World Development Vol. 54, pp.

139–155, 2014
Ó 2013 Elsevier Ltd. All rights reserved.
0305-750X/$ - see front matter
www.elsevier.com/locate/worlddev
http://dx.doi.org/10.1016/j.worlddev.2013.08.004

Long-Term Effects of Civil Conflict on Women’s

Health Outcomes in Peru
F. GRIMARD and S. LASZLO *
McGill University, Montreal, Canada
Summary. — Peru’s internal conflict resulted in over 69,000 deaths and disappearances from 1980 to 2000. We investigate the long-term
health effects on women exposed to this conflict in utero and in early life. Utilizing recent Demographic and Health Surveys (DHSs) and
district-level conflict data, we find that exposure in utero has long lasting impacts on a woman’s height (an indicator of long-term health),
even controlling for life-cycle factors (education and wealth) and the availability of public health centers. We find no long-term effects on
short term health (anemia and Body Mass Index (BMI)) or psychosocial indicators (domestic abuse).
Ó 2013 Elsevier Ltd. All rights reserved.

Key words — health, armed conflict, Peru, fetal origins, life-course models

1. INTRODUCTION The second identification challenge is to find a credible

Recent literature in economics and epidemiology has
documented important and long-lasting effects of shocks
experienced in early life, either in utero or in early infancy.
The usual story is that shocks to nutrition in a critical stage
of human development can have permanent deleterious effects
on an individual’s health over her life-course because they
have permanent effects on her physiology (Barker, 1998). A
number of recent papers have documented long-lasting effects
of such shocks on adult height (Case & Paxson, 2010), on
adult socioeconomic outcomes (Almond, Chay, & Lee, 2005;
Currie, 2009), on adult self-reported health, on child
development (Currie, 2009), and on educational out-
comes (Cutler & Lleras-Muney, 2008). Evidence has been
gathered in both developed countries (Almond, 2006; Currie
& Hyson, 1999; Currie & Moretti, 2007; Oreopoulos, Stabile,
Walld, & Roos, 2008) and developing countries (Maccini &
Yang, 2009), from childhood (Paxson & Schady, 2005) to
old age (Case, Fertig, & Paxson, 2005; Grimard, Laszlo, &
Lim, 2010).
Making the case for this critical period programing hypoth-
esis is complicated by two major problems of identification.
First, the effects of shocks in utero or in early infancy on out-
comes later in life will be confounded by their indirect effects
though the acquisition of human capital and the determina-
tion of adult socioeconomic factors. Put differently, if nutri-
tional shocks in a critical period have effects on cognitive
ability (Walker, Chang, Powell, & Grantham-McGregor,
2005), then educational attainment will also be affected.
Meanwhile, a vast literature in labor economics documents
important returns to education, which influence earnings, both
of which influence adult health through health information or
investment in health inputs. And past investment in education
may affect risk and time preferences in ways that influence
adult behavior and hence outcomes ranging from the
socioeconomic to the psychosocial, all of which affect the indi-
vidual’s quality of life. These life-course or pathways channels
are qualitatively important because they can help reverse some
of the negative effects experienced by shocks in the critical
period. They are also policy relevant because they imply that
pro-human capital investments can counter the long-term
effects of these shocks.
139
source of exogenous variation in outcomes during the critical
period which can affect nutrition. Early generations of studies
of this issue pointed to maternal socioeconomic outcomes
(Barker, 1998; Currie, 2009). The notion here is that more
able, educated, and wealthy mothers will be better fed during
pregnancy and breastfeeding and so infant outcomes will be
better. The problem with this strategy is that such variation
is not exogenous if we consider the likely genetic transmission
of, say, cognitive ability. One influential study linked exposure
to the 1918 influenza pandemic in utero and adult economic
outcomes (Almond, 2006). In this case, the swiftness and vir-
ulence of the flu through the US population generated a cred-
ible source of exogenous variation and the children of
pregnant women exposed during the pandemic fared much
worse than their counterparts with mothers who escaped the
flu in their pregnancy months. Similarly, Maccini and Yang
(2009) exploited regional variation in rainfall at birth to ex-
plain a host of adult socioeconomic outcomes (especially
height), and in doing so finding evidence of the critical period
programing.
In this paper, we consider the effects of the Peruvian civil
conflict of the 1980s and 1990s on women’s health and psycho-
social outcomes. We focus on Peru because it provides a un-
ique opportunity to seek for evidence of the critical period
programing hypothesis due to the extensive geographic and
time variation in the intensity and extent of conflict. It is pos-
sible to do so thanks to the work published by the Truth and
Reconciliation Commission, which investigated and reported
* We thank Kim Lehrer, Louis Hotte, and participants at the 2010 Can-
adian Economic Association meetings, DIAL and McGill’s Department of
Economics Summer Workshop. This paper also greatly benefited from the
comments of two anonymous referees. We are grateful to the Demo-
graphic and Health Survey, as well as Peru’s Truth and Reconciliation
Committee for making their data available. We are especially grateful to
Dr. Loyola Garcia-Frias for providing us with the census of health
infrastructure data and to Catherine Vandermeulen for research assistance
during field research in 2007. This project has benefited from funding by
McGill’s Centre for Developing Areas Studies of the Institute for the
Study of International Development. Final revision accepted: August 10,
2013.
140 WORLD DEVELOPMENT
detailed events data at the district level covering a 20-year per-
iod. The conflict is believed to have caused significant eco-
nomic losses to affected households due to theft and
destruction of homes, agricultural plots, and other farm assets.
In extreme cases, people lost loved ones who were also income
earners. They also abandoned their land or dwelling to seek
shelter to avoid being caught in the fighting, and such displace-
ment can lead to susceptibility to disease. In addition, prenatal
stress induced by the conflict likely affected individuals’ cogni-
tive development (Entringer, Buss, Kumsta, Hellhammer, &
Wadhwa, 2009; King & Laplante, 2005; Laplante, Brunet,
Schmitz, Ciampi, & King, 2008). The literature cited in Entrin-
ger et al. (2009) suggests prenatal stress affects cognitive devel-
opment via neurological factors. There are thus three possible
mechanisms through which exposure to conflict in early life
can lead to lower health status later in life, independently of
their effects on determining socioeconomic status: (i) shocks
to nutrition resulting from the death of income earners and
the loss or theft of assets, (ii) shocks to health because of
unsanitary environments during displacement, and (iii) prena-
tal psychosocial stress shocks. Similarly, prenatal stress can
influence long-term outcomes via its effect on birth outcomes.
Indeed, Mansour and Rees (2012) documented the effect of
conflict-related prenatal stress on birth weight in the context
of the al-Aqsa Intifada. Similarly, Yehuda, Mulherin Engel,
Brand, Seckl, Marcus, and Berkowitz (2005) found a positive
association between uterine exposure to the 2001 World Trade
Center attacks and the probability of low cortisol levels in the
offspring, a risk factor in the ability to cope with stress.
We focus on women because it is believed that women tend
to suffer disproportionately from adverse shocks. The develop-
ment literature has documented a great deal of evidence that in
times of economic hardship, girls bear the brunt of intra-
household re-allocation in the face of negative shocks to dis-
posable income (Escobal, 2007; United Nations, 2008). While
Peru ranks 17th out of 102 non-OECD countries in the
OECD’s Social Institutions and Gender Index, a rating indic-
ative of low gender discrimination, Peruvian women nonethe-
less experience higher poverty and unemployment rates than
men, and domestic violence and psychological and sexual
abuse against Peruvian women is unfortunately a common
reality. 1 These facts and the central stage that women hold
in the Millennium Development Goals justify our focus on
considering the long-term effects of civil conflict on women’s
outcomes. In addition, we are able to answer these questions
thanks to the information contained in the primary data set
we utilize, which focuses on women.
We utilize three datasets to conduct our analysis. Our out-
come variables come from the 5th round of the Peru Demo-
graphic and Health Survey (DHS), which is a continuous
survey from 2004 to 2008 of over 41,000 women aged 15 to
49 years. From this data set, we extract information on wo-
men’s health and psychosocial outcomes: height, Body Mass
Index (BMI), anemia, and domestic abuse. This dataset also
provides us with important socioeconomic controls such as
wealth, education, and ethnicity. The main explanatory vari-
able that allows us to measure stress in early infancy comes
from Peru’s Truth and Reconciliation Commission report
(Comisión de Verdad y Reconciliación (CVR), 2003). The
CVR provides detailed data at the district level of violence
over the 1980–2000 period. We construct from these data
dummy variables on whether a woman’s birth district experi-
enced deaths and/or disappearances in her month and year
of birth and adjacent months/years. The CVR estimates that
over 69,000 Peruvians died or disappeared over the course
of the conflict at the hands of the Shining Path or government
forces. These variables, which exhibit considerable space and
time variation, provide us with the basis for the empirical
strategy we use to determine whether the negative shocks in
early life have long-lived effects. To capture long-term effects,
we consider only women over the age of 18 years, so we re-
strict our sample to include only women born between 1970
and 1990. Finally, we use the 1992 health infrastructure
census, provided by Peru’s Ministry of Health, Censo de
Infraestructura Sanitaria.
Our paper thus contributes to several literatures. First, it
contributes to the literature on critical period programing
(Barker, 1998; Case et al., 2005; Case & Paxson, 2010).
Second, it contributes to an important emerging literature
on the microeconomic effects of armed conflict. Akresh and
colleagues have investigated the effects of armed conflict on
children’s health outcomes in the contexts of the Burundian ci-
vil war and Rwandan genocide (Akresh & DeWalque, 2010;
Akresh, Verwimp, & Bundervoet, 2011; Bundervoet,
Verwimp, & Akresh, 2009). Blattman has investigated the
socioeconomic and psychosocial effects of abduction and child
soldiering in the context of the Lord’s Resistance Army in
Uganda (Annan & Blattman, 2010; Annan, Blattman, Mazur-
ana, & Carlson, 2011). León (2012), Sanchez (2010), Galdo
(2013) and Laszlo and Santor (2009) have utilized data from
the CVR to explain socioeconomic outcomes: the first finds
short- and long-term effects of violence on educational
attainment, the second on children, height for age z-scores,
the third on monthly earnings and the fourth uses CVR data
to instrument for migration patterns in analyzing migrants’
access to credit.
We find that civil conflict events during the year preceding
birth have measurable deleterious effects on women’s height,
even controling for observable adult socioeconomic status
such as education and wealth. Events at older ages have either
no effect, or weaker effects. We also find little to no evidence of
the effects of conflict in early life on short-term health or psy-
chosocial indicators. That we find such little persistence in ef-
fects over the long run suggests a number of encouraging
underlying explanations. First, it could be that Peruvian
women are resilient. Such resilience has been documented in
Annan and Blattman (2010) and Annan et al. (2011) who
show that former child soldiers and abductees are able to rein-
tegrate into society, limiting the long-term effects of abduction
on socioeconomic and psychosocial outcomes. Second, though
related, they suggest that positive events can reverse the
deleterious effects of negative shocks in utero and during early
infancy. Indeed, the 1990s saw a tremendous increase of pub-
licly funded social programs in Peru, and especially in poor
rural regions that would have been particularly hard hit by
the conflict.
2. PERU’S CIVIL CONFLICT AND SHOCKS AT BIRTH
We first discuss the possible mechanisms through which the
Peruvian internal conflict could be associated with worse in-
fant outcomes. The existing economics literature on the micro-
economic effects of conflict have identified two important
channels through which conflict could worsen childhood out-
comes. First, Akresh and DeWalque (2010), Akresh et al.
(2011) and Bundervoet et al. (2009) argue, in the contexts of
the conflicts in Burundi and Rwanda, that conflict created
unanticipated shocks to income (theft or destruction of assets
including livestock and grain stocks, death, or abduction of
income earner). Since these shocks are unanticipated, affected
households are unable to adjust in the short-term, and so
 LONG-TERM EFFECTS OF CIVIL CONFLICT ON WOMEN’S HEALTH OUTCOMES IN PERU
already poor households are likely to face food shortages, and
hence an inability to provide adequate nutrition. 2 There is lit-
tle reason to believe that the Peruvian internal conflict is any
different than the Burundian or Rwandan contexts in this
case. 3 Indeed, Tome VIII of the CVR report (CVR, 2003) esti-
mates that the average peasant household suffered material
losses worth over US$ 5,000 (through dwelling, tools, cook-
ware, plows, harvesters, livestock, and potato seeds). This
amount is considerable for poor peasant households. In addi-
tion, land was often abandoned because of conflict (CVR,
2003). It is estimated that nationally, over 30,000 ha were ren-
dered unusable because of terrorism (CVR, 2003). In the
department of Junı́n alone, one of Peru’s major bread baskets,
over 13,000 ha were rendered unusable due to terrorism
(CVR, 2003). If we include labor shortages, the report identi-
fies over 134,000 ha rendered unusable. There is little question
that the extent of theft and destruction severely impacted peas-
ant households’ ability to feed themselves. Barker’s critical
period programing model builds on nutritional deficiencies
in early life, deficiencies which are likely to have affected indi-
viduals in utero or in early life during the Peruvian conflict.
Second, in the contexts of the Burundian and Rwandan con-
flicts, Akresh and DeWalque (2010), Akresh et al. (2011) and
Bundervoet et al. (2009) also argue that displacement during
conflict can have negative effects on child outcomes. The
mechanism at play here would predominantly affect child out-
comes through unsanitary environments and hence disease. In
Peru, the internal conflict forced between 500,000 to one mil-
lion individuals to become displaced (Internal Displacement
Monitoring Centre (IDMC), 2007, 2009). The long-term ef-
fects of the conflict on the internally displaced peoples (IDPs)
continue to be felt today, as IDPs continue to be socially and
economically excluded (IDMC, 2007, 2009). Nearly 50% of
the IDPs, according to the CVR report (CVR, 2003) and the
IDMC (2007, 2009), eventually returned to their original loca-
tion (the remaining half stayed in their new locations, mostly
slums in urban Lima). Indeed, Laszlo and Santor (2009) doc-
ument the role that the internal conflict played in determining
migration patterns. Price and Bohara (2012) find that armed
conflict in the context of Nepal had a negative effect on
ante-natal health care utilization. This suggests another and
related channel through which child outcomes might have
been affected by the Peruvian conflict.
A third important channel may also operate. Beyond delete-
rious physical effects of conflicts, the internal conflict in Peru is
associated with severe psychological effects. Indeed, the CVR
documents numerous psychosocial effects of the conflict: fear
and distrust, disintegration of kinship, and community ties
(orphanhood, loss of loved ones, breakdown of the family,
insecurity, community stigma, disruption to normal mourning
rituals, and loneliness) and harm to personal identity (sexual
assault, torture, humiliation, desperation, and emotional dis-
tress) (CVR, 2003). Laplante and Holguin (2006) also provide
an analysis of the mental health consequences of the Peruvian
conflict. The medical literature provides evidence of noxious
neurological effects of prenatal psychosocial stress (Entringer
et al., 2009; King & Laplante, 2005; Laplante et al., 2008;
Marmot, 1997). This literature finds that maternal psychoso-
cial stress during pregnancy leads to lower fetal development,
which in turn leads to lower cognitive ability and worse mental
health outcomes. Meanwhile, Camacho (2009) finds evidence
of the effect of prenatal stress on low birth weights in the con-
text of terrorist attacks in Colombia.
In summary, whether through malnutrition, disease expo-
sure during forced displacement, or prenatal psychosocial
stress, the effects of the Peruvian internal conflict has a
141
reasonable chance to inflict negative consequences to early
life development. Through either physiological or neurolog-
ical effects, there is good reason to suspect that exposure to
the conflict in utero or early infancy would have direct ef-
fects on adult health outcomes. In other words, the brutality
of the conflict provides an environment in which the critical
period programing mechanism might take effect. Unlike
Mansour and Rees (2012), our paper does not aim to isolate
the effects of one causal mechanism over the other. In fact,
in our opinion, there is good reason to believe that all
mechanisms are present and, for instance, the psychological
(stress) factor could be persistent months, even years after
the date of the events. 4
That said, there are also important other mechanisms pres-
ent in conflict situations, including in Peru’s internal conflict.
Neurological effects of prenatal exposure to stress in various
settings have been found to include reductions in cognitive
and linguistic functioning and memory (Entringer et al.,
2009; King & Laplante, 2005; Laplante et al., 2008). Sanchez
(2010) finds that exposure to conflict in Peru affects adversely
nutrition outcomes as proxied by height-for-age z-scores of
young children. The consequences of these effects might very
well translate into lower educational attainment. Indeed,
León (2012) looks at the effects of the Peruvian internal con-
flict on educational attainment. His results show that violent
events in the individuals’ birth district have a negative effect
on years of schooling. These effects are largely driven by
events in early childhood (which he defines as 2 years before
birth to age 3) and in pre-school years (ages 4–6). Indeed, the
internal conflict also disrupted the provision of educational
services: schools were often destroyed, teachers killed, and
parents were reluctant to send their children to school for
fears of insecurity (CVR, 2003; León, 2012). Using data from
the ENAHO, Galdo (2013) finds that early-life (the first
36 months) exposure to the Peruvian civil conflict contrib-
uted to a 4% fall in adult monthly earnings. However, con-
trary to León (2012), he suggests that the connection
between early-life exposure to conflict and adult earnings
operates through health, as he finds, like Sanchez (2010) in
a secondary data set from a school census that children’s
height-for-age z-scores appear negatively affected by the civil
war events. Thus, factors at birth and over the lifetime might
both be in operation.
3. CONCEPTUAL FRAMEWORK
(a) Long-term effects of shocks in utero and at birth
The conceptual framework for this analysis follows closely
that in Grimard et al. (2010), which is inspired by the work
by Grossman (1972) and Maccini and Yang (2009). Specifi-
cally, health (h) at any given point in time t is a function of
health in all previous periods, human capital investments (E)
in all previous periods, wealth (Y), and initial health (h0). In
addition, local community programs and local environments
(C) in all periods are also believed to affect health. As outlined
in Grimard et al. (2010), this process is summarized by the fol-
lowing health production function:
hit ¼ H ðhi0 ; hi1 ;. . . hit1 ; Ei1 ; . . . Eit ; Y i0 ; . . . Y it ; C i0 ; .. . C it ; X i Þ ð1Þ
where Xi represents time-invariant individual and regional
characteristics. Since health in any given period is itself a
recursive function of health, education, and wealth in previous
periods, we consider the following reduced-form version of the
health production function:
142
hit ¼ hðh0 ; Ei1 ; . . . Eit ; Y i0 ; . . . Y it ; C i0 ; . . . C it ; X i Þ
Since we are particularly interested in critical period program-
ing effects, a key variable in (2) is initial health h0, which we
assume to be determined by unobserved genetic factors (G)
and community-level characteristics (C0), which is the dis-
trict-level conflict shock:
hi0 ¼ gðC i0 ; Gi ; X i Þ
The system comprised of Eqns. (2) and (3) can be expressed as
a linear approximation in a reduced form relationship between
current health and initial shocks (in utero or in early infancy)
as follows:
hit ¼ b0 þ b1 C i0 þ bE Eit þ bY Y it þ bC C it þ bX Xi þ it
An identification issue arises when critical period programing
mechanisms may be at play alongside life-course mechanisms.
Both mechanisms are models from the epidemiological and so-
cial determinants of health literatures that link conditions in
early life to outcomes in later life. The critical period program-
ing model, based on Barker (1998) and Barker, Eriksson, For-
sén, and Osmond (2002), posits that negative shocks in utero
and early infancy affect fetal and infant development in ways
which have permanent effects on the individual’s health. In
other words, these early shocks can cause a permanent down-
ward shift in an individual’s health profile over their life. In
Eqn. (4), the coefficient b1 would capture this effect. Con-
versely, the life-course models (Ben-Shlomo & Kuh, 2002;
Kuh & Wadsworth, 1993) propose that conditions in early-life
affect the determination of human capital acquisition, which in
turn affects adult socioeconomic conditions and hence health
outcomes in later life. That is, adverse shocks in utero and
early infancy will influence an individual’s ability to invest in
education. Since education is critical in generating income,
low education leads to low income, which leads to poor adult
outcomes. Because both mechanisms are likely to exist, the
effect of in utero or early infancy exposure to adverse shocks
on adult health is confounded by a direct (critical period
programing) effect via physiological sequelae and by an indi-
rect (life-course) effect via their effects on adult socioeconomic
status.
(b) Empirical strategy
As both mechanisms can exist simultaneously and are
not mutually exclusive, our identification strategy consists
of testing for the presence of critical period programing
effects and their persistence controling for adult socioeco-
nomic status. This is the strategy employed in Grimard
et al. (2010) which we follow here. Specifically, we first
consider an empirical specification similar to Eqn. (4) that
omits adult socioeconomic factors (namely education and
wealth), then sequentially add adult socioeconomic
conditions. If the effects of shocks at birth persist when
controling for adult socioeconomic conditions, then we
uncover evidence of critical period programing. If the
effects of these shocks weaken when controling for adult
socioeconomic conditions, then we find evidence of
life-course effects.
We begin by considering a reduced-form version of Eqn. (4):
X
T
hid ¼ b0 þ bt CVRðtÞid þ bX X id þ id
t¼s
where hid is the health status of individual i 2 {1, . . . N} in dis-
trict d 2 {1, . . . D} observed in the 2004–08 DHS, CVR(t) is a
WORLD DEVELOPMENT
ð2Þ dummy indicating whether the CVR recorded events (deaths
and disappearances) experienced by individual i in district d
during the tth month or year before (if t < 0) or after (if
t > 0) birth and X includes the individual’s demographic char-
acteristics (excluding education and wealth). Finally, id is the
stochastic error term.
The main questions we are interested in are whether
ð3Þ conflict experienced in utero or in early life have long-lasting
effects on a woman’s health, and whether higher incomes and
education accumulated over her life course can mitigate or
protect the individual from these shocks. If we find that the
conflict variables are statistically significant when observed
in utero or shortly thereafter, then we have preliminary
ð4Þ evidence of the presence of fetal origins or critical period
programing mechanism. We then add wealth and education
to the equation to see whether these adult socioeconomic
characteristics are significantly predictive, which would point
to life-course mechanisms. If the conflict variables maintain
their predictive power when controling for the life-course
variables, then the evidence in favor of the fetal original
mechanism becomes stronger. 5 We should note, of course,
that these are not two competing hypotheses: both fetal
origins or critical period programing and life-course mecha-
nisms can be in operation at the same time for a given
individual. Thus, from (5), we proceed by sequentially adding
the individual’s education (Eid) and wealth (Yid). We are
interested both in their coefficients and in the coefficients on
the conflict variables. Education and wealth are believed to
be extremely strong determinants of adult socioeconomic
status.
X
T
hid ¼ b0 þ bt CVRðtÞid þ bX X id þ bE Eid þ id ð6Þ
t¼s
X
T
hid ¼ b0 þ bt CVRðtÞid þ bX X id þ bE Eid þ bY Y id þ id ð7Þ
t¼s
It is usual in such analyses to estimate these sorts of models
using district-level fixed effects to account for unobserved
district level variation. We will thus also estimate Eqns.
(5)–(7) with district fixed effects. However, these fixed effects
are almost certainly correlated with the main independent
variables of interest since the conflict variables are also at
the district-level. There is good reason to believe that the
conflict was both affected by and affected unobserved
district-level characteristics which themselves could influence
both the fetal origins or critical period programing and
life-cycle mechanisms. The conflict might be more likely to
occur in a district with high poverty rates or with the
presence of schools (schools were often used to recruit rebel
fighters). Conversely, the conflict might have impacted
the district’s ability to offer health services to pregnant
women and their infants. In other words, including both
the conflict variables and the district fixed effects could be
problematic.
In summary, our empirical strategy is to estimate regres-
sions (5)–(7) by Ordinary Least Squares (OLS) or fixed effects
and test whether the district-level conflict shock in utero or in
early infancy has long lasting effects on women’s health out-
comes. If we find that the effects of the shocks are significant,
and that they persist when controling for adult socioeco-
nomic factors, then our results would suggest the presence
ð5Þ of critical period programing effects. If we find that the effects
weaken when controling for adult socioeconomic status, then
the results would suggest that life-course mechanisms are also
in effect.
 LONG-TERM EFFECTS OF CIVIL CONFLICT ON WOMEN’S HEALTH OUTCOMES IN PERU
4. DATA
(a) DHS
We employ the 5th wave of Peru DHS, which is a continu-
ous survey with five yearly cycles from 2004 to 2008. The total
sample size is 46,073 households, and is nationally representa-
tive. We utilize the individual recode for women between the
ages of 15 and 49 years, covering a sample of 41,648 women.
The individual recode files include detailed information on wo-
men’s health (mostly reproductive), their children’s health, as
well as numerous socioeconomic characteristics. Information
about the DHS is available at: http://www.measuredhs.com/
start.cfm. While the DHS for Peru includes earlier waves
(1986, 1991–92, 1996, 2000), including those used in Paxson
and Schady (2005), we selected the 2004–08 wave because
many of the women born during the conflict would have
reached adulthood by 2004. We are thus only now able to look
at the long-term health effects (if any) of the conflict. In addi-
tion, the sample size is large, which will be important given the
sample restrictions we are forced to make.
Indeed, a major limitation of the DHS is that to match
events in utero or in early infancy, we require location codes.
The DHS only provides current district. This means that we
can only observe birth district for non-migrants. While
53.5% of the individual recode sample reported never moving,
there is a concern that this leads to a selection bias because of
importance of IDPs during the internal conflict (IDMC, 2007,
2009). In fact, Laszlo and Santor (2009) show that the conflict
was an important determinant of migration patterns during
this period. Nevertheless, León (2012) shows, using the Peru-
vian census, that the effect of the conflict did not have impor-
tant differential effects across migrants and non-migrants in
terms of their years of education (while the magnitudes of
his point estimates are different, they lie within one standard
error). We also restrict our sample’s by age range. Since we
are interested in adult outcomes, we restrict our sample to
those born between 1970 and 1990. While the conflict began
in 1980, we include women born in the previous decade to
act as controls (our main results would nevertheless remain
the same if we were to restrict the sample to women born be-
tween 1980 and 1990).
We consider a number of dependent variables to capture
physical and psychosocial health. Specifically, we consider
height, weight, and anemia for physical health. Height is an
important health indicator of long-term health (Almond,
2006; Case et al., 2005; Elo & Preston, 1992; Maccini & Yang,
2009; Strauss & Thomas, 2008), and according to research ci-
ted in Elo and Preston (1992) is pre-determined in early child-
hood. Height is thus an adequate measure of health to capture
long-lasting effects of conditions very early in life. The avail-
ability of weight data allows us to construct a BMI, which is
also indicative of overall health (O’Donnell, van Doorslaer,
Wagstaff, & Lindelow, 2008). The medical literature provides
hormonal explanations linking prenatal stress and long-term
effects on adult BMI (Dahlgren et al., 2001; Mueller & Bale,
2006; Welberg & Seckl, 2001).
We extract from the DHS whether the woman is anemic. An
insufficiency of red blood cells means that the body does not
receive enough oxygen, which in severe cases can seriously
harm human organs (such as the brain and the heart) and
can even cause death (WHO, 2008). Medical literature links
iron deficiency in early life with anemia (Oski, 1993) and
prenatal stress has been shown to lead to offspring iron
deficiencies among non-human primates (Coe, Lubach, &
Shirtcliff, 2007).
143
In addition we consider psychosocial variables. The DHS
reports data on whether individuals were victims of any emo-
tional or physical domestic abuse. It also asks if individuals
experienced any domestic violence. The WHO produced a
report on violence and health in 2002, outlining the relation-
ship between domestic abuse and its long-term consequences:
mental health disorders (including depression), chronic pain,
gastrointestinal disorders, reproductive health consequences
in addition to the physical injuries sustained during the abuse
(WHO, 2002).
Table 1 presents the mean and standard deviations for the
sample of women born between 1980 and 1990 from the
2004–08 continuous DHS data set. Women have an average
height of 152.2 cm and about 26% are reported to be affected
with anemia. Sixty-six percent of women appear to have a
BMI that is outside the normal range. These women have
on average 10 years of education and do not appear to be in
the poorest quintile of the wealth distribution, as measured
by the DHS. Indeed only 9% are reported to be in the poorest
quintile whereas the other women appear about equally likely
to be in the other quintiles. Finally, in terms of domestic
violence, 26–32% of women have reported being victims of
emotional or physical domestic abuse. 6
(b) CVR
In 2003, Peru’s Truth and Reconciliation Commission
(Comisión de Verdad y Reconciliación, CVR) produced a 9
tome final report on their investigations of the Peruvian inter-
nal conflict between 1980 and 2000. This final report (CVR,
2003) collected detailed information about violent events that
took place where the perpetrators where either from the Sen-
dero Luminoso, a Mao-inspired rebel movement whose goal
was to eliminate the drastic economic and social inequalities
experienced in Peru, the Movimiento Revolucionar Tupac
Amarú (MRTA, infamous for its siege of the Japanese em-
bassy in 1997), government forces, and local rondas campesin-
as. 7 The report outlines the time-space progression of the
conflict, which originated in the 1980s mostly with the Sendero
Luminoso’s subversive activities in the rural Sierra (especially
in the Ayacucho department) and by the capture of its leader
Abimael Guzmán in 1992 had reached urban Lima. With
Guzmán’s capture, the Sendero Luminoso activities were
drastically hampered, which translated in a sharp drop in
violent activity. Over the two decades, it is estimated that
almost 70,000 Peruvians died or disappeared.
Tome I of the CVR (2003) provides a thorough description
of country-wide deaths and disappearances between 1980 and
2000. Country-wide there were two severe peaks in 1984 and
another in 1989 with over 4000 and 2500 deaths and disap-
pearances, respectively, and a sharp fall following Guzmán’s
capture in 1992 (see gráfico 23 of Tome I of the (CVR,
2003)). Our identification strategy relies on birth district con-
flict, and the report documents significant heterogeneity in
the location of the conflict, measured here as the number of
districts experiencing deaths and disappearances. Gráfico 24
from Tome 1 of the CVR, reproduced here in Figure 1, shows
the yearly number of deaths and disappearances by year (solid
line) as well as the number of districts (dashed line) experienc-
ing deaths and disappearances. The peak of the conflict took
place from 1987 to 1992, when the conflict was at its most
widespread geographically. To put in perspective, of a total
1893 districts in CVR dataset, 56.06% did not experience
any deaths or disappearances. Of the remaining 832 districts,
the average district experienced at least one death or disap-
pearance per year for 4.1 years. One district experienced at
144 WORLD DEVELOPMENT

Table 1. Sample descriptive statistics

Variable Number of observations Mean Standard deviation
Height (mm) 9592 1521.68 57.08
Anemic (0/1) 8676 0.26 –
Normal BMI 9592 0.58 –
Victim of emotional abuse 4457 0.29 –
Victim of physical abuse 4457 0.37 –
Age 9592 25.38 6.29
Years of education 9592 9.90 3.92
Poorest wealth quintile 9592 0.10 –
Second poorest wealth quintile 9592 0.23 –
Middle wealth quintile 9592 0.23 –
Second richest wealth quintile 9592 0.20 –
Richest wealth quintile 9592 0.24 –
Martital status 9592 0.48 –
Urban 9592 0.64 –
Mother tongue is native language 9592 0.12 –
Mother has no schooling 7087 0.25 –
Mother has primary schooling 7087 0.44 –
Mother has secondary schooling 7087 0.22 –
Mother has post-secondary schooling 7087 0.10 –
Father has no schooling 6850 0.10 –
Father has primary schooling 6850 0.45 –
Father has secondary schooling 6850 0.31 –
Father has post-secondary schooling 6850 0.14 –
District CVR at age 2 9592 0.06 –
District CVR at age 1 9592 0.08 –
District CVR at age 0 9592 0.09 –
District CVR at age 1 9592 0.10 –
District CVR at age 2 9592 0.12 –
District CVR at age 3 9592 0.14 –
District CVR at age 4 9592 0.14 –
District CVR at age 5 9592 0.14 –
Total public health centers in district (1992) 9499 7.79 8.54
Total public health centers in district (1996) 9499 10.92 10.29
Total public health centers in district (1999) 9499 10.43 10.03

Figure 1. Number of deaths and disappearances reported to the CVR and number of districts where the events occurred, by year.
Source: Grafico 24, Tome I, CVR (2003).
 LONG-TERM EFFECTS OF CIVIL CONFLICT ON WOMEN’S HEALTH OUTCOMES IN PERU
least one death or disappearance every year. Our identification
strategy takes this geographic variation into account. How-
ever, we note that it considers location-birth year effects of
the conflict, so that the pattern of exposure to conflict changes
for each woman born in a different district-year.
We use an indicator for whether the district experienced any
conflict-related deaths and disappearances of CVR. Other eco-
nomic studies which we are aware of that also used CVR data
(Laszlo & Santor, 2009; León, 2012; Sanchez, 2010) also use
birth district-level deaths and disappearances. While the
CVR also publishes district-level data on other forms of vio-
lence (assaults, sexual assaults, kidnappings, etc.), we focus
on the most extreme cases of deaths and disappearances. To
construct our main independent variable, we identified
whether a district experienced any deaths and disappearances
for each year. We then match this district-year value to the
birth district-year of the individual in the DHS. We include
in our regressions this value annually for each of the two years
before birth, at birth and for first five years of life. 8
Table 1 reports the summary statistics for these variables.
For instance, 15% of sample women were born in a district
which experienced conflict-related deaths and disappear-
ances in the year of their birth. Similarly, 17% of women
were born in a district which experienced conflict-related
deaths and disappearances a year after their birth. At age 5,
21% of the sample lived in districts which experienced deaths
and disappearances. Using observations on those individuals
who never moved from their birth districts, our identification
strategy then relies on the shocks that these women
faced compared to those women residing in districts that did
not directly experience conflict-related deaths and disappear-
ances.
(c) Public health infrastructure census
We also appeal to census data on health infrastructure, pro-
vided by Peru’s Ministry of Health, Censo de Infraestructura
Sanitaria. 9 The census, intended to capture the universe of
public and private health facilities, was conducted in 1992,
1996, and 1999. While there is a wealth of information avail-
able, we use the district-level public health infrastructure data
to help us assess the role that public health plays in determin-
ing women’s health outcomes.
5. RESULTS
(a) Main results – height
Table 2 shows the coefficients from estimating Eqns. (5)–(7)
on the conflict variables, years of education, and wealth
dummies for women’s height. While we do not report the
coefficients on other controls, all specifications include
controls for marital status, whether mother’s native tongue
is indigenous, current location is urban and birth year
dummies. Columns (1) to (3) report the OLS estimates, while
columns (4) to (6) include district-level fixed effects. In
columns (1) and (4), we provide the results of Eqn. (5). We
add education in columns (2) and (5), corresponding to Eqn.
(6). Finally, in columns (3) and (6) we add wealth quintiles
(the poorest quintile as the reference group), corresponding
to Eqn. (7).
The results on height show a pattern similar to what has
been found in other contexts: shocks at birth are important
determinants of height. The results show that district exposure
to conflict deaths and disappearances in the year prior to birth
145
(corresponding to the individual’s in utero exposure) has a
negative effect on a woman’s stature. Conflict at other ages
is not significantly estimated. The effect is particularly strong
(in magnitude and significance levels) in the OLS specifica-
tions, though the magnitude becomes slightly (but insignifi-
cantly) lower when controling for wealth. The OLS results
thus reveal a pattern which is consistent with the critical
period programing hypothesis: the effect is robust to including
education and wealth. The implication of these results is that
effects of shocks in utero or at birth on height are irreversible,
and is consistent with the literature on the determinants of
height, which suggest that height is predetermined by an early
age (Elo & Preston, 1992) and that the uterine environment is
particularly important. In other words, while education and
wealth are positive predictors of height, they are not enough
to reverse the negative consequences of conflict shocks on
height.
The general pattern of results suggests an additional avenue
to explore: could wealthier households better sustain the
shock than poor households? In other words, could children
born into richer homes be better insulated from the civil
conflict shock than poorer ones, and thus the long-term effect
on their height be lessened? Unfortunately, we do not have
any information on household wealth at birth. The results
in Table 2 already control for maternal tongue and urban
residence. Additionally, we can use parental education
information from the DHS. We generate variables for
maternal and paternal educational attainment (none,
primary, secondary, and post-secondary), and the descriptive
statistics for these are shown in Table 1. The most common
educational attainment level for both mothers and fathers is
primary schooling. We re-run the analysis from Table 2
including these variables (post-secondary schooling is the
omitted category), and the results are shown in Table 3. Since
parental education is only available as of the 2006 round of
the survey, we re-run Eqn. (7) on a reduced sample and
compare the results to those in Table 2. Columns (1) and
(4) in Table 3 repeat the results from Table 2 (columns (3)
and (6)) for the OLS and district fixed-effects, respectively.
Columns (2) and (5) repeat the same exercise on the reduced
sample. Finally, columns (3) and (6) add the parental educa-
tion variables.
Comparing the original results with the results using the re-
stricted sample, we find similar conclusions: controling for
education and current wealth, the conflict in the year preced-
ing birth has a statistically and significantly negative effect
on heights. The result is larger in magnitude, but within one
standard deviation, and so we can not reject that they are
the same. Controling for maternal and paternal education,
we find no measurable change in the results on the conflict
variables: these remain highly stable. Maternal education,
however, is a very strong predictor of a woman’s height: rela-
tive to post-secondary schooling women with less educated
mothers are shorter. The effect is monotonic in maternal edu-
cation. More educated and thus presumably wealthier mothers
will have taller children, but the evidence here is suggestive
that this does not insulate their children from the conflict
shock.
In fact, it seems that none of the variables that capture cur-
rent wealth or wealth at birth diminish the negative effect of
conflict on height. This conclusion suggests that mechanisms
other than income are present in affecting children’s height
when the shock occurs in utero. Stress has been identified as
a potential mechanism in the fetal origins or critical period
programing hypothesis, and we are unable to rule it out here
(see the literature cited in Section 2).
146 WORLD DEVELOPMENT

Table 2. OLS and fixed effects regressions on height for women born between 1970 and 1990
OLS District FE Province FE Department FE
(1) (2) (3) (4) (5) (6) (7) (8)
District CVR at age 2 1.973 1.71 0.902 4.658 4.922 4.55 2.577 2.346
[3.171] [3.105] [2.981] [3.001] [2.953]* [2.956] [3.005] [2.990]
District CVR at age 1 7.082 7.284 6.967 6.107 5.973 5.753 5.237 4.746
[2.831]** [2.888]** [2.984]** [2.717]** [2.799]** [2.881]** [2.895]* [2.914]
District CVR at age 0 1.419 1.927 1.021 1.337 0.954 1.426 0.973 0.247
[2.808] [2.800] [2.769] [2.775] [2.742] [2.767] [2.696] [2.665]
District CVR at age 1 1.236 0.735 0.201 0.301 0.749 1.088 1.293 1.397
[2.871] [2.863] [2.846] [2.908] [2.896] [2.920] [2.882] [2.877]
District CVR at age 2 1.725 1.176 1.33 3.114 2.521 2.45 2.211 2.311
[2.338] [2.262] [2.267] [2.287] [2.221] [2.239] [2.246] [2.272]
District CVR at age 3 4.145 3.892 4.179 2.975 2.895 3.053 3.145 2.903
[2.376]* [2.333]* [2.238]* [2.210] [2.184] [2.159] [2.132] [2.124]
District CVR at age 4 2.389 1.153 1.364 1.835 0.598 0.692 0.795 0.349
[2.384] [2.372] [2.478] [2.449] [2.417] [2.473] [2.493] [2.424]
District CVR at age 5 0.765 0.811 1.515 2.459 2.789 2.947 1.987 2.596
[2.244] [2.200] [2.196] [2.293] [2.248] [2.249] [2.243] [2.232]
Years education 3.751 2.71 3.269 2.69 2.894 2.834
[0.216]*** [0.217]*** [0.233]*** [0.234]*** [0.219]*** [0.222]***
2nd wealth quintile 3.174 1.161 1.249 2.811
[2.193] [2.403] [2.173] [2.208]
3rd wealth quintile 9.016 4.818 7.248 8.441
[3.051]*** [3.166] [2.925]** [2.922]***
4th wealth quintile 17.076 10.441 13.1 14.437
[3.218]*** [3.327]*** [3.128]*** [3.083]***
5th wealth quintile 32.182 21.466 26.068 28.577
[3.514]*** [3.445]*** [3.240]*** [3.261]***
Constant 1,510.28 1,477.45 1,478.96 1,506.87 1,479.82 1,479.76 1,507.15 1,491.54
[3.721]*** [3.942]*** [4.212]*** [4.104]*** [4.377]*** [4.520]*** [10.011]*** [5.475]***
Number of districts 608 608 608
Number of provinces 173
Number of departments 25
R-squared 0.06 0.1 0.12 0.05 0.10 0.11 0.16 0.13
Joint F-test 2.47** 2.33** 2.23** 1.33 1.31 1.32 1.00 0.96
N = 9592. Standard errors in parentheses are clustered at the district-level.All specifications include controls for marital status, mother’s native tongue is
indigenous, current location is urban and birth year dummies.
*
Significant at 10%.
**
Significant at 5%.
***
Significant at 1%.

(b) District-level effects
Using district fixed effects in Tables 2 and 3, the existence of
conflict-related deaths and disappearances still has a signifi-
cantly negative effect on height when experienced in the year
prior to birth, though the magnitudes of the coefficients are
smaller. That said, the conflict variables are jointly statistically
insignificant: the correlation between the district fixed effect
and the conflict variable may bias the estimation of the effect
of the conflict variable. 10 Appendix A explores the relation-
ship between the district-level fixed effects in heights and the
district-level conflict variable. It finds indeed a correlation
and, furthermore, that it is exposure in the year of birth that
most severely impacts the district-level fixed effect estimate.
This would then suggest that including the district-level fixed
effects in Eqns. (5)–(7) is likely to cause problems in estimating
the effect of the conflict variable rather than directly leading to
an invalidation of the critical period programing hypothesis.
As an attempt to circumvent this issue, we re-ran the regres-
sion using province or department fixed effects and we get sim-
ilar estimates (columns (7) and (8) in Table 2): the effect of
exposure to conflict in the year preceding birth is negatively
associated with height, the magnitudes of the effects are within
a standard deviation of the district fixed effects estimate,
though the department fixed effect estimate is insignificant.
Similarly, we fail to reject that the fixed effects estimates of
conflict exposure are jointly different from zero. This is not
surprising since the location effects at the province and depart-
ment levels likely suffer from the same concerns as the district
fixed effects. 11
(c) Conflict variable: timing
The issue of timing is a double edged sword: events may spill
over from one year to the next, which may bias our results in
Table 2. Table 4 presents the results from re-estimating Eqn.
(7), using each of the conflict variables one at a time. While
the results indicate that conflict at different periods are indeed
negatively associated with height, we find that the effect is
strongest in the year preceding birth. This result is consistent
with the fetal origins or critical period programing hypothesis
and suggests that further investigation around the birth year is
 LONG-TERM EFFECTS OF CIVIL CONFLICT ON WOMEN’S HEALTH OUTCOMES IN PERU 147

Table 3. OLS and fixed effects regressions on height for women born between 1970 and 1990, controling for parents’ education
OLS District FE
(1) (2) (3) (4) (5) (6)
District CVR at age 2 0.902 1.378 1.167 4.55 4.69 4.215
[2.981] [3.772] [3.782] [2.956] [3.647] [3.656]
District CVR at age 1 6.967 8.739 8.767 5.753 8.262 8.225
[2.984]** [3.743]** [3.592]** [2.881]** [3.598]** [3.527]**
District CVR at age 0 1.021 1.406 0.867 1.426 1.575 2.039
[2.769] [3.337] [3.296] [2.767] [3.238] [3.223]
District CVR at age 1 0.201 0.198 0.651 1.088 0.16 0.403
[2.846] [2.996] [2.928] [2.920] [3.036] [3.019]
District CVR at age 2 1.33 3.235 3.547 2.45 4.487 4.735
[2.267] [2.754] [2.717] [2.239] [2.786] [2.787]*.
District CVR at age 3 4.179 3.452 3.287 3.053 2.688 2.654
[2.238]* [2.481] [2.465] [2.159] [2.427] [2.425]
District CVR at age 4 1.364 2.892 2.881 0.692 2.182 1.95
[2.478] [2.936] [2.987] [2.473] [2.906] [2.951]
District CVR at age 5 1.515 1.383 1.194 2.947 1.826 1.734
[2.196] [2.664] [2.664] [2.249] [2.717] [2.753]
Years education 2.71 2.57 2.038 2.69 2.438 2.035
[0.217]*** [0.253]*** [0.269]*** [0.234]*** [0.275]*** [0.290]***
2nd wealth quintile 3.174 4.673 5.274 1.161 1.766 2.08
[2.193] [3.024] [3.068]* [2.403] [3.209] [3.233]
3rd wealth quintile 9.016 8.71 8.563 4.818 7.102 6.367
[3.051]*** [3.933]** [3.922]** [3.166] [4.071]* [4.051]
4th wealth quintile 17.076 18.354 16.546 10.441 14.56 12.197
[3.218]*** [4.148]*** [4.190]*** [3.327]*** [4.375]*** [4.402]***
5th wealth quintile 32.182 33.512 27.609 21.466 25.529 20.082
[3.514]*** [4.439]*** [4.387]*** [3.445]*** [4.442]*** [4.496]***
Mom has no education 28.658 24.836
[3.749]*** [3.739]***
Mom has primary education 22.022 19.081
[3.028]*** [3.069]***
Mom has secondary education 11.53 10.259
[2.724]*** [2.894]***
Dad has no education 0.515 0.273
[4.073] [3.991]
Dad has primary education 0.106 0.508
[3.039] [3.022]
Dad has secondary education 1.286 0.113
[2.997] [3.015]
Constant 1,478.96 1,478.16 1,507.44 1,479.76 1,482.17 1,507.71
[4.212]*** [4.980]*** [6.038]*** [4.520]*** [5.938]*** [6.892]***
Observations 9592 6736 6736 9592 6736 6736
F-test (CVR) 2.23** 2.17** 2.22** 1.32 1.26 1.3
F-test (mom’s ed) 22.54*** 16.67***
F-test (dad’s ed) 0.14 0.02
F-test (mom’s and dad’s ed) 14.90*** 11.40***
R-squared 0.12 0.11 0.13 0.11 0.11 0.13
Number of districts 586 490 490
Standard errors in parentheses are clustered at the district-level. Same controls as in Table 2.
*
Significant at 10%.
**
Significant at 5%.
***
Significant at 1%.

warranted. Indeed, considering yearly measures rather than
monthly, we lose some degree of precision in the estimates:
the events occurring in the birth year will take on very different
meanings for an individual born on January 1st compared to
December 31st in the same year in the same district. Because
of this, we now move to estimating Eqns. (5)–(7) in columns
(1) and (3) replacing the yearly conflict measures with monthly
conflict measures. This allows us to be more precise regarding
the timing of the conflict events relative to the timing of the
individual’s birth. Because of the collinearity concern men-
tioned earlier, we re-estimate this relationship separating out
the timing: 24 to 10 months before birth, the 9 months before
birth and the 12 months after birth. We do this for the full
specification (Eqn. (7)) in columns (4), (5), and (6).
The results in the first three columns of Table 5 point statis-
tically insignificant effects of district-level conflict only. How-
ever, the F-test for joint significance of the conflict variables
is strongly significant. Columns (4), (5), and (6) show that
148 WORLD DEVELOPMENT

Table 4. OLS and fixed effects regressions on height for women born between 1970 and 1990, sensitivity analysis
(1) (2) (3) (4) (5) (6) (7) (8)
District CVR at age 2 4.96
[2.772]*
District CVR at age 1 8.284
[2.188]***
District CVR at age 0 5.407
[2.315]**
District CVR at age 1 4.046
[2.398]*
District CVR at age 2 3.711
[2.145]*
District CVR at age 3 5.695
[2.097]***
District CVR at age 4 4.188
[1.975]**
District CVR at age 5 2.497
[1.780]
Years education 2.711 2.718 2.719 2.713 2.714 2.705 2.702 2.711
[0.216]*** [0.216]*** [0.215]*** [0.215]*** [0.216]*** [0.216]*** [0.216]*** [0.216]***
2nd wealth quintile 3.194 3.229 3.204 3.197 3.231 3.181 3.249 3.289
[2.182] [2.188] [2.189] [2.184] [2.191] [2.192] [2.190] [2.189]
3rd wealth quintile 8.991 8.966 8.946 8.943 8.988 9.067 8.987 9.005
[3.049]*** [3.045]*** [3.050]*** [3.055]*** [3.056]*** [3.052]*** [3.054]*** [3.056]***
4th wealth quintile 17.241 17.112 17.111 17.172 17.22 17.187 17.238 17.265
[3.218]*** [3.220]*** [3.225]*** [3.228]*** [3.235]*** [3.229]*** [3.232]*** [3.237]***
5th wealth quintile 32.379 32.191 32.176 32.219 32.301 32.315 32.322 32.351
[3.515]*** [3.508]*** [3.522]*** [3.533]*** [3.531]*** [3.515]*** [3.520]*** [3.538]***
Constant 1,479.03 1,478.91 1,478.94 1,479.03 1,479.00 1,479.08 1,479.22 1,479.14
[4.203]*** [4.212]*** [4.208]*** [4.208]*** [4.209]*** [4.212]*** [4.205]*** [4.218]***
R-squared 0.12 0.12 0.12 0.12 0.12 0.12 0.12 0.12
N = 9592. Standard errors in parentheses are clustered at the district-level. Same controls as in Table 2.
*
Significant at 10%.
**
Significant at 5%.
***
Significant at 1%.

the effects of each conflict measure is statistically and signifi-
cantly negative on adult height, the effects being strongest
for the 9 months preceding birth. This evidence suggests that
the period in utero and the year preceding that have the stron-
gest fetal origins or critical period programing impact on adult
height. Given the discussion above about the possible spill-
overs of the effects from one period to the next, it is not sur-
prising to see this pattern in the data here as well.
(d) The role of public health
Our main results point to some evidence of long-term effects
of the civil conflict in utero and in early childhood, a result
barely weakened by including adult socioeconomic status
(education and wealth). Given the potential of life-course
mechanisms in our sample that might not have been ade-
quately picked up by education and wealth, the next step is
to investigate whether public health interventions have been
able to mitigate some of the negative effects of the conflict
on women’s health. This is a particularly important avenue
of investigation for policy. Indeed, the 1990s saw significant
health reforms, largely beginning in 1994 (Cotlear, 2000;
Rousseau, 2007; Valdivia, 2002). These reforms came on the
heels of a near “complete collapse” of public health services
(Rousseau, 2007) in the 1980s as a result of civil conflict and
the deep economic crisis. In the 1990s, under President Alberto
Fujimori, the Peruvian government invested tremendously in
public health and public health infrastructure (Cotlear, 2000;
Rousseau, 2007; Valdivia, 2002), including through the Peru-
vian Social Fund—FONCODES (Van Der Gaag, 1995).
We thus augment our regression in Eqn. (7) with data from
the Health Infrastructure Census. Specifically, we include the
total number of public health facilities at the district-level in
1992, 1996, and 1999. We have three hypotheses here. First,
the availability of health facilities should be associated with
better health outcomes. Second, the effect of civil conflict
should be smaller in regions where there were public health
facilities. Two identification problems immediately arise in
that the first observation for health infrastructure (1992) oc-
curs (i) after the women in our sample were born and (ii) once
the civil conflict is greatly reduced. Third, since public health
investments increased in the 1990s, we expect the health effects
to be more widespread in 1992 and 1996.
Table 6 presents the results of Eqn. (7) augmented with the
health infrastructure data, using the monthly conflict variables.
We repeat the results from columns (4), (5), and (6) in Table 5,
so that they provide the benchmark for comparison. Including
district-level public health centers does not measurably change
the results. We can test an additional hypotheses with the infra-
structure data. First, the availability of public health facilities
should improve health outcomes. Second, since height is indic-
ative of long-term health and is believed pre-determined very
early in life, we would expect the 1992 census data to be the
most significantly correlated with height.
We observe that our first hypothesis holds. Total public
health centers in 1992 have a positive and significant effect
 LONG-TERM EFFECTS OF CIVIL CONFLICT ON WOMEN’S HEALTH OUTCOMES IN PERU 149

Table 5. OLS regressions on height for women born between 1970 and 1990, using monthly CVR indicators
OLS District FE
(1) (2) (3) (4) (5) (6)
CVR 24 and 10 months before birth 2.487 1.771 1.666 4.912
[2.670] [2.654] [2.601] [2.109]**
CVR 0 to 9 months before birth 3.909 4.623 3.199 5.498
[3.598] [3.594] [3.457] [2.263]**
CVR 0 to 12 months after birth 3.527 2.43 1.547 4.453
[2.686] [2.615] [2.490] [2.039]**
Years education 3.746 2.716 2.712 2.719 2.712
[0.215]*** [0.215]*** [0.216]*** [0.215]*** [0.216]***
2nd wealth quintile 3.08 3.143 3.14 3.085
[2.193] [2.189] [2.191] [2.193]
3rd wealth quintile 8.811 8.877 8.857 8.844
[3.058]*** [3.055]*** [3.053]*** [3.062]***
4th wealth quintile 16.926 17.059 16.999 17.008
[3.222]*** [3.222]*** [3.221]*** [3.229]***
5th wealth quintile 31.878 32.093 31.957 32.006
[3.494]*** [3.502]*** [3.487]*** [3.513]***
Constant 1,510.21 1,477.50 1,479.07 1,479.09 1,479.04 1,479.09
[3.705]*** [3.944]*** [4.219]*** [4.213]*** [4.212]*** [4.214]***
Joint F-test 4.25*** 3.35** 2.21*
R-squared 0.06 0.1 0.12 0.12 0.12 0.12
N = 9592. Standard errors in parentheses are clustered at the district-level. Same controls as in Table 2.
*
Significant at 10%.
**
Significant at 5%.
***
Significant at 1%.

Table 6. OLS regressions on height for women born between 1970 and 1990, controling for the role of public health centers
OLS District FE
(1) (2) (3) (4) (5) (6)
CVR 24 and 10 months before birth 4.912 5.358
[2.109]** [2.080]**
CVR 0 to 9 months before birth 5.498 6.08
[2.263]** [2.265]***
CVR 0 to 12 months after birth 4.453 5.076
[2.039]** [1.977]**
Years education 2.712 2.692 2.719 2.7 2.712 2.692
[0.216]*** [0.217]*** [0.215]*** [0.217]*** [0.216]*** [0.217]***
2nd wealth quintile 3.143 2.866 3.14 2.856 3.085 2.795
[2.189] [2.197] [2.191] [2.200] [2.193] [2.200]
3rd wealth quintile 8.877 8.686 8.857 8.656 8.844 8.64
[3.055]*** [3.068]*** [3.053]*** [3.068]*** [3.062]*** [3.076]***
4th wealth quintile 17.059 16.476 16.999 16.398 17.008 16.4
[3.222]*** [3.222]*** [3.221]*** [3.225]*** [3.229]*** [3.231]***
5th wealth quintile 32.093 30.922 31.957 30.754 32.006 30.79
[3.502]*** [3.447]*** [3.487]*** [3.441]*** [3.513]*** [3.460]***
Total public health centers in district (1992) 0.357 0.359 0.363
[0.187]* [0.185]* [0.185]*
Total public health centers in district (1996) 0.067 0.077 0.075
[0.422] [0.418] [0.418]
Total public health centers in district (1999) 0.277 0.283 0.283
[0.382] [0.380] [0.380]
Constant 1,479.09 1,474.95 1,479.04 1,474.88 1,479.09 1,474.93
[4.213]*** [3.823]*** [4.212]*** [3.822]*** [4.214]*** [3.818]***
Observations 9592 9499 9592 9499 9592 9499
R-squared 0.12 0.12 0.12 0.12 0.12 0.12
Standard errors in parentheses are clustered at the district-level. Same controls as in Table 2.
*
Significant at 10%.
**
Significant at 5%
***
Significant at 1%.
150 WORLD DEVELOPMENT

Table 7. Probit marginal effects estimates of the conflict on other physical health indicators
Anemia Normal BMI
(1) (2) (3) (4) (5) (6)
CVR 24 and 10 months before birth 0.015 0.015 0.015 0.001 0.000 0.001
[0.028] [0.028] [0.028] [0.029] [0.029] [0.029]
CVR 0 to 9 months before birth 0.008 0.008 0.010 0.037 0.036 0.037
[0.026] [0.026] [0.026] [0.033] [0.033] [0.033]
CVR 0 to 12 months after birth 0.020 0.021 0.018 0.007 0.008 0.004
[0.020] [0.020] [0.020] [0.029] [0.029] [0.029]
Years education 0.001 0.003 0.004 0.008
[0.002] [0.002] [0.002]** [0.002]***
2nd wealth quintile 0.039 0.118
[0.019]** [0.026]***
3rd wealth quintile 0.038 0.213
[0.023] [0.028]***
4th wealth quintile 0.050 0.204
[0.024]** [0.032]***
5th wealth quintile 0.065 0.183
[0.026]** [0.032]***
Observations 8702 8702 8702 9592 9592 9592
Pseudo R-squared 0.0045 0.0046 0.0054 0.0645 0.065 0.071
Joint F-test 2.49 2.58 1.95 3.55 3.70 3.23
Standard errors in parentheses are clustered at the district-level. Same controls as in Table 2.
**
Significant at 5%.
***
Significant at 1%.

on height: the availability of public health services improves
health outcomes. The way to see whether this is true here is
to compare the results on birth district-level deaths and disap-
pearances across columns (1) and (2), (3) and (4), and (5) and
(6). The coefficient on the conflict variable remains relatively
unchanged in magnitude when controling for district-level
health facilities. The second hypothesis suggests that health ef-
fects of public health investments should become more wide-
spread over time. We find that only the public health
facilities in 1992 are statistically significant. One possible inter-
pretation is that the closer in time to the conflict, the more
likely health facilities can mitigate the negative effects on
height. Another possible interpretation is that public health
became so widespread by 1996 and 1999 that the variation

Table 8. Probit marginal effects estimates of the conflict on psychosocial indicators

Emotional abuse Psychological abuse
(1) (2) (3) (4) (5) (6)
CVR 24 and 10 months before birth 0.010 0.012 0.018 0.015 0.015 0.007
[0.044] [0.044] [0.044] [0.048] [0.048] [0.048]
CVR 0 to 9 months before birth 0.059 0.057 0.056 0.071 0.070 0.069
[0.048] [0.048] [0.049] [0.046] [0.047] [0.047]
CVR 0 to 12 months after birth 0.058 0.055 0.053 0.059 0.057 0.055
[0.036]* [0.036] [0.036] [0.038] [0.038] [0.039]
Years education 0.009 0.007 0.005 0.002
[0.002]*** [0.002]*** [0.002]*** [0.002]
2nd wealth quintile 0.044 0.051
[0.021]** [0.022]**
3rd wealth quintile 0.056 0.062
[0.028]** [0.029]**
4th wealth quintile 0.002 0.013
[0.009] [0.031]
5th wealth quintile 0.037 0.077
[0.030] [0.033]**
Pseudo R-squared 0.0536 0.0570 0.0606 0.0292 0.0302 0.0362
Joint F-test 3.25 3.07 3.15 3.01 2.85 2.67
Standard errors in parentheses are clustered at the district-level. Same controls as in Table 2.
*
Significant at 10%.
**
Significant at 5%.
***
Significant at 1%.
 LONG-TERM EFFECTS OF CIVIL CONFLICT ON WOMEN’S HEALTH OUTCOMES IN PERU
in these variables falls to the extent that they no longer explain
variation in health status. This interpretation has support gi-
ven that Valdivia (2002) finds that the expansion in facilities
did not lead to an increase in the equitability in health services
utilization.
Grimard and Laszlo (2011) consider the role that the con-
flict played in future health service investment. Using the same
data as used in this paper, they find that the change in district-
level public health centers between 1993 and 1996 was posi-
tively associated with whether the district experienced conflict
between 1980 and 1992, indicating post-conflict targeting of
health services. This does not cause a problem in our analysis.
Indeed, the coefficients on the conflict variables in Table 5 do
not change when adding the levels of public health. Even if
public health service placement and conflict are jointly deter-
mined, if the critical period programing hypothesis is in oper-
ation, then we would expect little effect of public health
interventions unless they specifically deal with the in utero
mechanisms described above (nutrition, prenatal stress, and
displacement). We also rule out the possibility that public
health investments had little impact on overall health since
the linear coefficients on the health centers in Table 5 are sig-
nificantly positively associated with height. Yet, this effect
could be due to some positive self-selection (i.e. health centers
may have been put in place in conflict areas where the health
levels were relatively better) in the placement of government
health services, something that Grimard and Laszlo (2011)
find some mixed evidence. This positive self-selection, if it ex-
ists, might still be consistent with the critical period program-
ing hypothesis concerning the effects of the conflict that
happened in the past.
(e) Other health indicators
Thus far, this paper follows the literature on the long-term
social determinants of health using height as an indicator for
long-term health. While this might be seen as restrictive, we
conducted similar analyses using different measures of health
by utilizing the wealth of health related information in the
DHS. Specifically, we consider anemia, BMI, and self-re-
ported measures of domestic emotional or physical abuse.
These results, reported in Table 7 (anemia and BMI) and in
NOTES
1. Gender Index, OECD. http://genderindex.org/country/peru, last
accessed May 18, 2010.
2. Food shortages during the Peruvian internal conflict were made all the
more worse by disruption of transportation routes. Since these shocks are
likely felt fairly uniformly within districts, we do not consider these here as
our empirical strategy will include district fixed effects.
3. Fürst et al. (2010) find weak effects of conflict in Côte d’Ivoire on
household socioeconomic position.
4. The psychological effects do likely persist beyond pregnancy: a
pregnant woman may be traumatized for years after the conflict, and
this could affect the way her children are raised. For a general discussion
of the effect of armed conflict on women, see Mazurna (2010).
5. Events well before the in utero period should have limited effects on
long-term health (Maccini & Yang, 2009).
151
Table 8 (emotional and physical domestic abuse), show little
to no statistical relation between birth month conflict with
these alternative health measures (we also ran this using birth
year instead of birth month, and the results are similar). The
effects of the socioeconomic condition variables on anemic sta-
tus and the probability of having a normal BMI, however, do
have predictive power in meaningful directions. Relative to
individuals in the poorest wealth quintile, wealthier individu-
als are less likely to be anemic and less likely to have a normal
BMI. Similarly, more educated and wealthier individuals are
more likely to have a normal BMI. Since both anemia and
BMI status have to do with current nutrition, and good nutri-
tion is positively associated with both education and wealth,
these results are to be expected. These socioeconomic variables
are also highly significant in predicting emotional and physical
abuse: more educated women and the richest women are least
likely to be victims of abuse, relative to the poorest wealth
quintile. These results are consistent with intuition. Mean-
while, the 2nd and third wealth quintiles are associated with
more abuse relative to women in the poorest quintile, which
may also reflect the fact that the poorest wealth quintile wo-
men are the least likely to be married and so have less of a
chance of being abused by their spouse (marital status is by
far the most important predictor of abuse in these regressions,
results not shown here).
6. CONCLUSIONS
In this paper, we investigate the long-term health implica-
tions for women of strong negative shocks at birth, namely
the Peruvian internal conflict during the 1980s. Using data
from the Peru DHS on women’s health and psychosocial out-
comes (anthropometrics, anemia, and domestic abuse), and
data from Peru’s Truth and Reconciliation Committee Report
on deaths and disappearances in birth districts, we investigate
the importance of critical period programing and life-course
mechanisms of the socioeconomic determinants of health.
Specifically, we find that the effects of the shock at birth are
long-lived, especially for height, even controling for adult
socioeconomic conditions such as education and wealth.
6. While height is typically considered as a measure of long-term health,
it is worthwhile noting that, by construction, it is also significantly
correlated with normal BMI status (q = 0.0369), and insignificantly
correlated with anemia (q = 0.0031) or being a victim of emotional
(q = 0.0070) or physical abuse (q = 0.0082). Normal BMI status is,
however, positively correlated with anemia (q = 0.0396), and negatively
correlated with bring a victim of emotional (q = 0.0262) and physical
abuse (q = 0.0325). Finally, being a victim of emotional abuse is highly
positively correlated with being a victim of physical abuse (q = 0.5289).
7. The full report and data can be freely accessed on line at: http://
www.cverdad.org.pe/ifinal/index.php.
8. In the empirical analysis, we will also explore the monthly frequency
of the CVR data by considering months before and months after birth.
9. We thank Dr. Loyola Garcia-Frias from Peru’s Ministerio de Salud
(Oficina General de Estadı́stica e Informática) for giving us access to the
Health Infrastructure Census.
152 WORLD DEVELOPMENT
10. Districts exposed to conflict are likely to also be different in other
ways which might be meaningful in determining a woman’s height. For
instance, the conflict might have been more likely in poor districts which
also had worse access to prenatal care. Conversely, one can imagine that a
district which has experienced conflict might have consequently been
unable to provide prenatal care for months (or years) following the event
itself. Indeed, many of the worst years of the internal conflict coincided
with a severe macroeconomic crisis and the Peruvian government was not
likely in a position to rebuild or invest in a given district. Rousseau (2007)
describes the state of public health in the 1980s as being “in a state of
functional collapse.”
11. Using district fixed effects may be demanding on the variance-
covariance matrix since the number of districts is large—over 600. In
unreported regressions, available upon request from the authors, we tried
restricting our attention to the urban sample, where there are much fewer
districts and women’s heights are higher on average and the variance
larger. Our conclusions hold, though the coefficients are larger in
REFERENCES
Akresh R. & DeWalque, D. (2010). Armed conflict and schooling: Evidence
from the 1994 Rwandan genocide. World Bank Policy Research
Working Paper #4606.
Akresh, R., Verwimp, P., & Bundervoet, T. (2011). Civil war, crop failure,
and the health status of young children. Economic Development and
Cultural Change, 59(4), 777–810.
Almond, D. (2006). Is the 1918 influenza pandemic over? Long-term
effects of in utero influenza exposure in the post-1940 US population.
Journal of Political Economy, 114(4), 672–712.
Almond, D., Chay, K., & Lee, D. (2005). The costs of low birth weight.
Quarterly Journal of Economics, 102(3), 1031–1083.
Annan, J., & Blattman, C. (2010). The consequences of child soldiering.
Review of Economics and Statistics, 92(4), 882–898.
Annan, J., Blattman, C., Mazurana, D., & Carlson, K. (2011). Civil war,
gender and reintegration in Northern Uganda. Journal of Conflict
Resolution, 55, 877–908.
Barker, D. J. P. (1998). Mothers, babies and health in later life (2nd ed.).
Edinburgh: Churchill Livingstone.
Barker, D. J. P., Eriksson, J. G., Forsén, T., & Osmond, C. (2002). Fetal
origins of adult disease: Strength of effects and biological basis.
International Journal of Epidemiology, 31(6), 1235–1239.
Ben-Shlomo, Y., & Kuh, D. (2002). A life course approach to chronic
disease epidemiology: Conceptual models, empirical challenges and
interdisciplinary perspectives. International Journal of Epidemiology,
31(2), 285–293.
Bundervoet, T., Verwimp, P., & Akresh, R. (2009). Health and civil war in
rural Burundi. Journal of Human Resources, 44(2), 536–563.
Camacho, A. (2009). Stress and birth weight: Evidence from terrorist attacks.
American Economic Review: Papers & Proceedings, 98(2), 511–515.
Case, A., & Paxson, C. (2010). Causes and consequences of early life
health. Demography, 47(Suppl), S65–S85.
Case, A., Fertig, A., & Paxson, C. (2005). The lasting impact of childhood
health and circumstance. Journal of Health Economics, 24(2), 265–289.
Coe, C., Lubach, G., & Shirtcliff, E. (2007). Maternal stress during
pregnancy predisposes for iron deficiency in infant monkeys impacting
innate immunity. Pediatric Research, 61(5), 520–524.
Comisión de Verdad y Reconciliación (CVR) (2003). Informe Final. Lima:
CVR.
Cotlear, D. (2000). Peru: Reforming health care for the poor. World Bank
Human Development Department LCSHD Paper Series No. 57.
Currie, J. (2009). Healthy, wealthy and wise: Socioeconomic status, poor
health in childhood, and human capital development. Journal of
Economic Literature, 47(1), 87–122.
Currie, J., & Hyson, R. (1999). Is the impact of health shocks cushioned
by socioeconomic Status? The case of low birth weight. American
Economic Review Papers and Proceedings, 89(2), 246–250.
magnitude and the district fixed effects estimates are statistically signif-
icant. Specifically, we found that the effect of conflict in the year before
birth to be the strongest, and this effect is maintained in all fixed effects
specifications. Thus, adding the various location fixed effects contributes
to increasing the variance of the estimation but does not appear to remove
any unobserved location bias that may potentially be there. We suspect
that since the magnitude of the conflict effect is larger here in the base OLS
without location dummies for the urban sector, the increased variance of
the district FE estimates of the conflict is not enough to remove the
significance of the conflict variables.
12. We do not report the results from the first two specifications, as the
analysis has so far shown little difference in the conflict variables when we
include education and wealth. We thus opt for the more complete
specification. We did run this analysis for the other two specifications but
the results are not very different in this case either. Full results are
available from the authors upon request.
Currie, J., & Moretti, E. (2007). Biology as destiny? Short- and long-run
determinants of intergenerational transmission of birth weight. Journal
of Labor Economics, 25(2), 231–263.
Cutler, D., & Lleras-Muney, A. (2008). Education and health: Evaluating
theories and evidence. In R. F. Schoeni, J. House, G. A. Kaplan, & H.
Pollack (Eds.), Making Americans healthier. New York: Russell Sage
Foundation.
Dahlgren, J., Nilsson, C., Jennische, E., Ho, H.-P., Eriksson, E.,
Niklasson, A., et al. (2001). Prenatal cytokine exposure results in
obesity and gender-specific programing. American Journal of Physiol-
ogy-Endocrinology and Metabolism, 281(2), E326–E334.
Elo, I. T., & Preston, S. H. (1992). Effects of early-life conditions on adult
mortality: A review. Population Index, 58(2), 186–212.
Entringer, S., Buss, C., Kumsta, K., Hellhammer, D. H., & Wadhwa, P.
D. (2009). Prenatal psychosocial stress exposure is associated with
subsequent working memory performance in young women. Behavioral
Neuroscience, 123(4), 886–893.
Escobal, J. (2007). Trade liberalization and children: Understanding and
coping with children’s vulnerabilities. 2020 Focus Brief on the World’s
Poor and Hungry People. Washington, DC: IFPRI.
Fürst, T., Tschannen, A. B., Raso, G., Acka, C. A., de Savigny, D., Girardin,
O., et al. (2010). Effect of an armed conflict on relative socioeconomic
position of rural households: Case study from western Côte d’Ivoire.
Emerging Themes in Epidemiology, 7(6), <http://www.ncbi.nlm.nih.gov/
pmc/articles/PMC2945336/pdf/1742-7622-7-6.pdf>.
Galdo, J. (2013). Long-run labor market impacts of early life exposure to
civil war: Evidence from the shining path in Peru. Economic Develop-
ment and Cultural Change, 61(4), 789–823.
Grimard, F., Laszlo, S., & Lim, W. (2010). Health, aging and childhood
socioeconomic conditions in Mexico. Journal of Health Economics,
29(5), 630–640.
Grimard, F., & Laszlo, S. (2011). Civil conflict and the political economy of
delivering health services in Peru. McGill working Paper.
Grossman, M. (1972). On the concept of health capital and the demand
for health. Journal of Political Economy, 80(2), 223–255.
King, S., & Laplante, D. P. (2005). The effects of prenatal maternal stress
on children’s cognitive development: Project Ice Storm. Stress, 8(1),
35–45.
Kuh, D. J., & Wadsworth, M. E. (1993). Physical health status at 36 years
in a British national birth cohort. Social Science and Medicine, 37(7),
905–916.
Internal Displacement Monitoring Centre (IDMC). (2007). Peru: Com-
pensation pending for victims of forced displacement. <http://lib.ohchr.
org/HRBodies/UPR/Documents/Session2/PE/IDMC_PER_UPR_Sn
2_2 008anx_Comp ensationpendingforvictimsofforceddisplacement.
pdf> (Last accessed 19.05.10).
 LONG-TERM EFFECTS OF CIVIL CONFLICT ON WOMEN’S HEALTH OUTCOMES IN PERU
Internal Displacement Monitoring Centre (IDMC). (2009). Peru :
Reparations begin but IDPs excluded. A profile of the internal
displacement situation. <http://www.internal-displacement.org/coun-
tries/peru> (Last accessed 19.05.10).
Laplante, D. P., Brunet, A., Schmitz, N., Ciampi, A., & King, D. (2008).
Project Ice Storm: Prenatal maternal stress affects cognitive and
linguistic functioning in 5 year-old children. Journal of the American
Academy of Child and Adolescent Psychiatry, 47(9), 1063–1072.
Laplante, L. J., & Holguin, M. R. (2006). The Peruvian Truth Commis-
sion’s mental health reparations: Empowering survivors of political
violence to impact public health policy. Health and Human Rights, 9(2),
136–163.
Laszlo, S., & Santor, E. (2009). Migration, social networks and credit:
Empirical evidence from Peru. Developing Economies, 47(4),
383–409.
León, G. (2012). Civil conflict and human capital accumulation: The long-
term effects of political violence in Perú. Journal of Human Resources,
47(4), 991–1022.
Mansour, H., & Rees, D. I. (2012). Armed conflict and birth weight:
Evidence from the al-Aqsa Intifada. Journal of Development Econom-
ics, 99(1), 190–199.
Marmot, M. G. (1997). Early life and adult disorder: Research themes.
British Medical Bulletin, 53(1), 3–9.
Mazurna, D. (2010). Understanding the gendered legacies of armed conflict:
Women’s rights and lives during armed conflict and transition periods and
governance IDRC research report. <http://idl-bnc.idrc.ca/dspace/bit-
stream/10625/43897/1/130394.pdf>.
Maccini, S., & Yang, D. (2009). Under the weather: Health, schooling,
and economic consequences of early-life rainfall. American Economic
Review, 99(3), 1006–1026.
Mueller, B., & Bale, T. (2006). Impact of prenatal stress on long-term
body weight is dependent on timing and maternal sensitivity. Physi-
ology and Behavior, 88(4-5), 605–614.
O’Donnell, O., van Doorslaer, E., Wagstaff, A., & Lindelow, M. (2008).
Analyzing health equity using household survey data. Washington, DC:
The World Bank.
Olson, M., Sarna, N., & Swamy, A. (2000). Governance and growth: A
simple hypothesis explaining cross-country differences in productivity
growth. Public Choice, 102, 341–364.
Oreopoulos, P., Stabile, M., Walld, R., & Roos, L. (2008). Short-,
medium-, and long-term consequences of poor infant health. Journal of
Human Resources, 43(1), 138–188.
Oski, F. (1993). Iron deficiency in infancy and childhood. New England
Journal of Medicine, 329, 190–193.
Paxson, C., & Schady, N. (2005). Child health and economic crisis in Peru.
World Bank Economic Review, 19(2), 203–223.
Price, J. I., & Bohara, A. K. (2012). Maternal health care amid political
unrest: The effect of armed conflict on antenatal care utilization in
Nepal. Health Policy and Planning, 1–11.
Rousseau, S. (2007). The politics of reproductive health in Peru: Gender
and social policy in the Global South. Social Politics, 14(1), 93–125.
Sanchez, A. (2010). Transitory shocks and long-term human capital
accumulation: The impact of conflict on physical health in Peru. Banco
Central de Reserva del Peru Working Paper #2010-020.
Strauss, J., & Thomas, D. (2008). Health over the life-course. In T. P.
Schutlz, & J. Strauss (Eds.). Handbook of development economics (Vol.
4). Amsterdam: North-Holland.
United Nations. (2008). Linking disaster risk reduction and poverty
reduction: Good practices and lessons learned. A publication of the
Global Network of NGOs for Disaster Risk Reduction. <http://
Table 9. Estimated fixed effects
Mean
No controls for Adult SES
Estimated fixed effect 37.89
Constant 1547.48
Controling for education
Estimated fixed effect 32.89
Constant 1513.24
Controling for education and wealth
Estimated fixed effect 34.51
Constant 1516.86
153
www.unisdr.org/eng/about_isdr/isdr-publications/14_Linking_Disaster_
Risk_Reduction_Poverty_Reduction/Linking_Disaster_Risk_Reduc-
tion_Poverty_Reduction.pdf>.
Valdivia, M. (2002). Public health infrastructure and equity in the
utilization of outpatient health care services in Peru. Health and Policy
Planning, 17(Suppl. 1), 12–19.
Van Der Gaag, J. (1995) (Private and public initiatives: Working together
for health and education.). Washington, DC: World Bank and
International Bank for Reconstruction and Development.
Walker, S. P., Chang, S. M., Powell, C. A., & Grantham-McGregor, S. M.
(2005). Effects of early childhood psychosocial stimulation and
nutritional supplementation on cognition and education in growth-
stunted Jamaican children: Prospective cohort study. The Lancet,
366(9499), 1804–1807.
Welberg, L., & Seckl, J. (2001). Prenatal stress, glucocorticoids and the
programing of the brain. Journal of Neuroendocrinology, 13(2),
113–128.
World Health Organization (2002). In E. G. Krug, L. L. Dahlberg, J. A.
Mercy, A. B. Zwi, & R. Lozano (Eds.), World report on violence and
health. Geneva: WHO.
World Health Organization (2008). In B. de Benoist, E. McLean, I. Egli, &
M. Cogswell (Eds.), Worldwide prevalence of anaemia 1993–2005:
WHO Global Database on Anaemia. Geneva: WHO.
Yehuda, R., Mulherin Engel, S., Brand, S. R., Seckl, J., Marcus, S. M., &
Berkowitz, G. S. (2005). Transgenerational effects of posttraumatic
stress disorder in babies of mothers exposed to the World Trade Center
attacks during pregnancy. Journal of Clinical Endocrinology & Metab-
olism, 90(7), 4115–4118.
APPENDIX A
This appendix proposes a two-stage approach to explore
further the issue of correlated fixed effects. It is inspired by a
technique used in Olson, Sarna, and Swamy (2000) who face
a similar problem in estimating the effect of institutions on
growth. The first stage consists in estimating district-fixed ef-
fects variants of Eqns. (5)–(7) where we do not include to con-
flict variables:
hid ¼ a0 þ aX X id þ ðaE Eid þ aY Y id Þ þ fed þ mid ð8Þ
where fed is the district-level fixed effect and mid is a stochastic
error term. From (8), we take the estimated fixed effects f ^ed ,
which we then regress against the district-level conflict vari-
ables in the second stage:
X
T 
f ^ed ¼ ct CVRðtÞd þtd ð9Þ
t¼s
 Pd
where CVRðtÞd ¼ n1d ni¼1 CVRðtÞid is as defined in the main text.
Intuitively, this approach allows us to evaluate whether the
district-level effects on a woman’s health later in life is driven
by the conflict experienced in that district, and more specifi-
Std. Dev. Min Max N
166 41 607
154 61 607
153 57 607
154 WORLD DEVELOPMENT

cally, whether the timing of that effect (whether in utero or in
early infancy) matters in ways consistent with the fetal origins
or critical period programing hypothesis.
We proceed to explore the relationship between district-level
fixed effects and district-level conflict. We do so by first
estimating the fixed effects model in (8), where we do not
include district-level conflict data. From this estimation, we
retrieve the estimated fixed effects. Table 9 presents the
descriptive statistics of these fixed effects for each of the three
specifications: first where we exclude any adult socioeconomic
characteristics (education and wealth), second where we
include education and third where we include both education
and wealth. For reference, we also include the constant term in
these regressions. The constant terms should be viewed a
woman’s estimated height in the event that all independent
variables take on the value zero. Note that the estimated

Table 10. OLS regressions of estimated fixed effects

(1) (2) (3)
District CVR at age 2 15.16 14.411 12.659
[27.791] [24.902] [24.757]
District CVR at age 1 7.185 4.378 4.898
[27.985] [25.743] [25.203]
District CVR at age 0 36.643 32.419 30.169
[28.846] [25.060] [25.222]
District CVR at age 1 24.581 19.581 20.695
[34.421] [33.102] [33.931]
District CVR at age 2 38.36 36.406 40.038
[27.446] [25.286] [25.238]
District CVR at age 3 13.587 13.986 15.05
[25.225] [23.010] [22.830]
District CVR at age 4 18.601 19.71 20.525
[26.599] [24.88] [24.391]
District CVR at age 5 12.033 15.237 15.115
[24.42] [22.955] [22.928]
Observations 607 607 607
Joint F-test 26.97*** 25.41*** 26.87***
FE model includes E N Y Y
FE model includes E and Y N N Y
R-Squared 0.23 0.23 0.23
Standard errors in parentheses are clustered at the district-level. All specifications include controls for marital tatus, mother’s native tongue is indigenous,
current location is urban and birth year dummies.
***
Significant at 1%.

Table 11. OLS regressions of estimated fixed effects on yearly CVR indicators
Dependent variable: estimated district fixed effect
(1) (2) (3) (4) (5) (6) (7) (8)
District CVR at age 2 144.939
(10.936)***
District CVR at age 1 129.625
(9.440)***
District CVR at age 0 125.568
(8.734)***
District CVR at age 1 119.01
(8.220)***
District CVR at age 2 113.93
(7.460)***
District CVR at age 3 108.14
(7.435)***
District CVR at age 4 107.009
(8.282)***
District CVR at age 5 102.546
(7.761)***
Observations 607 607 607 607 607 607 607 607
R-Squared 0.19 0.20 0.22 0.22 0.22 0.21 0.21 0.19
Standard errors in parentheses are clustered at the district-level. All specifications include controls for marital status, mother’s native tongue is indigenous,
current location is urban and birth year dummies.
***
Significant at 1%.
 LONG-TERM EFFECTS OF CIVIL CONFLICT ON WOMEN’S HEALTH OUTCOMES IN PERU
constant terms all lie within 30 mm of the mean height
reported in Table 1. The estimated fixed effects range
from roughly 170 to +66 mm, with an average of just
about 40 to 45 mm. These estimates all lie within a reason-
able range.
The second step in this procedure is to run a district-level
regression of the estimated fixed effects on district-level aver-
age exposure to the conflict, as in Eqn. (9). The results of this
estimation are found in Table 10 and suggest that it is expo-
sure in the year of birth that most severely impact the dis-
trict-level fixed effect estimate. In fact, exposure in years
other than the birth year are statistically insignificant in these
regressions. Controling for education and wealth does not do
much to change the estimated coefficients on conflict. The F-
tests for the joint significance of the exposure variables are
however quite strong, and stronger in significance level than
the only significant coefficient. This suggests that collinearity
in the exposure variables is a possible problem. There are
two reasons why we might expect high correlations between
the effects of conflict on height to spillover from one year to
the next. First, since many acts actually spanned multiple
years, the correlation would be by construction. Second, the
Available online at www.sciencedirect.com
ScienceDirect
155
discussion in Section 2 suggested several mechanisms with
which the prenatal exposure to conflict might impact height:
nutrition shocks, displacement, and stress. It is reasonable to
believe that these mechanisms persist for some time after the
punctual conflict event. That is to say, women who experience
violence in their district may feel the nutritional, displacement,
and stress effects well beyond the date in which the violence
happened in their district. To illustrate further, she may still
feel the consequences of a violent event that occurred months,
perhaps years, after the event.
Thus, we estimate Eqn. (9) where we include the district-le-
vel mean exposure for each year t at a time. These results for
the specification which includes both education and wealth are
provided in Table 11. 12 The estimated coefficients are (almost
perfectly) monotonically decreasing in magnitude as we move
rightward in the table, and they are always significant. The
estimated coefficient in the birth year, perhaps surprisingly,
are not the largest in magnitude. In fact, it is statistically indis-
tinguishable from the effect in the year preceding birth, mea-
surably smaller than two years before birth and measurably
larger than years subsequent of birth.