.00437) (.00863) (.12) 

Price index 12 -.0167*** -.0189*** -.0177*** -.0251*** -.0292 -.00638** 
(.00358) (.00315) (.00353) (.0036) (.0144) (.00298) 

Price index 12 No TA -.0124*** -.0156*** -.0116*** -.0286*** -.025 -.00719** 
(.00395) (.00424) (.00394) (.00939) (.0251) (.00276) 

TA price index 12 -.0245*** -.0244*** -.0266*** -.0256*** -.0362*** .152 
(.00298) (.00371) (.00344) (.00339) (.00759) (.117) 

N 3648 1824 1824 2112 480 3168 
District FE 
Year FE 

Source:  Authors. 
Notes:    FE = fixed effects; TA = tantalum. *** p < 0.01, ** p < 0.05, * p < 0.1. 

12 



Figure 5.1 First-stage regressions on extended periods 

Source:  Authors. 

Figure 5.2 Lowess estimation of first stage 

Source:  Authors. 

Table 5.2 puts forward the measures of conflict, Conflict Events 12 and Conflict Exposure 12, the 
coefficient of which is again strongly significant and negative. Significance is actually increased with 
linear district-specific trends.7 

7 These results hold if we use quadratic instead of linear trends. 

13 



Table 5.2 First-step specifications 
 

Dependent variable Conflict events 12 Conflict exposure 12 
Trend None 

(1) 
Linear 

(2) 
None 

(3) 
Linear 

(4) 
Price index 12 -.0120*** 

(.0039) 
-.0167*** 
(.0036) 

-.0014* 
(.0007) 

-.0022*** 
(.0006) 

N 3,648 3,648 3,648 3,648 
District FE 

Year FE 
Source:  Authors.  
Notes:    FE = fixed effects. *** p < 0.01, ** p < 0.05, * p < 0.1; 

Infant Mortality 
Table 5.3 presents cross-section regressions over the sample of children born between 1997 and 2004, for 
all children (columns 1–2), only boys (columns 3–4) or only girls (columns 5–6). We find strong 
evidence for the nonrandom distribution of violence. Table 5.3 suggests a downward bias of naive OLS 
regressions. A potential explanation may be that conflicts are more likely to target more wealthy (and 
therefore healthy) households in a looting-driven warfare. The results also identify a gender-specific 
impact of conflict episodes on child mortality. Girls are more adversely affected by conflict than boys. 
According to column 6, a change by 1 standard deviation in the number of conflict events (that is, by 
about 14 conflict events) increases the probability of a girl’s dying within the first year of life by about 7 
percent. The magnitude of this effect is far from trivial, inasmuch as it constitutes a doubling of girls’ 
mortality (at mean value). Adding the full set of controls in Panel B of Table 5.3 leaves the coefficients of 
interest virtually unchanged. That is very reassuring with respect to the exclusion restriction. Conditional 
on household wealth, the exclusion restriction is more likely to hold.8 Given the magnitude and 
significance of the F-statistics on the instrument, it is very unlikely that the analysis should be invalidated 
by a problem of weak instrument. 
 

8 Figure A.1 in appendix further supports the exclusion restriction. It graphically shows the coefficients obtained by 
regressing girl mortality on the price index over a moving window of 96 months. It is striking to see how the coefficients are 
consistently around zero in the preconflict period. The negative response of girls’ mortality to mineral price shocks is particularly 
strong during the period of investigation. As shown on the right panel, the relationship is even stronger in highly conflictive 
districts. 

14 



Table 5.3 Regressions of intensity of conflict on child mortality 

 

Source:  Authors. 
Notes:   2SLS = two-stage least squares; FE = fixed effects; HH = household; IV = instrumental variables;  dev = deviation;

OLS = ordinary least squares.*** p < 0.01, ** p < 0.05, * p < 0.1. Correction for sampling weights. Sample of 
residents only. Standard errors clustered at the village level. Dummies for ethnic groups (10) and religion (9) are 
included in all regressions of Panel B. 

Being part of a multiple birth is robustly significant and increases mortality, although the result 
hinges on only 2 percent of the sample. Being a first child is likely to be detrimental as well, although the 
impact is less robust because it loses significance once the sample is split by gender, following the 
increase of the standard errors. In line with the literature, we find strong and robust harmful effects for 

Dependent variable Mortality rate at 12 months 
All Boys Girls 

OLS 2SLS OLS 2SLS OLS 2SLS 
(1) (2) (3) (4) (5) (6) 

Panel A: 
Conflict events 12 -0.0007 0.0008 -0.0017*** -0.0041 0.0002 0.0051 ** 

(0.0005) (0.0021) (0.0006) (0.0032) (0.0008) (0.0023) 
F-stat of IV 33.65*** 28.16*** 31.64*** 

Panel B: 
Conflict events 12 -0.0007 0.0016 -0.0016*** -0.0031 0.0002 0.0055 ** 

(0.0005) (0.0022) (0.0006) (0.0031) (0.0008) (0.0024) 
Twin  0.1801***   0.1807 ***   0.1420 ***   0.1409 ***   0.2369 ***   0.2343 *** 

(0.0407) (0.0405) (0.0475) (0.0467) (0.0682) (0.0685) 
1st child 0.0204 * 0.0214 * 0.0263 0.0254 0.0114 0.0137 

(0.0117) (0.0116) (0.0194) (0.0190) (0.0164) (0.0159) 
# brothers alive 0.0054 0.0055 0.0088* 0.0088* 0.0012 0.0014 

(0.0037) (0.0037) (0.0053) (0.0052) (0.0045) (0.0044) 
# sisters alive 0.0058* 0.0060* 0.0038 0.0036 0.0071 0.0073* 

(0.0032) (0.0032) (0.0045) (0.0045) (0.0043) (0.0042) 
Malaria exposure 0.0045 0.0041 0.0044 0.0046 0.0042 0.0030 

(0.0030) (0.0029) (0.0037) (0.0038) (0.0028) (0.0026) 
Rain 25y (dev) -0.0064 -0.0077 -0.0120 -0.0108 -0.0049 -0.0065 

(0.0073) (0.0075) (0.0081) (0.0083) (0.0093) (0.0095) 
Temp 25y (dev) 0.0078 0.0077 0.0200** 0.0203** -0.0031 -0.0023 

(0.0060) (0.0061) (0.0084) (0.0081) (0.0078) (0.0079) 
Mother—no education  0.0545* 0.0540*  0.0224  0.0232 0.0837** 0.0836** 

(0.0326) (0.0324) (0.0420) (0.0417)  (0.0380) (0.0379) 
Mother—primary 0.0190 0.0182 0.0045 0.0056 0.0297 0.0298 

(0.0278) (0.0276) (0.0388) (0.0383) (0.0332) (0.0333) 
Mother—secondary -0.0054 -0.0061 -0.0368 -0.0354 0.0192 0.0191 

(0.0268) (0.0265) (0.0381) (0.0379) (0.0291) (0.0290) 
Wealth index -0.0000 -0.0000 -0.0000 -0.0000* -0.0000 -0.0000 

(0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) 
HH size -0.0082***   - 0.0081***  - 0.0091***  - 0.0091***  - 0.0073***  - 0.0070*** 

(0.0021) (0.0021) (0.0031) (0.0030) (0.0017) (0.0017) 
Female-headed HH -0.0184 -0.0186 -0.0266* -0.0259* -0.0120 -0.0108 

(0.0123) (0.0122) (0.0142) (0.0138) (0.0219) (0.0214) 
F-stat of IV 44.88*** 34.62*** 42.41*** 

District & month FE 
N 10,397 10,397 5,242 5,242 5,155 5,155 

15 



children whose mother does not record any years of formal education (see, for instance, Behrman and 
Wolfe 1987; Strauss and Thomas 1995). As expected, wealth decreases infant mortality. These results are 
extremely robust to alternative specifications. Household size is also negatively associated with infant 
mortality. In Table 5.4, we assess the robustness of the results to the addition of a linear trend (column 1), 
the inclusion of nonresident households in the sample (column 2), and the use of the alternative 
measurement of violence based on the number of months of exposure (columns 3–4). The addition of a 
linear trend improves the efficiency of the point estimates while, as expected, the precision and the size of 
the conflict coefficients decrease with the extended sample. Adopting the alternative measure of violence 
gives even higher coefficients for the variable of interest. According to the coefficient reported in column 
4, an increase of 1 standard deviation in violence translates into about a 13 percent increase in the 
likelihood of a girl’s dying within the first 12 months of life. 

Table 5.4 Regressions of conflict on girl mortality: Alternative specifications 

Dependent variable Girls’ mortality at 12 month 
Conflict events 12 Conflict exposure 12 

2SLS 2SLS 2SLS 2SLS 
(1) (2) (3) (4) 

Panel A: No controls 
Conflict 0.0058** 0.0033* 0.0426** 0.0469** 

(0.0024) (0.0018) (0.0196) (0.0120) 
F-stat of IV 125.7*** 31.81*** 20.86*** 65.38*** 

Panel B: Full set of controls 
Conflict 0.0058** 0.0038** 0.0450** 0.0489** 

(0.0028) (0.0019) (0.0201) (0.0241) 
F-stat of IV 110.9*** 37.88*** 23.04*** 44.50*** 

Linear trend 
Includes nonresidents 
District & month FE 
N 5,155 6,680 5,155 5,155 

Source:  Authors. 
Notes:    2SLS = two-stage least squares; FE = fixed effects; IV = instrumental variables.*** p < 0.01, ** p < 0.05, * p < 0.1. 

Correction for sampling weights. Sample of residents only. Standard errors clustered at the village level. Full set 
of controls in Panel B as in Panel B of Table 5.3.  

Despite the addition of mother and household characteristics in Panel B of Tables 5.3 and 5.4, we 
cannot be certain unobserved child characteristics are not driving the relationship between violence and 
infant mortality. To reduce that concern, we introduce mother fixed effects. Intuitively, we compare the 
mortality of children with the same mother differently exposed to conflict. The gender imbalance in infant 
mortality is further confirmed in Table 5.5. The magnitude of the effect is even stronger. According to the 
results reported in column 6, an increase in conflict of 1 standard deviation would translate into a 9 
percent increase in the likelihood of a girl’s dying before she turns one year old. Said differently, an 
increase of 1 standard deviation in conflict would be responsible for the death of 454 additional girls out 
of the sample of 5,155 girls. As for the previous specifications, the mother fixed effects results do not 
depend on the inclusion of climate and individual characteristics (Panel B of Table 5.5). Table 5.6 further 
indicates that the results are robust to the inclusion of a district-specific time trend and the use of the 
alternative definition of conflict (months of exposure). 

16 



Table 5.5 Regressions of conflict on infant mortality: Mother fixed effects 

17 

Dependent variable Mortality rate at 12 months 

All Boys Girls 
OLS 2SLS OLS 2SLS OLS 2SLS 
(1) (2) (3) (4) (5) (6) 

Panel A: 
Conflict events 12 -0.001 0.000 -0.001* -0.003 -0.001 0.006** 

[-0.002, 0.000] [-0.004, 0.005] [-0.002, 0.000] [-0.007, 0.003] [-0.002, 0.001] [ 0.000, 0.010] 
F-stat of IV 42.2*** 32.6*** 34.1*** 

Panel B: 
Conflict events 12 -0.001 0.001 -0.001 -0.002 -0.001 0.006** 

[-0.002, 0.000] [-0.003, 0.005] [-0.002, 0.000] [-0.006, 0.003] [-0.002, 0.001] [-0.000, 0.011] 
Twin 0.124** 0.124*** 0.107* 0.107* 0.140* 0.139* 

[ 0.021, 0.214] [ 0.033, 0.213] [-0.019, 0.222] [-0.021, 0.227] [-0.022, 0.262] [-0.038, 0.264] 
1st child -0.000 -0.001 0.012 0.012 -0.018 -0.022 

[-0.046, 0.042] [-0.043, 0.045] [-0.034, 0.054] [-0.032, 0.054] [-0.070, 0.048] [-0.080, 0.043] 
# brothers 0.057*** 0.058*** 0.081*** 0.081*** 0.010 0.014 

[ 0.044, 0.072] [ 0.044, 0.072] [ 0.059, 0.098] [ 0.059, 0.099] [-0.010, 0.034] [-0.008, 0.036] 
# sisters 0.044*** 0.044*** -0.001 -0.001 0.060*** 0.063*** 

[ 0.028, 0.059] [ 0.028, 0.058] [-0.025, 0.021] [-0.025, 0.023] [ 0.039, 0.081] [ 0.040, 0.088] 
Malaria index 0.007 0.005 0.007 0.007 0.004 -0.003 

[-0.002, 0.016] [-0.007, 0.015] [-0.008, 0.020] [-0.010, 0.021] [-0.014, 0.024] [-0.021, 0.019] 
Rain 25y -0.011 -0.017* -0.016* -0.004-0.009     -0.010 

[-0.027, 0.006] [-0.029, 0.006] [-0.033, 0.002] [-0.031, 0.001] [-0.031, 0.016] [-0.037, 0.014] 
Temp. 25y 0.010 0.010 0.013 0.012 0.003 0.002 

[-0.005, 0.020] [-0.006, 0.022] [-0.004, 0.029] [-0.004, 0.028] [-0.015, 0.018] [-0.017, 0.019] 
F-stat of IV 47.6*** 36.6*** 37.7*** 

Mother FE 
Month FE 
N 10,397 10,397 5,242 5,242 5,155 5,155 

Source:  Authors. 
Notes:   2SLS = two-stage least squares; FE = fixed effects; IV = instrumental variables; OLS = ordinary least squares. *** p < 0.01, ** p < 0.05, * p < 0.1. Correction for 

sampling weights. Sample of residents only. Confidence intervals produced by wild bootstrap (percentile-t method). 



Table 5.6 Alternative specifications with mother fixed effects 

Dependent variable: Girls’ mortality at 12 months 
Conflict events 12 Conflict exposure 12 

2SLS 2SLS 2SLS 2SLS 
(1) (2) (3) (4) 

Panel A: No controls 
Conflict 0.006** 0.004* 0.049** 0.049** 

[ 0.001, 0.010] [ 0.000, 0.008] [ 0.014, 0.083] [ 0.014, 0.083] 
F-stat of IV 34.2***  24.9*** 27.2*** 27.4*** 

Panel B: Full set of controls 
Conflict 0.006** 0.005** 0.052** 0.051** 

[ 0.001, 0.010] [ 0.000, 0.009] [ 0.006, 0.088] [ 0.006, 0.088] 
F-stat of IV 37.8*** 27.4*** 28.2*** 28.4*** 

Linear trend 
Includes nonresidents 
Mother FE 
Month FE 
N 5,155 6,680 5,155 5,155 

Source:  Authors.  
Notes:  2SLS = two-stage least squares; FE = fixed effects; IV = instrumental variables.*** p < 0.01, ** p < 0.05, * p < 0.1. 

Correction for sampling weights. Confidence intervals produced by wild bootstrap (percentile-t method). Full 
set of controls in Panel B as in Panel B of Table 5.5.  

As a final robustness test we aggregate the data at the district level and run panel regressions on 
mortality rates by gender. The results, presented in Table 5.7, are qualitatively identical to those obtained 
from cross-section regressions using individual observations. According to the estimate reported in 
column 6 of Panel A, an increase by 1 standard deviation (12.3) in conflict events would increase the 
likelihood of a girl’s dying within the first 12 months of life by about 7.1 percentage points. That would 
correspond to an increase of about 94 percent, at mean value. Such an increase is actually almost identical 
to the one obtained with the cross-sectional estimates but lower than our preferred specification with 
mother fixed effects. As in the context of cross-section analysis, these results remain unaffected by the 
inclusion of aggregated climatic variables (Panel B of Table 5.7), the introduction of district-specific time 
trends (Panel A of Table 5.8), and the use of an alternative measurement of violence (Panels B–D of 
Table 5.8). The impact of an increase by one standard deviation in violence on girls’ mortality increases 
to about 10–12 percent using the alternative measure of conflict. 

The results obtained in this section consistently suggest that considering conflict events as 
nonrandom events and consequently exploiting mineral price variations as exogenous shocks on the 
likelihood of conflicts makes a significant difference when estimating the impact of conflict on infant 
mortality. A robust pattern emerges: when we control for the endogeneity of conflict intensity, we find 
that girls are substantially more affected by violence than boys. The next section attempts to identify the 
factors behind this gender-specific effect. 

18 



Table 5.7 Panel regressions on the impact of conflict on child mortality 

Source:  Authors. 
Notes:  2SLS = two-stage least squares; FE = fixed effects; IV = instrumental variables; OLS = ordinary least squares. 

*** p < 0.01, ** p < 0.05, * p < 0.1. Sample of residents only. Standard errors clustered at the district level. Full set 
of controls in Panels B and D. 

Table 5.8 Regressions of conflict on girl mortality: Alternative panel specifications 
Dependent variable: Girls’ mortality at 12 months 

2SLS
Panel A: District, month FE, all controls, linear trends 
Conflict events 12 0.0063** 

(0.0028) 
F stat of IV 85.69*** 
N 2,876 
Panel B: District, month FE 
Conflict exposure 12 0.0439** 

(0.0210) 
F stat of IV 30.43*** 
N 2,876 
Panel C: District, month FE, all controls 
Conflict exposure 12 0.0381** 

(0.0185) 
F stat of IV 39.18*** 
N 2,876 
Panel D: District, month FE, all controls, linear trends 
Conflict exposure 12 0.0467** 

(0.0204) 
F stat of IV 62.00*** 
N 2,876 

Source:  Authors. 
Notes:  2SLS = two-stage least squares; FE = fixed effects; IV = instrumental variables.*** p < 0.01, 

** p < 0.05, * p < 0.1. Sample of residents only. Standard errors clustered at the district level. Full set of controls 
except in Panel C.   

Dependent variable:             Mortality rate at 12 months 
All Boys Girls 

OLS 2SLS OLS 2SLS OLS 2SLS 
Panel (1) (2) (3) (4) (5) (6) 
Panel A: District, month FE 

Conflict events 12 -0.0008* 0.0035 -0.0008 0.0005 -0.0001 0.0058** 

(0.0005) (0.0033) (0.0005) (0.0038) (0.0006) (0.0027) 

F-stat of IV 52.09*** 52.09*** 52.09*** 

N 2,876 2,876 2,876 2,876 2,876 2,876 

Panel B: District, month FE, all controls 
Conflict events 12 -0.0009* 0.0036 -0.0009* 0.0015 -0.0000 0.0051** 

(0.0005) (0.0030) (0.0005) (0.0036) (0.0006) (0.0025) 

F-stat of IV 62.91*** 62.91*** 62.91*** 

N 2,876 2,876 2,876 2,876 2,876 2,876 

19 



6. BEHAVIORAL VERSUS BIOLOGICAL FACTORS

Two broad classes of factors may drive these findings. First, households may discriminate against girls 
when resources within households become more limited during times of warfare. In other words, gender 
imbalances in infant mortality would be the result of behavioral factors in a situation in which sons are 
favored by parents.9 An alternative explanation relates to the biological vulnerability of boys in utero that 
would generate a strong selection effect at birth. As explained by Valente (forthcoming), gender-based 
mortality selection in utero may be explained either by the fact that a male fetus is frailer than its female 
counterpart (Shettles 1961; Kraemer 2000; Mizuno 2000; Catalano et al. 2006) or by the ability of 
females in poor conditions to favor female offspring due to the lesser variance in reproductive success of 
girls compared with boys, the so-called Trivers and Willard (1973) hypothesis. Distinguishing behavioral 
factors from biological ones is far from obvious from an empirical point of view (Garenne 2003; Mu and 
Zhang 2011). In this section, we propose various strategies to test the plausibility of the two mechanisms. 

We conjecture that the behavioral hypothesis would be consistent with several exacerbating or 
mitigating factors. The results of tests for behavioral factors are reported in Table 6.1, where only the 
coefficients of interest are presented.10 First, the literature on gender discrimination suggests that when 
the decisionmaking power is in the hands of women, such discrimination against girls should be at least 
partly corrected (Thomas 1990, 1994; Duflo 2003, 2012). Interacting the conflict measure with a dummy 
capturing female-headed households, we do not find any evidence for such a mitigating factor (Panel A of 
Table 6.1). In contrast, we could also expect widows to particularly value the survival of a male offspring 
in a context in which inheritance rules are generally discriminatory against women. If this is true, we 
should observe more discrimination against girls of widows during violence. Interacting conflict with a 
dummy for widows, however, no differentiated impact is found (Panel B). 

Consistent with the literature on sibling rivalry (Morduch 2000; Akresh and Edmonds 2011), we 
expect the composition of the household, and in particular the number of brothers and sisters to have an 
additional impact on girls’ mortality. Controlling for household size, when a girl has several brothers, we 
expect gender discrimination to be even stronger. The reverse is expected when a girl has several sisters, 
because the burden of gender discrimination would be shared among sisters. The results of these 
exercises, reported in Panels C and D, suggest once more that the behavioral hypothesis has very little 
explanatory power.11 

9 Another possibility would be that girls have been systematically targeted by armed violence. Even though gender-based 
(sexual) violence has been a dramatic phenomenon in eastern DRC (Peterman, Palermo, and Bredenkamp 2011), since we are 
dealing with infants, this is not likely to be the major mechanism explaining gender imbalances. 

10 The appendix reports the full tables for the interested reader. 
11 Controlling for the number of siblings similar to Akresh and Edmonds (2011) or using a dummy for having at least one 

brother or one sister does not change the results. 

20 



Table 6.1 Regressions of conflict on girl mortality: Behavioral factors 
Dependent variable: Girls’ mortality at 12 months 

Conflict events 12 Conflict events 12 
2SLS 

(1) 
2SLS 

(2) 
2SLS 

(3) 
2SLS 

(4) 
 

Panel A: Female-headed HH 
Conflict 0.0056** 0.0062** 0.0447** 0.0504** 

(0.0024) (0.0028) (0.0198) (0.0194) 
Conflict*Female-Headed HH -0.0008 -0.0032 0.0055 -0.0124 

(0.0031) (0.0033) (0.0208) (0.0194) 
F-stat of IV 21.65*** 53.97*** 12.48*** 21.98*** 
Panel B: Widow 
Conflict 0.0054** 0.0057** 0.0443** 0.0482** 

(0.0024) (0.0028) (0.0201) (0.0239) 
Conflict*Widow 0.0011 0.0005 0.0080 0.0043 

(0.0028) (0.0028) (0.0173) (0.0164) 
F-stat of IV 21.15*** 55.83*** 11.54*** 22.42*** 
Panel C: Number of brothers 
Conflict 0.0062** 0.0064** 0.0479** 0.0508** 

(0.0029) (0.0032) (0.0217) (0.0255) 
Conflict*# brothers -0.0007 -0.0006 -0.0025 -0.0020 

(0.0010) (0.0010) (0.0046) (0.0052) 
F-stat of IV 22.19*** 59.77*** 11.54*** 22.61*** 
Panel D: Number of sisters 
Conflict 0.0058** 0.0061* 0.0470** 0.0508* 

(0.0027) (0.0032) (0.0212) (0.0260) 
Conflict*# sisters -0.0002 -0.0002 -0.0012 -0.0014 

(0.0008) (0.0009) (0.0037) (0.0038) 
F-stat of IV 22.38*** 68.24*** 11.53*** 23.29*** 
Panel E: Son preference 
Conflict 0.0061** 0.0068** 0.0478** 0.0537** 

(0.0030) (0.0032) (0.0231) (0.0257) 
Conflict*son pref. -0.0008 -0.0009 -0.0044 -0.0044 

(0.0006) (0.0007) (0.0030) (0.0032) 
F-stat of IV 30.46*** 72.89*** 14.48*** 25.88*** 
Linear trend 
District & month FE 
Full set of controls 

Source:  Authors.  
Notes:   2SLS = two-stage least squares; FE = fixed effects; HH = household; IV = instrumental variables.
              *** p <  0.01, ** p < 0.05, * p < 0.1. Correction for sampling weights. Sample of residents only.  Standard errors 

            clustered at  the village level.  

21 



An alternative strategy to assess the role of behavioral factors consists in introducing an 
interaction term between violence and a proxy for son preference at the household level. We define boy 
preference following Jayachandran and Kuziemko (2011) in subtracting the ideal number of sons reported 
by each mother to the number of boys alive in her household. If our gender imbalance is driven by 
discrimination, we should find a stronger discrimination among households whose boy preference is 
stronger. The results of this test, in Panel E, show no additional effect due to stronger boy preference. We 
also test the robustness of these results to dividing the Jayachandran and Kuziemko (2011) proxy for son 
preference by the same indicator for daughter preference, to disentangle the preference for boys from the 
preference for large family. No evidence of stronger probability of girls’ dying within the first year of life 
is found in households with a stronger boy preference. That remains true when we adopt a different 
measure for boys’ preference, namely the ratio between the number of boys alive and the ideal number of 
sons reported by mothers. 12 Since we do not find any evidence for the existence of heterogeneous 
impacts along the dimensions that should reveal the presence of gender discrimination, we conclude that 
the behavioral hypothesis is unlikely to constitute a major explanation for the gender imbalances in infant 
mortality due to the civil war in DRC. Of course, the absence of evidence is only suggestive of an 
evidence-of-absence effect. But at least, contrary to initial expectation, the behavioral factors seem not to 
be driving the gender imbalances in mortality in times of conflict in DRC. 

The main alternative explanation for the gender-specific effect hinges on a biological hypothesis. 
As reported above, medical evidence shows that male fetuses are more vulnerable than their female 
counterparts. In particular, recent studies have provided strong evidence that the sex ratio at birth 
decreases following a worsening of the pregnancy environment, as a consequence of a civil war (Valente, 
forthcoming), terrorist attacks (Catalano et al. 2006), or prolonged economic crisis (Fernandez et al. 
2011). If this has been the case in DRC, the higher impact of violence on girls’ mortality rate may be a 
direct consequence of the selection induced by higher mortality of boys in utero. We therefore test 
whether conflict intensity affected the sex ratio at birth in the panel framework. To reduce the number of 
undefined values for the ratio when either no boy or no girl is born in a particular month, we define the 
sex ratio for each region and district as the number of boys born over the total number of births (replacing 
missing values by 0.5 when no birth is recorded for a particular month). The average ratio in our sample 
stands at 0.511, which is slightly lower than half the value of 1.03 of the sex ratio (boys over girls) at 
birth given by Anderson and Ray (2012) for the whole SSA (compared with 1.06 in developed countries). 
We find some tentative evidence, reported in Table 6.2, that the sex ratio at birth significantly decreases 
following violence experienced during pregnancy. According to the results reported in column 2 of Panel 
A, an increase by 1 standard deviation in the number of violent events experienced in utero reduces the 
sex ratio by about 14 percent, at the mean value of the sex ratio at birth. Using the alternative measure of 
conflict (months of exposure) provides very similar results (column 4). 

To further investigate the strength of the selection effect in utero, we assess the impact of 
violence experienced in utero and during the first 12 months of life on the sex ratio among one-year-old 
children (Panels B–D of Table 6.2). We first look at the impact of violence experienced while in utero on 
the sex ratio among one-year-old children. Results turn (barely) insignificant but are still negative (Panel 
B). We then assess the separate impact of violence in utero versus violence during the first 12 months of 
life on sex ratio in Panel C. Finally, in Panel D, we test the cumulative impact of violence from 
conception until the first birthday on sex ratio. The results jointly confirm the importance of violence 
experienced in utero in driving the sex ratio. The higher mortality rate among girls caused by violence 
during the first 12 months of life, as documented in this study, is only marginally compensating for the 
strong selection effect against boys in utero. 

12 Results not reported in the paper; available upon request. 

22 



Table 6.2 Regressions of conflict on girl mortality: Biological factors 
Dependent variable: Sex ratio (boys/total births) 

Source:  Authors.  
Notes:  2SLS = two-stage least squares; FE = fixed effects; IV = instrumental variables. *** p < 0.01, ** p < 0.05, * p < 0.1. 

Correction for sampling weights. Sample of residents only.  Standard errors clustered at the village level. 

In the absence of evidence for behavioral factors, this analysis suggests that gender imbalances in 
infant mortality are at least partially driven by the selection induced by higher vulnerability of boys in 
utero. 

Conflict events 12 Conflict exposure 12 
2SLS 2SLS 2SLS 2SLS 
(1) (2) (3) (4) 

Panel A: Sex ratio at birth 

Conflict in utero -0.0068 -0.0071* -0.0540 -0.0553* 
(0.0041) (0.0041) (0.0341) (0.0331) 

F-stat of IV 26.07*** 32.99*** 43.16*** 49.56*** 
N 2,598 2,598 2,598 2,598 

Panel B: Sex ratio at 12 months—impact in utero 

Conflict in utero -0.0055 -0.0062 -0.0442 -0.0484 
(0.0043) (0.0043) (0.0351) (0.0345) 

F-stat of IV 26.07*** 32.99*** 43.16*** 49.56*** 
N 2,598 2,598 2,598 2,598 

Panel C: Sex ratio at 12 months—separate impacts in utero and first year 

Conflict in utero -0.0058 -0.0068 -0.0444 -0.0455 
(0.0063) (0.0061) (0.0343) (0.0301) 

Conflict 0.0007 0.0019 -0.0014 0.0045 
(0.0074) (0.0073) (0.0362) (0.0364) 

F-stat of IV (in utero) 29.37*** 20.87*** 21.20*** 30.54*** 
F-stat of IV (Conflict) 13.84*** 35.25*** 20.85*** 27.71*** 
N 2,598 2,598 2,598 2,598 

Panel D: Sex ratio at 12 months—total impact in utero and first year 

Conflict in utero & 12 months -0.0056** -0.0052** -0.0568* -0.0536** 
(0.0026) (0.0025) (0.0292) (0.0273) 

F-stat of IV 29.69*** 39.41*** 15.05*** 20.41*** 
N 2,291 2,291 2,291 2,291 

Linear trend  
District and month FE 

23 



7. CONCLUSIONS

In this paper we analyzed the impact on infant mortality of the armed conflict afflicting the DRC from 
1997 to 2004. This study differs from existing microlevel studies in a major way. Relying on a credible 
instrumental variable approach, we control for the nonrandom timing and location of conflict violence. 
This is particularly relevant when we exploit within-district variations inasmuch as we show that our 
instrumental variables results significantly differ from our ordinary least squares findings, raising some 
concerns about potential bias in the existing studies relying on geographical variation in exposure to 
conflicts. 

We find that experiencing violence substantially increases mortality rates among infants, but only 
for girls. This pattern is robust to many different specifications, controlling for district and month fixed 
effects, rainfall anomalies, and mother fixed effects. Why does violence affect especially girls’ mortality? 
Two broad classes of factors could explain the gender imbalance we uncover: behavioral and biological. 
According to the former, girls may be more discriminated against by households faced with difficult 
circumstances in times of conflict, to safeguard their male offspring. The latter, instead, would ascribe to 
purely biological factors the gender-specific resilience to conflict. For instance, boys and girls may 
feature different resistance levels to negative shocks. 

We adopt several strategies to assess the different roles of behavioral versus biological factors in 
explaining our result. We find no support for gender discrimination to be driving the higher mortality 
rates among girls. Instead, we find suggestive evidence that gender imbalances in infant mortality in times 
of warfare are mainly driven by biological factors. More specifically, we show that violence is more fatal 
for male fetuses than for their female counterparts. In turn, the higher vulnerability of boys in utero 
induces a selection in the sample of children born in conflict-affected regions. Overall, our results suggest 
that more attention should be paid to understanding possible selection in utero in studies assessing the 
impact of shocks in early life. 

Our analysis also delivers a critical policy recommendation. Gender-specific warfare damages 
have sometimes led scholars and policymakers to call for gender-based targeted interventions. Although 
they may be grounded on good motives and may help in reducing gender discrimination in general, our 
study warns that these policies may miss their targets if they fail to account for the possible selection in 
utero. In other words, despite providing some evidence of gender imbalances in infant mortality, our 
paper suggests that any policy should be drawn on a sound understanding of the sources of such gender 
bias. As biological factors in utero are found to be a more prominent explanation than the standard 
behavioral factors, our paper resets the priority to policies aiming at enhancing the resilience of (pregnant) 
women to violent experiences. Policies ensuring high coverage of multiple micronutrient supplementation 
and other nutrition-sensitive programs directly addressing pregnant women, family planning to delay the 
age of first pregnancy, or educational interventions designed to increase spacing between births may 
therefore all prove comparatively more effective in reducing infant mortality in times of violence than 
policies targeting gender discrimination during the first year of life (Black et al. 2013; Bhutta et al. 2013; 
Ruel and Alderman 2013; WHO 2014). 

24 



APPENDIX: SUPPLEMENTARY TABLES 

Table A.1 Regressions of conflict on girl mortality: Interaction with female-headed HH 

Sample of residents Whole sample 

OLS 2SLS 2SLS 2SLS 2SLS 2SLS 
Variable [1] [2] [3] [4] [5] [6] 
Panel A: 
Conflict events -0.0000 

(0.0008) 
0.0052 

**(0.0023) 
0.0061 

**(0.0024) 
0.0035 * 
(0.0018) 

Conflict*female-headed HH 0.0011 
(0.0018) 

0.0003  
(0.0033)- 

0.0025  
(0.0035) 

-0.0014  
(0.0029) 

Female-headed HH 0.0006 
(0.0240) 

0.0093  
(0.0347) 

0.0171  
(0.0351) 

-0.0055  
(0.0520) 

0.0201  
(0.0512) 

0.0212  
(0.0329) 

Exposure to conflict 0.0427 ** 
(0.0194) 

0.0480 
**(0.0196) 

0.0073 
(0.0207) 

-0.0090 
(0.0205) 

Exposure*female-headed HH 

F-stat of IV 20.19*** 64.18*** 12.25*** 10.34*** 17.24*** 
District and month FE yes yes yes yes yes yes 
Linear trend – – yes – yes – 
N 5,155 5,155 5,155 5,155 5,155 6,080 
Panel B: 
Conflict events 0.0000 

(0.0008) 
0.0056 

**(0.0024) 
0.0062 

**(0.0028) 
0.0039 ** 
(0.0019) 

Conflict*female-headed HH 0.0008 
(0.0017) 

-0.0008 
(0.0031) 

-0.0032 
(0.0033) 

-0.0019 
(0.0028) 

Female-headed HH -0.0172 
(0.0242) 

-0.0059 
(0.0350) 

0.0024 
(0.0351) 

-0.0205 
(0.0533) 

0.0070 
(0.0514) 

0.0062 
(0.0335) 

Exposure to conflict 0.0447 
**(0.0198) 

0.0504 
**(0.0239) 

0.0055 
(0.0208) 

-0.0124 
(0.0194) 

Exposure*female-headed HH 

F-stat of IV 21.65** 53.97*** 12.48*** 21.98*** 20.07*** 
District and month FE yes yes yes yes yes yes 
Control yes yes yes yes yes yes 

yes yes yes yes yes yes Ethnicity (10) and religion (9) 

Linear trend – – yes – yes – 
N 5,155 5,155 5,155 5,155 5,155 6,080 

Source:  Authors. 
Notes:    HH = household; OLS = ordinary least square; 2SLS = two-stage least squares; FE = fixed effects; IV = instrumental 

variables; N = number. Correction for sampling weights. Standard errors clustered at the village level.

25 



Table A.2 Regressions of conflict on girl mortality: Interaction with mother’s being a widow 

Sample of residents Whole sample 

OLS 2SLS 2SLS 2SLS 2SLS 2SLS 
Variable [1] [2] [3] [4] [5] [6] 

Panel A: 
Conflict events -0.0001 

(0.0007) 
0.0050 

**(0.0024) 
0.0058 

**(0.0024) 
0.0034* 
(0.0018) 

Conflict*widow 0.0017 
(0.0018) 

0.0009  
(0.0031) 

0.0004  
(0.0030) 

-0.0003  
(0.0025) 

Widow 0.0048(0.0265) 0.0073  
(0.0334) 

0.0088  
(0.0334) 

0.0036  
(0.0435) 

0.0082  
(0.0431) 

0.0249  
(0.0332) 

Exposure to conflict 0.0423 
**(0.0197) 

0.0467 
**(0.0199) 

Exposure*widow 0.0062 
(0.0184) 

0.0037 
(0.0175) 

F-stat of IV 19.31*** 63.65*** 10.39*** 33.08*** 16.86*** 
District and month FE yes yes yes yes yes yes 
Linear trend – – yes – yes – 
N 5,155 5,155 5,155 5,155 5,155 6,080 
Panel B: 
Conflict events -0.0001 

(0.0008) 
0.0054 

**(0.0024) 
0.0057 

**(0.0028) 
0.0038 

**(0.0019) 
Conflict*widow 0.0016 

(0.0018) 
0.0011 

(0.0028) 
0.0005 

(0.0028) 
-0.0003 
(0.0025) 

Widow 0.0142(0.0252) 0.0017 
(0.0290) 

0.0092 
(0.0293) 

-0.0037 
(0.0387) 

0.0102 
(0.0382) 

0.0215(0.0314) 

Exposure to conflict 0.0443 
**(0.0201) 

0.0482 
**(0.0239) 

Exposure*widow 0.0080 
(0.0173) 

-0.0043 
(0.0164) 

F-stat of IV 21.15*** 55.83*** 11.54*** 22.42*** 20.12*** 
District and month FE yes yes yes yes yes yes 
Control yes yes yes yes yes yes 

yes yes yes yes yes yes Ethnicity (10) and religion (9) 

Linear trend – – yes – yes – 
N 5,155 5,155 5,155 5,155 5,155 6,080 

Source:  Authors. 
Notes:      HH = household; OLS = ordinary least square; 2SLS = two-stage least squares; FE = fixed effects; IV = instrumental 

variables; N = number. Correction for sampling weights. Standard errors clustered at the village level. 

26 



Table A.3 Regressions of conflict on girl mortality: Interaction with number of brothers 

Sample of residents Whole sample 

OLS 2SLS 2SLS 2SLS 2SLS 2SLS 
Variable [1] [2] [3] [4] [5] [6] 

Panel A:

Conflict events 0.0005 
(0.0009) 

0.0061 ** 
(0.0028) 

0.0065 
**(0.0029) 

0.0043** 
(0.0022) 

Conflict*# brothers -0.0002 
(0.0003) 

-0.0009  
(0.0009) 

-0.0008  
(0.0010) 

-0.0009  
(0.0008) 

# brothers 0.0013 
(0.0045) 

0.0054  
(0.0074) 

0.0042  
(0.0074) 

0.0072  
(0.0105) 

0.0060  
(0.00109) 

0.0060  
(0.0067) 

Exposure to conflict 0.0468 ** 
(0.0214) 

0.0498** 
(0.0214) 

Exposure*# brothers -0.0036 
(0.0047) 

-0.0031 
(0.0050) 

F-stat of IV 19.52*** 67.89*** 10.45*** 33.43*** 16.68*** 
District and month FE yes yes yes yes yes yes 
Linear trend – – yes – yes – 
N 5,155 5,155 5,155 5,155 5,155 6,080 
Panel B: 
Conflict events 0.0004 

(0.0010) 
0.0062** 
(0.0029) 

0.0064 
**(0.0032) 

0.0045 
**(0.0023) 

Conflict*widow -0.0002 
(0.0003) 

-0.0007 
(0.0010) 

-0.0006 
(0.0010) 

-0.0006 
(0.0008) 

Widow 0.0022 
(0.0049) 

0.0056 
(0.0075) 

0.0046 
(0.0078) 

0.0066 
(0.0104) 

0.0053 
(0.0111) 

0.0046 
(0.0067) 

Exposure to conflict 0.0479 ** 
(0.0217) 

0.0508 ** 
(0.0255) 

Exposure*widow -0.0025 
(0.0046) 

-0.0020 
(0.0052) 

F-stat of IV 22.19*** 59.77*** 11.54*** 22.61*** 20.05*** 
District and month FE yes yes yes yes yes yes 
Control yes yes yes yes yes yes 

yes yes yes yes yes yes Ethnicity (10) and religion (9) 

Linear trend – – yes – yes – 
N 5,155 5,155 5,155 5,155 5,155 6,080 

Source:  Authors. 
Notes:  HH = household; OLS = ordinary least square; 2SLS = two-stage least squares; FE = fixed effects; IV = instrumental 

variables; N = number. Correction for sampling weights. Standard errors clustered at the village level. 

27 



Table A.4 Regressions of conflict on girl mortality: Interaction with number of sisters 
Sample of residents      Whole sample 

OLS 2SLS 2SLS 2SLS 2SLS 2SLS 
Variable [1] [2] [3] [4] [5] [6] 

Panel A: 
Conflict events 0.0004 

(0.0009) 
0.0055** 
(0.0026) 

0.0061**(0.0029) 0.0043** 
(0.0022) 

Conflict*# sisters -0.0002 
(0.0003) 

-0.0003  
(0.0009) 

-0.0002  
(0.0009) 

-0.0009  
(0.0008) 

# sisters 0.0033 
(0.0045) 

0.0042  
(0.0074) 

0.0044  (0.0072) 0.0056  
(0.0097) 

0.0066  
(0.00097) 

0.0051 
(0.0072) 

Exposure to conflict 0.0453** 
(0.0208) 

0.0498** 
(0.0214) 

Exposure*# sisters -0.0036 
(0.0047) 

-0.0031 
(0.0050) 

F-stat of IV 20.33*** 73.90*** 10.37*** 33.82*** 18.92*** 
District and month FE yes yes yes yes yes yes 
Linear trend – – yes – yes – 
N 5,155 5,155 5,155 5,155 5,155 6,080 
Panel B: 
Conflict events 0.0003 

(0.0010) 
0.0058** 
(0.0027) 

0.0061*(0.0032) 0.0039* 
(0.0021) 

Conflict*# sisters -0.0001 
(0.0003) 

-0.0002 
(0.0008) 

-0.0002 (0.0009) -0.0001 
(0.0008) 

# sisters 0.0080 
(0.0051) 

0.0090 
(0.0076) 

0.0100 (0.0077) 0.0102 
(0.0097) 

0.0122 
(0.0102) 

0.0087 
(0.0072) 

Exposure to conflict 0.0470** 
(0.0212) 

0.0508* 
(0.0260) 

Exposure*# sisters -0.0012 
(0.0037) 

-0.0014 
(0.0038) 

F-stat of IV 22.38*** 68.24*** 11.53*** 23.29*** 23.83*** 
District and month FE yes yes yes yes yes yes 
Control yes yes yes yes yes yes 

yes yes yes yes yes yes Ethnicity (10) and religion (9) 

Linear trend – – yes – yes – 
N 5,155 5,155 5,155 5,155 5,155 6,080 

Source:  Authors. 
Notes:  HH = household; OLS = ordinary least square; 2SLS = two-stage least squares; FE = fixed effects; IV = instrumental 

variables; N = number. Correction for sampling weights. Standard errors clustered at the village level. 

28 



Table A.5 Regressions of conflict on girl mortality: Interaction with a proxy for son preference 

Sample of residents Whole sample 

OLS 2SLS 2SLS 2SLS 2SLS 2SLS 
Variable [1] [2] [3] [4] [5] [6] 

Panel A: 
Conflict events 0.0004 

(0.0007) 
0.0054* 
(0.0028) 

0.0064** 
(0.0029) 

0.0038** 
(0.0021) 

Conflict*son pref. -0.0000 
(0.0002) 

-0.0008  
(0.0007) 

-0.0008  
(0.0007) 

-0.0007  
(0.0006) 

Son preference 0.0023 
(0.0031) 

0.0062  
(0.0054) 

0.0061  
(0.0055) 

0.0096  
(0.0074) 

0.0066  
(0.00077) 

0.0067 
(0.0049) 

Exposure to conflict 0.0422* 
(0.0218) 

0.0498** 
(0.0213) 

Exposure*son pref. -0.0043 
(0.0032) 

-0.0043 
(0.0034) 

F-stat of IV 29.58*** 81.44*** 13.90*** 35.83*** 32.58*** 
District and month FE yes yes yes yes yes yes 
Linear trend – – yes – yes – 
N 4,726 4,726 4,726 4,726 4,726 5,556 
Panel B: 
Conflict events 0.0004 

(0.0007) 
0.0061** 
(0.0030) 

0.0068** 
(0.0032) 

0.0044* 
(0.0022) 

Conflict*son pref. 0.0000 
(0.0002) 

-0.0008 
(0.0006) 

-0.0009 
(0.0007) 

-0.0008 
(0.0005) 

Son preference 0.0017 
(0.0041) 

0.0053 
(0.0058) 

0.0052 
(0.0060) 

0.0082 
(0.0075) 

0.0083 
(0.0079) 

0.0064 
(0.0054) 

Exposure to conflict 0.0478** 
(0.0231) 

0.0537** 
(0.0257) 

Exposure*son pref. -0.0044 
(0.0030) 

-0.0044 
(0.0032) 

30.46*** 72.89*** 14.48*** 25.88*** 35.19*** 
yes yes yes yes yes yes 
yes yes yes yes yes yes 
yes yes yes yes yes yes 

– – yes – yes – 

F-stat of IV 
District and month FE 
Control 
Ethnicity (10) and religion (9) 

Linear trend 
N 4,726 4,726 4,726 4,726 4,726 5,556 

Source:  Authors. 
Notes:  HH = household; OLS = ordinary least square; 2SLS = two-stage least squares; FE = fixed effects; IV = instrumental 

variables; N = number. Correction for sampling weights. Standard errors clustered at the village level. 

29 



Figure A.1 Placebo test on girls’ sample 

Source:  Authors. 

30 



REFERENCES 

Aguero, J., and A. Deolalikar. 2012. “Late Bloomers? Identifying Critical Periods in Human Capital Accumulation. 
Evidence from the Rwanda Genocide.” Unpublished. 

Akresh, R., S. Bhalotra, M. Leone, and U. Osili. 2012a. “War and Stature: Growing up during the Nigerian Civil 
War.” American Economic Review: Papers and Proceedings 102 (3): 273–277. 

Akresh, R., and E. V. Edmonds. 2011. “Sibling Rivaly, Residential Rivalry and Constraints on the Availability of 
Child Labor.” NBER Working Paper 17165. 

Akresh, R., L. Lucchetti, and H. Thirumurthy. 2012b. “Wars and civil health: Evidence from the Eritrean-Ethiopian 
Conflict. Journal of Development Economics 99 (2): 330–340. 

Alderman, H., J. Hoddinott, and B. Kinsey. 2006. “Long Term Consequences of early Childhood Malnutrition.” 
Oxford Economic Papers 58: 450–474. 

Almond, D., and J. Currie. 2011. “Killing Me Softly: The Fetal Origins Hypothesis.” The Journal of Economic 
Perspectives 25 (3): 153–172. 

Anderson, S., and D. Ray. 2012. “Missing Women: Age and Disease.” Review of Economic Studies 77: 1262–1300. 

Behrman, J. R., and B. L. Wolfe. 1987. “How Does Mother’s Schooling Affect Family Health, Nutrition, Medical 
Care Usage, and Household Sanitation?” Journal of Econometrics 36 (1–2): 185–204. 

Bhutta, Z. A., J. K. Das, A. Rizvi, M. F. Gaffey, N. Walker, S. Horton, P. Webb, A. Lartey, and R. E. Black. 2013. 
“Evidence-Based Interventions for Improvement of Maternal and Child Nutrition: What Can be Done and 
at What Cost?” The Lancet 382 (9890): 452–477. 

Black, R. E., C. G. Victora, S. P. Walker, Z. A. Bhutta, P. Christian, M. de Onis, M. Ezzati, S. Grantham-McGregor, 
J. Katz, R. Martorell, and R. Uauy. 2013. “Maternal and Child Undernutrition and Overweight in Low-
Income and Middle-Income Countries.” The Lancet 382 (9890): 427–451. 

Blattman, C., and J. Annan. 2010. “The Consequences of Child Soldiering.” The Review of Economics and 
Statistics, 92 (4): 882–898. 

Bruckner, M., and A. Ciccone. 2010. “International Commodity Prices, Growth and the Outbreak of Civil War in 
Sub-Saharan Africa.” Economic Journal 120: 519–534. 

Bundervoet, T. 2010. “Assets, Activity Choices, and Civil War: Evidence from Burundi.” World Development, 38 
(7): 955–965. 

Bundervoet, T., P. Verwimp, and R. Akresh. 2009. “Health and Civil War in Rural Burundi.” Journal of Human 
Resources, 44(2): 536–563. 

Camacho, A. 2008. “Stress and Birth Weight: Evidence from Terrorist Attacks.” The American Economic Review: 
Papers and Proceedings 98(2): 511– 515. 

Catalano, R., T. Bruckner, A. Marks, and B. Eskenazi. 2006. “Exogenbous Shocks to the Human Sex Ratio: The 
Case of September 11, 2001 in New York City.” Human Reproduction 21 (12): 3127–3131. 

Chen, S., N. V. Loayza, and M. Reynal-Querol. 2008. “The Aftermath of Civil War.” The World Bank Economic 
Review, 22. (1): 63–85. 

Coghlan, B., R. J. Brennan, P. Ngoy, D. Dofara, B. Otto, M. Clements, and T. Stewart. 2006. “Mortality in the 
Democratic Republic of Congo: A Nationwide Survey.” Lancet 367: 44–51. 

Congdon Fors, H., and O. Olsson. 2004. “Congo: The Size of Predation.” Journal of Peace Research 41: 321–336. 

Danish Epidemiology Science Centre. 1999. “Nutritional Status and Mortality of Refugee and Resident Children in 
a Non-Camp Setting during Conflict: Follow-Up Study in Guinea-Bissau.” British Medical Journal, 319 
(7214): 878–881. 

Davis, D., and J. Kuritsky. 2002. “Violent Conflict and Its Impact on Health Indicators in Sub-Saharan Africa, 1980 
to 1997.” Paper presented at the annual meeting of the International Studies Association, New Orleans, LA. 

31 



de Walque, D. 2005. “Selective Mortality during the Khmer Rouge Period in Cambodia.” Population and 
Development Review 31 (2): 351–368. 

Domingues, P., and T. Barre. 2013. “The Health Consequences of the Mozambican Civil War: An Anthropometric 
Approach.” Economic Development and Cultural Change 61 (4): 755–788. 

Duflo, E. 2003. “Grandmothers and Granddaughters: Old-Age Pensions and Intrahousehold Allocation in South 
Africa.” World Bank Economic Review 17 (1): 1–25. 

———.  2012. Women Empowerment and Economic Development.” Journal of Economic Literature 50 (4): 1051–
1079. 

Fernandez, S., R. Medina, J. Britton, and A. Fogarty. 2011. “The Association between Living through a Prolonged 
Economic Depression and the Male: Female Birth Ratio: A Longitudinal Study from Cuba, 1960-2008.” 
American Journal of Epidemiology 174 (12): 1327–1331. 

Gambino, T. 2011. Democratic Republic of Congo,” In Background Case Study to the World Development Report 
2011: Conflict, Security and Development. Washington, DC: World Bank, 

Garenne, M. 2003. “Sex Differences in Health Indicators among Children in African DHS Surveys.” Journal of 
Biosocial Science 35: 601–614. 

Ghobarah, H. A., P. Huth, and B. Russett. 2003. “Civil Wars Kill and Maim People Long After the Shooting Stops.” 
American Political Science Review, 97 (2), 189–202. 

Goma Epidemiology Group. 1995. “Public Health Impact of Rwandan Refugee Crisis: What Happened in Goma 
Zaire, in July 1994?” Lancet 345 (8946): 339–344. 

Guha-Sapir, D., and O. D’Aoust. 2010. “Demographic and Health Consequences of Civil Conflict.” World 
Development Report 2011 Background Paper, World Bank. 

Guha-Sapir, D., and W. G. van Panhuis. 2004. “Conflict-related Mortality: An Analysis of 37 Datasets.” Disasters 
28 (4): 418–428. 

Guha-Sapir, D., W. G. van Panhuis, O. Degomme, and V. Teran. 2005. “Civil Conflicts in Four African Countries: 
A Five-Year Review of Trends in Nutrition and Mortality.” Epidemiologic Reviews 27: 67–77. 

Hsiang, S., M. Burke, and E. Miguel. 2013. “Quantifying the Influence of Climate on Human Conflict.” Science, 
342 (6151): 1235367. 

Internal Displacement Monitoring Center. 2011. “IDPs Need Further Assistance in Context of Continued Attacks 
and Insecurity.” Accessed March 2011. www.internal-displacement.org/sub-saharan-africa/democratic-
republic-of-the-congo/2011/idps-need-further-assistancein-context-of-continued-attacks-and-insecurity. 

International Rescue Committee. 2011.  “IRC Study Reveals 31,000 Die Monthly in Congo Conflict and 3.8 Million 
Died in Past Six Years. When Will the World Pay Attention?” Accessed March 2011. 
www.rescue.org/news/irc-study-reveals-31000-die-monthly-congo-conflict-and-38-million-died-past-six-
years-when-will-3538. 

Jayachandran, S., and I. Kuziemko. 2011. “Why Do Mothers Breastfeed Girls less than Boys? Evidence and 
Implications for Child Health in India.” Quarterly Journal of Economics 126 (3): 1485–1538. 

Jenkins, J. C., S. J. Scanlan, and L. Peterson. 2007. “Military Famine, Human Rights, and Child Hunger: A Cross-
National Analysis, 1990-2000.” The Journal of Conflict Resolution, 51 (6), 823–847. 

Kiros, G., and D. Hogan. 2001. “War, Famine and Excess Child Mortality in Africa: The Role of Parental 
Education.” International Journal of Epidemiology 30: 447–455. 

Kraemer, S. 2000. “The Fragile Male.” British Medical Journal 321: 1609–1612. 

Kudamatsu, M., T. Persson, and D. Stromberg. 2012. “Weather and Infant Mortality in Africa.” CEPR Discussion 
Paper 9222. 

Mansoor, H., and D. Rees. 2012. “Armed Conflict and Birth Weight: Evidence from the Al-Aqsa Intifada.” Journal 
of Development Economics 99: 190–199. 

32 
 

http://www.internal-displacement.org/sub-saharan-africa/democratic-republic-of-the-congo/2011/idps-need-further-assistancein-context-of-continued-attacks-and-insecurity.
http://www.internal-displacement.org/sub-saharan-africa/democratic-republic-of-the-congo/2011/idps-need-further-assistancein-context-of-continued-attacks-and-insecurity.
http://www.rescue.org/news/irc-study-reveals-31000-die-monthly-congo-
http://www.rescue.org/news/irc-study-reveals-31000-die-monthly-congo-


Matsuura, K., and C. Willmott. 2009. “Terrestrial Air Temperature: 1900–2008 Gridded Monthly Time Series 
(version 2.01).” Accessed March 2011. http://climate.geog.udel.edu/. 

Médecins Sans Frontières. 2003. “Mortality among Displaced Former UNITA Members and their Families in 
Angola: A Retrospective Cluster Survey.” British Medical Journal 327 (7416): 650–660. 

Minoiu, C., and O. Shemyakina. 2012. “Child Health and Conflict in Cote d’Ivoire.” American Economic Review: 
Papers and Proceedings 102 (3): 294–299. 

Mizuno, R. 2000. “The Male/Female Ratio of Fetal Deaths and Births in Japan.” Lancet 356: 738–739. 

Morduch, J. 2000. “Sibling Rivalry in Africa.” American Economic Review 90 (2): 405–409. 

Mu, and Zhang. 2011. “Why does the Great Chinese Famine Affect the Male and Female Survivors Differently? 
Mortality Selection versus Son Preference.” Economics and Human Biology 9: 92–105. 

Peterman, A., T. Palermo, and Bredenkamp. 2011. “Estimates and Determinants of Sexual Violence against Women 
in the Democratic Republic of Congo.” American Journal of Public Health 101 (6): 1060–1067. 

Prunier, G. 2009. Africa’s World Wars: Congo, the Rwandan Genocide, and the Making of a Continental 
Catastrophe. Oxford: Oxford University Press. 

Raleigh, C., A. Linke, H. Hegre, and J. Karlsen. 2010. “Introducing ACLED: An Armed Conflict Location and 
Event Dataset.” Journal of Peace Research 47 (5): 651–660. 

Restrepo, J. A., M. Spagat, and J. F. Vargas. 2006. “Special Data Feature; the Severity of the Colombian Conflict: 
Cross-Country Datasets versus New Micro-Data.” Journal of Peace Research 43(1): 99–115. 

Ruel, M. T., and H. Alderman. 2013. “Nutrition-Sensitive Interventions and Programmes: How Can they Help to 
Accelerate Progress in Improving Maternal and Child Nutrition?” The Lancet, 382 (9891): 536–551. 

Sanchez de la Sierra, R. 2013. “On the Origin of States: Stationary Bandits and Taxation in Eastern Congo.” SSRN 
Working Paper. http://ssrn.com/abstract=2358701. 

Shettles, L. 1961. “Conception and Birth Sex Ratios.” Obstetrics and Gynecology 18: 122–130. 

Singh, K., U. Karunakara, G. Burnham, and H. Kenneth. 2005. “Forced Migration and Under-five Mortality: A 
Comparison of Refugees and Hosts in Northwestern Uganda and Southern Sudan.” European Journal of 
Population 21: 247–270. 

Stearns, J. 2011. Dancing in the Glory of Monsters: The Collapse of the Congo and the Great War of Africa. New 
York: Public Affairs. 

Strauss, J., and D. Thomas. 1995. “Human Resources: Empirical Modeling of Household and Family Decisions” In 
The Handbook of Development Economics, edited by J. Behrman and T.N. Srinivasan, 64–83. Amsterdam: 
Elsevier. 

Thomas, D. 1990. “Intra-household Resource Allocation: An Inferential Approach.” Journal of Human Resources, 
25 (4): 635–664. 

———.  1994. “Like Father, Like Son or Like Mother, Like Daughter: Parental Education and Child Health.” 
Journal of Human Resources 29 (4): 950–989. 

Toole, M., S. Galson, and W. Brady. 1993. “Are War and Public Health Compatible?” Lancet 341: 1193–1196. 

Trivers, R., and D. Willard. 1973. “Natural Selection of Parental Ability to Vary the Sex Ratio of Offsprings.” 
Science 179: 90. 

Turner, T. 2007. The Congo Wars: Conflict, Myth and Reality. London: Zed Books. 

Valente, C. 2014. “Civil Conflict, Gender-Specific Fetal Loss, and Selection: A New Test of the Trivers-Willard 
Hypothesis.” Journal of Development Economics, forthcoming. 

Verpoorten, M. 2009. “Cattle Sales in War and Peacetime: A Study of Household Coping Rwanda, 1991-2001.” 
Journal of Development Economics 188 (1): 67–86. 

33 

http://climate.geog.udel.edu/
http://ssrn.com/abstract%3D2358701


2012. “Detecting Hidden Violence: The Spatial Distribution of Excess Mortality in Rwanda.” Political Geography, 
31 (1): 44–56. 

Verwimp, P., and J. Van Bavel. 2005. “Child Survival and Fertility of Refugees in Rwanda.” European Journal of 
Population 21: 271–290. 

Vlassenroot, K., and T. Raeymaekers. 2004. “The Politics of Rebellion and Intervention in Ituri: The Emergence of 
a New Political Complex?” African Affairs 103: 385–412. 

Waldron, I. 1998. “Sex Differences in Infant and Early Child Mortality: Major Causes of Death and Possible 
Biological Causes” In Too Young to Die: Genes or Gender? 64–83. New York: United Nations. 

Wilson, E. 1975. Sociobiology: The New Synthesis. Cambridge: Harvard University Press. 

World Health Organization. 2014. Public Health Risk Assessment and Interventions. Conflict and Humanitarian 
Crisis, Republic of South Sudan. Geneva.

34 
 





 
 



RECENT IFPRI DISCUSSION PAPERS 

For earlier discussion papers, please go to www.ifpri.org/pubs/pubs.htm#dp. 
All discussion papers can be downloaded free of charge. 

1373. Variable returns to fertilizer use and its relationship to poverty: Experimental and simulation evidence from Malawi. 
Aurélie Harou, Yanyan Liu, Christopher B. Barrett, and Liangzhi You, 2014. 

1372. Drought risk reduction in agriculture: A review of adaptive strategies in East Africa and the Indo-Gangetic Plain of 
South Asia. Nicola Cenacchi, 2014. 

1371. The political economy of MGNREGS spending in Andhra Pradesh. Megan Sheahan, Yanyan Liu, Christopher B. Barrett, 
and Sudha Narayanan, 2014. 

1370. Assessing progress made toward shared agricultural transformation objectives in Mozambique. Todd Benson, Tewodaj 
Mogues, and Sileshi Woldeyohannes, 2014. 

1369. Strategies to control aflatoxin in groundnut value chains. Wojciech J. Florkowski and Shashidhara Kolavalli, 2014. 

1368. The agrarian reform experiment in Chile: History, impact, and implications. Alberto Valdés and William Foster, 2014. 

1367. What dimensions of women’s empowerment in agriculture matter for nutrition-related practices and outcomes in Ghana? 
Hazel Jean L. Malapit and Agnes R. Quisumbing, 2014. 

1366. The Economywide effects of teff, wheat, and maize production increases in Ethiopia: Results of economywide modeling. 
Todd Benson, Ermias Engida, and James Thurlow, 2014. 

1365. Identifying agricultural expenditures within the public financial accounts and coding system in Ghana: Is the ten percent 
government agriculture expenditure overestimated? Samuel Benin, 2014. 

1364. Environmental migration and labor markets in Nepal. Jean-François Maystadt, Valerie Mueller, Ashwini Sebastian, 
2014. 

1363. The role of agriculture in the fast-growing Rwandan economy: Assessing growth alternatives. Xinshen Diao, Godfrey 
Bahiigwa, Angga Pradesha, 2014. 

1362. Understanding the context for agriculture for nutrition research: Identifying country typologies of child-stunting 
determinants. Eduardo Maruyama, Laurian J. Unnevehr, and Jiarui Wang, 2014. 

1361.  Assessing the economic benefits of sustainable land management practices in Bhutan. Ephraim Nkonya, Raghavan 
Srinivasan, Weston Anderson, and Edward Kato, 2014. 

1360. Examining the sense and science behind Ghana’s current blanket fertilizer recommendation. Antony Chapoto and Francis 
Tetteh, 2014. 

1359. Aid effectiveness in Ghana: How’s the L’Aquila Food Security Initiative Doing? Samuel Benin, Tsitsi Makombe, and 
Michael Johnson, 2014. 

1358. The other Asian enigma: Explaining the rapid reduction of undernutrition in Bangladesh. Derek D. Headey, John 
Hoddinott, Disha Ali, Roman Tesfaye, and Mekdim Dereje, 2014. 

1357. Can market-based approaches to technology development and dissemination benefit women smallholder farmers?: A 
qualitative assessment of gender dynamics in the ownership, purchase, and use of irrigation pumps in Kenya and 
Tanzania. Jemimah Njuki, Elizabeth Waithanji, Beatrice Sakwa, Juliet Kariuki, Elizabeth Mukewa, John Ngige, 2014. 

1356. The impact of shocks on gender-differentiated asset dynamics in Bangladesh. Muntaha Rakib and Julia Anna Matz, 2014. 

1355. Tractor owner-operators in Nigeria: Insights from a small survey in Kaduna and Nasarawa States. Hiroyuki Takeshima, 
Hyacinth Edeh, Akeem Lawal, and Moshud Isiaka, 2014. 

1354. Co-movement of major commodity price returns: A time-series assessment. Francesca de Nicola, Pierangelo De Pace, and 
Manuel A. Hernandez, 2014. 

1353.  Bargaining power and biofortification: The role of gender in adoption of orange sweet potato in Uganda. Daniel O. 
Gilligan, Neha Kumar, Scott McNiven, J. V. Meenakshi, and Agnes Quisumbing, 2014. 

1352. Quality healthcare and health insurance retention: Evidence from a randomized experiment in the Kolkata Slums. Clara 
Delavallade, 2014. 

http://www.ifpri.org/pubs/pubs.htm%23dp


 

INTERNATIONAL FOOD POLICY  
RESEARCH INSTITUTE 

www.ifpri.org  

IFPRI HEADQUARTERS 
2033 K Street, NW 
Washington, DC 20006-1002 USA  
Tel.: +1-202-862-5600 
Fax: +1-202-467-4439 
Email: ifpri@cgiar.org 
 

 
 

mailto:ifpri@cgiar.org

	Abstract
	Acknowledgments
	1.  Introduction
	2.  Historical Background
	3.  Data Sources and Sample Construction
	4.  Empirical Strategy
	Conflict
	Infant Mortality
	Cross-section Regressions

	Panel Regressions

	5.  Results
	Conflict
	Infant Mortality

	6.  Behavioral versus Biological Factors
	7.  Conclusions
	Appendix: Supplementary Tables
	References
	RECENT IFPRI DISCUSSION PAPERS