Civil wars, forced migration and malaria

Jose G. Montalvo

Department of Economics,

Universitat Pompeu Fabra

and IVIE�

Marta Reynal-Queroly

Universitat Pompeu Fabraz

and CESifo

This version: July 2006

Abstract

Our objetive in this paper is to consider the in�uence of the incidence of malaria in the

decision of population to migrate conditional on the intensity of a civil con�ict. It is well known

that violence and civil wars are basic explanations in force migration. However, the intensity

of the con�ict is just one of the determinants of the �ow of migrants. A rational decisor should

also consider the negative e¤ects of the migration decision. We construct a theoretical model to

explain the rational decision of individuals to migrate or not to in an environment where there

is a violent con�ict of certain intensity and some risk of contracting malaria conditional on

migrating. The empirical results con�rm the theoretical predictions of the model. The intensity

of a civil war increases the �ow of refugees, although in a lower proportion the higher is the

standard of living and the incidence of malaria in the country.

JEL classi�cation number: I18, I31, O15.

�Instituto Valenciano de Investigaciones EconomicasyWe thank J. Caballe and M. Reiter and participants in the European Economic Association Meetings for many use-

ful commnets. The authors are greatful for the �nancial support from the IVIE and the Spanish National Science Foun-

dation Grant SEC2004-03619. Corresponding author: Marta Reynal-Querol, Department of Economics, UPF, C/ Ra-

mon Trias Fargas 25-27, Barcelona 08005, phone (34) 93 542 2590; fax (34) 93 542 1746; email: marta.reynal@upf.eduzDepartment of Economics

1 Introduction

This paper investigates the relationship between civil wars and forced migration, and the role of

the incidence of malaria in that relationship. In recent years there has been a renewed interest for

analyzing the determinants of civil wars, either from the political science perspective (Fearon and

Laitin 2003 or Doyle and Sambanis 2000) or from the economic perspective (Collier and Hoe­ er

2004 or Montalvo and Reynal-Querol 2005). The literature has also considered the measurement

of the consequences of violence and civil con�icts. Collier et al. (2003) provide a review of the

literature on the cost of civil war. Some papers have proposed monetary measures of the direct costs

of violence and civil wars. For example Collier (1999) �nds that during civil wars countries tend to

grow around 2.2 percentage points more slowly than during peace periods. Murdoch and Sandler

(2002) show that civil wars reduce also growth across an entire region of neighboring countries.

However, some authors argue that this direct measure grossly underestimate the true cost of wars

and violence. The calculation of non-monetary costs of violence, associated mainly with its e¤ects

on life expectancy, has deserved less attention. Recently Soares (2005) has provided an estimation

of the welfare cost of violence across countries applying the willingness to pay approach in order

to calculate the social cost. Using this contingent valuation approach Soares (2005) estimates that

the civil con�ict in Colombia reduce in 2.2 years the life expectancy at birth, which corresponds

to a 9.7% reduction of the Colombian GDP. In Latin America the estimated social cost of violence

is 57%. Ghobarah, Huth and Russett (2003) �nd important e¤ects of civil war on the subsequent

increase in the incidence of death and disability due to particular infectious diseases and conditions

in the di¤erent population sub-groups. They use the WHO data on DALYs, disability-adjusted

life-years, which contains detailed information on major 23 diseases and conditions on categories of

the population distinguished by gender and 5 di¤erent age groups.

Another important dimension of the indirect consequences of civil wars is forced migration.

Intense civil wars displace large populations. This mass displacement has many e¤ects. For instance,

Salehyan and Skrede (2006) argue that refugee �ows are an important mechanism which spreads

violence across borders. The presence of a large number of refugees from another country, although

many times with ethnic connections with tribes in the country receiving refugees, increases the

risk of a civil con�ict in the receiving country. In fact, Salehyan and Skrede (2006) conclude that

refugees from neighboring countries increase the probability of a con�ict in the receiving country.

Montalvo and Reynal-Querol (2006) consider a di¤erent dimension of the e¤ect of civil wars in

forced migration. They look at the spread of malaria in the country receiving countries and �nd

1

that there is a clear link between refugees and the spread of malaria. Montalvo and Reynal-Querol

(2006) argue that the outbreak of a civil war or an important social con�ict very often generates

the movement of people �eeing from its consequences. If there is risk of malaria transmission in

the country, even if it is small, and the vector is present, then forced migration is a likely cause for

a serious public health concern. There are many reasons for the increase in malaria incidence as

a consequence of forced migration. First of all, most of the population that �ee from urban areas

is generally not immune to malaria. Secondly malaria incidence is high in rural areas where the

vector can live longer in a favorable environment. Also the anarchic situation caused by this social

unrest and the military importance on paved roads, force people to walk through unfamiliar rural

areas, dumps and forests in order to avoid areas of military activity, so actually helping facilitate

its incidence. In fact population movement (due to political con�icts or civil wars) is potentially

the most important factor in the transmission of malaria (conditional on the dynamics between

vector, parasite and environment)1 . The contact of a non-immune individual with an immune rural

population in a high risk area, also increases the risk of transmission. The importance of the contact

with immune individuals is critical because repeated infection amongst individuals of rural endemic

areas generates an immune response in the host, who controls the infection. Finally, migrants do

not carry prophylactic nor protective measures against the bitting of the mosquito (like nets).

In this paper we consider a di¤erent question. In principle the higher is the risk of being infected

when migrating the smaller should be the propensity to migrate given a particular intensity of the

civil con�ict. We construct a theoretical model to explain the relationship between the di¤erent

elements involved in the decision to migrate. Using the results of the model we estimate a speci�ca-

tion to quantify the impact of civil wars on refugees given the incidence of malaria in the country.

The second section contains the theoretical model. The third section discusses data sources and the

estimation results. Finally, the forth section present the conclusions.

2 The incidence of malaria and the propensity to migrate

Epidemiologist has used during many years mathematical models to analyze the spread of infectious

diseases2 . More recently researchers have considered the interconnections between epidemiology and

public policy, inserting some mathematical expression to characterize the spread of infectious diseases

1Cruz Marques (1987).2See for instance Anderson and May (1991).

2

in the context of optimally behaving individual3 . Rational epidemics and their control have been an

important topic of research in recent year for many economist. Geo¤ard and Philipson (1996) study

an economic model were agents can be infected with an infectious diseases and analyze the di¤erence

in terms of public health interventions, of using epidemiological model versus economic models.

Ahituz, Hotz and Philipson (1996) estimate the responsiveness of the demand for condoms to the

incidence of AIDS. Kremer (1996) considers how to include behavioral choices into an epidemiological

models of AIDS.

Our objective in this paper is to consider the in�uence of the incidence of malaria in the decision

of population to migrate conditional on the intensity of the civil war in the country. It is well

known that violence and civil wars are basic explanations in force migration. In this section we

analyze theoretically how the intensity of civil wars, the distribution of income and the incidence

of malaria determine the migration decision. Gersovitz and Hammer (2004) consider a model of

control of infectious diseases4 where a social planner maximizes the present discounted value of

social welfare subject to the epidemiological equations that describe the evolution of the infected

and the susceptible population. Our model does not consider a social planner but the decision of

individuals to migrate or not to in an environment where there is a violent con�ict of certain intensity

and some risk of contracting malaria conditional on migrating.

2.1 Social conditions in the spread of malaria

With the number of clinical cases of malaria on the rise there is an increasing concern on the economic

and public health burden of this disease. There are two basic approaches to the war against malaria.

Some researchers emphasize the need to continue the search for a vaccine in order to reduce the

number of infections. From the economic perspective papers like Kremer (2000a, 2000b) discuss

the institutional design of such an e¤ort in the face of the usual externalities associated with this

kind of problems. There is a di¤erent perspective which relies on the investigation of the causes

of the malaria that are not strictly epidemiological or immunological. In the introduction to the

�rst edition of Bruce-Chwatt�s reference book on malaria the emphasis is placed on the technical

problems5 . He observes that

3Gersovitz and Hammer (2003) present a simple description of epidemiological models and its usefulness in public

policy.4With the appropriate choice of parameters it could be easily applied to malaria.5 Increasing resistance of mosquitoes to known drugs, its changing behavior (i.e. not resting inside houses after

feeding), the DDT prohibition and new vectors resistent to insecticides.

3

�Malaria continues to be a major problem of tropical developing countries and the

recent years have seen the return of it to areas freed from the disease in the 1960�s.

The causes of this setback to a unique international health endeavor have often been

analyzed and commented upon. Certainly technical factors such as the resistance of

insect vectors to insecticides played a major role in these reverses of fortune. Not less

and probably more important were other, often imponderable factors. Inadequacy of

planning, administrative shortcomings, �nancial stringency, shortage of personnel, poor

training were equally responsible for the recent shift in strategy from malaria eradications

to malaria control..� (1978)

It is noticeable the change in the general vision of the problem from the �rst edition to the second

were the focus shifted from technical problems to adverse social and economic conditions.

�During the few years that have elapsed between the �rst and the second edition of

this book the world�s malaria situation has deteriorated. This is particularly evident in

many developing countries situated in the tropics, where the revised strategy of malaria

control (seen as an alternative to malaria eradication) has not been implemented because

of adverse social and economic conditions, due either to general recession or to various

internal di¢ culties�(1985).

Therefore as the e¤orts to eradicate malaria fade by the end of the sixties, as a consequence of

technical problems, the attempts to control the disease encounter di¤erent challenges based on the

internal problems of the countries where the incidence of malaria is the largest. The most relevant

internal problems are civil wars and social con�icts. Therefore the original objective of the war

against malaria, eradication of the diseases, was substitute by control measures to keep the diseases

at a low level. It is important to notice that technical problems were aggravated by the di¢ culties

derive from civil wars and social con�icts. Moreover many wars take place in countries with a high

incidence of malaria. For instance the Afghan civil war takes place in a country where the incidence

of malaria is high, and has generated six millions of refugees that, in principles, are at a very

high risk of being infected. Najera et al. (1992) distinguish di¤erent patterns of reported malaria

cases. The so called �group B�is the one that produces most of the cases of malaria. Najera et al.

(1992) argue that �these countries are characterized by recent e¤orts to increase the exploitation

of natural resources (through agricultural colonization of forest or jungle areas) or by civil war and

sociopolitical con�ict (including illegal drug trade) and large movements of refugees or other mass

4

migrations�. Therefore, the agricultural colonization of forest, usually based on arti�cial irrigation,

and sociopolitical con�icts are the basic social determinants of malaria. Mass migration of urban

population and the contact with immune, but infected, rural population in unpaved areas where the

vector lives favors also the spread of malaria.

In this paper we analyze the relationship between malaria and civil wars through the displacement

of people. We argue that while urban areas are mostly free of non-imported malaria, rural areas have

some characteristics that make them a very contagious place for non-immune civil population �eeing

from civil war or social unrest. We consider civil wars and social con�ict as one of the basic reasons

behind the observed increase in the incidence of malaria either directly (non-immune refugees get

in contact with infected individuals when they �ee through rural and rainforest areas to reach a

foreign country) or indirectly (civil wars make very di¢ cult or even impossible to keep active control

measures against malaria).

2.2 A theoretical model.

In this section we present a simple model to characterize the basic relationships among malaria,

migration and civil wars. The model explains the migration decision of individuals using a mechanism

di¤erent from the usual one, based on wage di¤erences. In developing and underdeveloped countries

many mass migrations take place as a consequence of a civil wars, armed con�icts or ethnic tensions.

However in many of these countries migration caused by a civil war involves to walk through damps,

forest and rural areas where malaria is endemic and, therefore, there is a certain probability of

getting infected by the disease. Therefore the individual has to decide between staying, and maybe

die as a consequence of the civil war, or leaving, and maybe get infected with the malaria parasite6 .

Initially we assume that a country can be divided into two basic parts: cities and rural areas.

In an endemic country, even though urban areas are not malaria endemic, rural areas are not a safe

place for non-immune individuals. People living in rural areas are usually immune to malaria while

people leaving in the city are not. The basic model considers that whenever there is a probability

larger than 0 of dying from a civil war or ethnic unrest (PCW>0) the individual can chose between

staying or leaving the city. We denote the decision of staying by s, and the decision to �ee by f .

Given that most civil wars have a political or ethnic component it seems reasonable to assume that

people �eeing from the cities will not stay inside the country but go to a di¤erent country. Individuals

6 In this sense the model is related with the inclusion of behavioral choices in epidemiological models as in Kremer

1996 or Phillipson 1996.

5

are indexed by the subindex j. The consumption of individuals in the city is heterogeneous and is

represented by c0j while the consumption when moving to the foreign country and in the refugees

camps is simply the survival or minimum consumption c1; which is equal for all the individuals. We

assume that c0j > c1 for all i. There is no production. If an individual stays in the city she dies

with probability PCW . If she stays in the city, and she is alive her level of health will be h0. If

she dies obviously consumption and health will be zero. If she decides to leave the city she will not

be infected in the �rst period but, afterwards, if she gets infected her health will drop to h1where

h0 > h1. The probability of being infected (state i) is characterized as follows:

Pr(ijni) = PM

Pr(iji) = 1

Pr(niji) = 0

where ni means not infected. The second and the third equations are convenient simpli�cations

that do not alter the qualitative results of the model. In the case of malaria the probability of

changing from the infected state to the non infected situation is not 0 but depends on the parasite

family that causes the infection7 . In particular while the P. vivax can live in the human liver for

close to four years, the P. malariae, the most extended variant, could persist for as many as 50

years which is basically consistent with our assumption. However, in general it is also possible that

after being infected, and without re-infection, an individual will change her status to not infected8 .

Notice also that PM could change over time and depend on the number of people that has abandon

the city up to time t. From the basic epidemiological model (see Anderson and May 1992) we can

represent the number of new cases of malaria infection as

di

dt= �1�2�1�2ni� i

where �1 is the number of bites per unit of time of the average mosquito, �2 is the proportion of

bites by infected mosquitoes that lead to human infection, �1 is the proportion of infected mosquitoes,

�2 is the total number of mosquitoes over total population, is the proportion of infected people

7Malaria is not a mortal illness if treated. For this reason we assume that once infected there is no death from

malaria. The addition of this probability of dying from malaria will complicate the model marginally without changing

the basic results.8We also assume that the individuals in the city are not inmune to the malaria parasite and that malaria is not a

problem in highly urbanized areas. Both are realistic assumption (see Najera et al. 1992).

6

that die and ni is the number of susceptible individuals9 . If we would consider that Pr(niji) is not

equal to 0 then we should subtract from the above expression the rate of recovery times the number

of infected10 . In order to keep the model as simple as possible we abstract from these conditions

although its introduction would not change the basic qualitative results11 .

Under this assumptions there are four possible states in this economy: an individual with con-

sumption c0j could stay in the city and survive (s; nd; c0j), stay in the city and die (s; d), �ee from

the city and is not infected (f; ni), and �ee but gets infected (f; i). Therefore the individual takes the

decision of migrating or staying by solving the following stochastic dynamic programming problem:

V (s; nd; c0j) = maxfu(c0j ; h0) + �[(1� PCW )V (s; nd; c0j) + PCW � 0]; (1)

u(c1; h0) + �[PMV (f; i) + (1� PM )V (f; ni)]g

V (s; d) = 0 (2)

V (f; ni) = u(c1; h0) + �[PMV (f; i) + (1� PM )V (f; ni)g (3)

V (f; i) =u(c1; h1)

1� � (4)

where V (s; nd; c0j), is the value function of an individual with consumption c0j that decided to

stay and did not die; V (s; d) is the value function of any individual that decided to stay and died;

V (f; ni), is the value function of an individual that decided to �ee and has not been infected; and

�nally V (f; i) is the value function of an individual that decided to �ee and has been infected. The

coe¢ cient � 2 (0; 1) is the discount factor.

Plugging program (4) into program (3), we obtain and expression for V (f; ni). Plugging this

resulting expression for V (f; ni) and (4), into program (1) we obtain the following stochastic dynamic

programming problem:

9Susceptible individuals are equal to non infected individuals if inmunity is not a possible state.10Gersovitz and Hammer (2004).11Philipson (1996) and Geo¤ard and Philipson (1996) consider the dynamics of AIDS infected populations where

the conditional probability of infection depends on the number of infected people. However notice that we assumed

before that individuals will not stay at the rural areas in the same country but will cross the border to a di¤erent

country. This implies that the number of infected individual will not increase in the origin country.

7

V (s; nd; c0j) = maxfu(c0j ; h0) + �(1� PCW )V (s; nd; c0i);1

1� �(1� PM )u(c1; h0) +

1

1� �(1� PM )�PM1� � u(c1;h1)]g (5)

We can now solve this problem and �nd the value function of the marginal consumer, the in-

dividual that is indi¤erent between staying or �eeing from the city. The value function of the

marginal consumer is V (s; nd; c0), where we denote by c0 the consumption of the marginal consumer

or reservation consumption. This value function has the following expression,

V (s; nd; c0) =u(c1; h0) +

�1��PMu(c1; h1)

[1� �(1� PM )](�(1� PCW )� u(c0; h0)

�(1� PCW )(6)

Therefore, the marginal consumer solves the following problem:

u(c1; h0) +�1��PMu(c1; h1)

[1� �(1� PM )](�(1� PCW )� u(c0; h0)

�(1� PCW )= maxfu(c0; h0) + �(1� PCW )V (s; nd; c0); (7)

u(c1; h0)(1

1� �(1� PM) + u(c1;h1)(

1

1� �(1� PM ))(�PM1� � )] (8)

From here we �nd an implicit function for the reservation consumption

u(c0; h0) =[1� �(1� PCW )][1� �(1� PM )]

[u(c1; h0) +�

1� �PMu(c1; h1)] (9)

Now we can de�ne the policy function given PCW ; PM ; c1; h0; h1;and c0. If c0i < c0, then the

optimal would be to �ee from the city. If c0i > c0 the optimal decision is to stay in the country.

That is the j individual with initial consumption in the city c0j will abandon the city if her

consumption is below the reservation consumption, c0 ,given the probability of dying in the civil

war and the probability of being infected with malaria if she leaves. The initial consumption only

increase the utility of staying, therefore if an individual has a consumption lower(higher) than the

marginal consumption, then the expected utility of staying will be lower(higher) than the expected

utility of �eeing. Therefore from the distribution of total consumption in the city we know that the

individual that will �ee are the ones whose consumption is below the consumption of the marginal

consumer

8

Z c0

c1

N(c)dF (c)

where N(c) is the proportion of individuals of type c.

We adopt the convention that indi¤erent individuals migrate, and then, an individual migrate if

and only if

PMV (f; i) + (1� PM )V (f; ni) � (1� PCW )V (s; nd) (10)

Comparative statics.

The comparative statics allows us to characterize the size of the �ow of migrants as a function

of the probability of dying from a civil war or being infected with malaria given a particular utility

function.

Proposition 1: The consumption of the marginal consumer, c0 increases with PCW ; the prob-

ability of dying if there is a civil war (intensity of the civil war).

Proof : The derivative of the RHS of the expression (9) with respect to PCW is positive. That is,

@[u(c1;h0)+

�1��PMu(c1;h1)

[1��(1�PM )] [1� �(1� PCW )]]@PCW

=u(c1; h0) +

�1��PMu(c1; h1)

[1� �(1� PM )]� > 0 (11)

This means that if PCW increases, the RHS increases. In order to maintain the equality, then

the LHS, u(c0; h0), also must increase. We know that u(c; h) is an increasing function of c. Given

that all individuals have h0 at the initial period, therefore it should be that c0 increases. Therefore

if the probability of dying because of civil war increases, so it does the consumption of the marginal

consumer:�

Corollary 1: The number of individuals that migrate as a consequence of the civil war increases

with PCW ;

Proof: We know that the individual that �ee are the ones whose consumption, c0j ; is below the

consumption of the marginal consumer, c0,

9

Z c0

c1

N(c)dF (c)

From Proposition 1, we know that c increases with PCW and, therefore, if c increases with

PCW , the number of individuals with a consumption below c increases, and therefore the number of

migrants increases.

Proposition 2: The consumption of the marginal consumer, c0, decreases with PM ; the proba-

bility of getting infected by the malaria parasite.

Proof: the derivative of the RHS of the expression (9) above with respect to PM is negative.

That is,

@[u(c1;h0)+

�1��PMu(c1;h1)

[1��(1�PM )] [1� �(1� PCW )]]@PM

=1� �(1� PCW )[1� �(1� PM )]2

�[u(c1; h1)� u(c1; h0)] < 0 (12)

This means that if PM increases, the RHS of (12) decreases. In order to maintain the equality,

the LHS, u(c0; h0), also must decrease. We also know that u(c; h) is an increasing function of c.

Given that all individuals have h0 at the initial period, therefore it should be that c0 decrease.

Therefore if the probability of dying because of civil war increases, consumption of the marginal

consumer decreases.�

Corollary 2: The number of refugees that migrates from civil war decreases with PM ; the

probability of being infected with the malaria parasite.

Proof: We know that the individual will �ee if her consumption, c0j ; is below the consumption

of the marginal consumer, c0,

Z c0

c1

N(c)dF (c)

From Proposition 2, we know that c decreases with PM , therefore, if c, decreases with PM , the

amount of people that has a consumption below c decreases, and therefore the number of refugees

decreases. Therefore we expect to see that migration will increase as a consequence of an increase

in the probability of dying from the civil war (more intense war) and decrease with the probability

of being infected.

10

3 Malaria and migration caused by civil wars.

In this section we analyze the empirical relationship between the intensity of civil wars and forced

migration conditional on the incidence of malaria. Therefore, following the results of section 2, we

estimate the e¤ect of the di¤erent elements considered in the theoretical discussion on the �ow of

migrants. Based in our model the intensity of the �ows of migrants should be positively related

with the existence/intensity of the civil war and negatively related with the probability of being

infected with the malaria parasite (PM ). The estimation of the e¤ect of malaria on migration is

complicated since, as we argued before, the incidence of malaria is endogenous to the size of the

migrating population. However, as argued in the section on social determinants of the incidence of

malaria, this probability can be speci�ed as function of the proportion of non-urban population and

the size of the irrigated land. In addition the distribution of consumption/income is also important

since the �ow of migrants depend on the accumulated density around the trigger point c. The rest

of the section discusses the sources of data and the results of the estimation.

3.1 Data sources

The endogenous variable of the regressions is the �ow of refugees per capita, REFPC (refugees

divided by population). There are two basic sources of information for the data on refugees: the

United Nations High Commission for the Refugees (UNHCR), and the US Committee For Refugees

(USCR). The data on refugees that we use comes from the United Nation High Commission for

the Refugees. Thanks to Susanne Schmeidl we have had access to the internal data of the UNHCR

from 1951 until 199912 . Following the UNHCR de�nition, refugees are persons recognized as refugees

under the 1951 United Nations Convention relating to the Status of Refugees or its 1967 Protocol, the

1969 Organization of African Unity (OAU) Convention Governing the Speci�c Aspects of Refugee

Problems in Africa, persons recognized as refugees in accordance with the UNHCR Statute, persons

granted humanitarian or comparable status and those granted temporary protection. This dataset

is organized by country of origin and country of asylum and provides information on the number of

refugees that arrive to the asylum country at time t coming from di¤erent origin countries. Notice

that, in principle, a high incidence of malaria should also deter internal migration. Unfortunately,

the data on internally displaced persons within countries in civil wars are very scarce for the obvious

12The data from 1951 to 1992 is not public and come from the work of Schemidl and Jenkins (2001). Schemidl and

Jenkins (2001) also describe the di¤erence between the data compiled by the UNHCR and the USCR. They argue

that the data from the UNHCR have higher quality than the ones coming from the USCR.

11

problems of counting displaced people in the middle of a violent con�ict.

We use two alternative datasets on civil wars. The �rst one comes from Doyle and Sambanis

(2000). This de�nition is nearly identical to the de�nition of Singer and Small (1982,1994). Doyle and

Sambanis (2000) (DSCW) de�ne civil war as an armed con�ict with the following characteristics: �(a)

it caused more than 1,000 deaths; (b) it challenged the sovereignty of an internationally recognized

state; (c) it occurred within the recognized boundary of that state; (d) is involves the state as a

principal combatant; (e) it included rebels with the ability to mount organized armed opposition

to the state; and (f) the parties were concerned with the prospects of living together in the same

political unit after the end of the war.�13 . For the sake of robustness we also use the dataset

of Uppsala/PRIO, which we name PRIOCW. This is a contested incompatibility that concerns

government and/or territory where the use of armed force between two parties, of which at least one

is the government of a state, results in at least 25 battle-related deaths yearly14 . We only consider

types 3 and 4 (internal armed con�icts).

We also consider an indicator of the intensity of the civil con�ict. The previous de�nitions of

civil war usually consider a threshold number of battle deaths to determine the civil war status of

a con�ict. They transform battle deaths into a dummy variable. However, the recent availability

of carefully constructed datasets on the number of battle deaths by con�ict allows us to specify the

intensity of a civil war with more precision. In particular, we use the data on annual battle deaths

in con�icts recorded in the Uppsala/PRIO Armed Con�ict Dataset, that have been presented in

Lacina and Gleditsch (2005).

We cannot include directly the contemporaneous incidence of malaria since it is clearly endoge-

nous. However, the probability of being infected depends on the percentage of not immune popula-

tion and also on the extension of country land where the vector can survive. As we argued before,

the infection is mostly associated with non-immune people �eeing from cities. For this reason we

include the proportion of population living in cities (URBPOP), which is obtained from the World

Development Indicators. We should also include proportion of land irrigated per capita (IRRIG)

since large extensions of water are basic for the reproduction of the vector. This last variable come

also from the World Development Indicators. Income distribution and the level of consumption are

two of the factors that explain the intensity of the �ow of migrants when there is a civil war. In

principle in the steady state c is function of the level of consumption of the country, that is, on the

13This de�nition is practically identical to Singer and Small (1994) in their Correlates of Wars project (COW).14The results are not a¤ected by the use of the de�nition of con�ict which includes only intermediate intensity

con�icts and civil wars.

12

standard of living measured in consumption terms. As a proxy we use the so called standard of

living (STLIV) from the World Development Indicators. We also include the level of development

of the country (LNGDP) per capita.

Civil wars are not the only possible reason for forced migration. Natural disasters could also

generate a large �ow of migrants. In order to avoid the omission of some important explanatory

variables we also include in the speci�cation a series of natural disasters (NATDIS). The data come

from the EM-DAT: the OFDA/CRED International Disaster Database15 . Since 1988 the WHO

collaborating Centre for Research on the Epidemiology of Disasters (CRED), has been maintaining

an Emergency Events database EM-DAT. EM-DAT was created with the initial support of the WHO

and the Belgian Government. The disasters database contains essential data on the occurrence

and e¤ects of mass disasters in the world from 1900 to the present day. The disaster data are

sub-divided into three types: natural, technological and con�icts. We only use natural disasters.

EMDAT contains essential core data on the occurrence and e¤ects of over 12.500 mass disasters in the

world from 1900 to present. The database is compiled from various sources, including UN agencies,

non-governmental organizations, insurance companies, research institutes and press agencies. The

OFDA/CRED o¤ers information on the occurrence, the number of people injured, killed, made

homeless and the total number a¤ected. There are many di¤erent types of natural disasters included

in the data base: drought, earthquake, extreme temperature, �ood, landslide, volcano, tidal wave,

wild�re and windstorm. In some speci�cations we show the separate e¤ect of drought since this is

the most likely source of long term e¤ect on the economic conditions of population and, therefore,

the natural disaster that can most likely generate migration.

Therefore the basic speci�cation can be written as

REFPCit = �0LNGDPit + �1CWit

+�2STLIVit + �3STLIVit � CWit

+�4IRRIGit + �5IRRIGit � CWit

+�6URBPOPit + �7URBPOPit � CWit

+�8OPENit +X

�kNATDISkit + ui + vit

where PREFPC is the proportion of the �ow of refugees from the origin country to any other

15EM-DAT: The OFDA/CRED International Disasater Database- www.cred.be/emdat-Universite Catholique de

Louvain-Brussels-Belgium.

13

country in a particular year over the population of the origin country, CW is a dummy variable

that takes value 1 if there is a civil war, STLIV is the standard of living index, URBPOP is the

proportion of urban population, IRRIG is the extension of land irrigated over the size of the country,

OPEN is the degree of openness measure as the proportion of import plus exports over GDP. The

openness variable is expected to capture the degree of porosity of boundaries. We also include in

some speci�cations droughts and other natural disasters that may generate also mass migration.

As we argued before, there are basically two groups of factors that a¤ect the incidence of malaria:

ecological conditions and social conditions. The ecological conditions include the African savannah,

the plains and valleys outside of Africa, the highlands, seashore and coastal areas. All these ge-

ographical conditions are country speci�c but time invariant and, therefore, are included in the

�country speci�c e¤ect�of our regression. Therefore, the individual speci�c e¤ect, ui; represents the

country speci�c likelihood of malaria infection given its ecological conditions.

3.2 Empirical results

Table 1 shows the results of the regression using the �xed e¤ect panel data speci�cation and the

de�nition of civil war in Doyle and Sambanis (2000). Column 1 shows that the occurrence of a civil

war increases the �ows of migrants while, given this, income per capita has no explanatory power

on the proportion of migrants over total population. Column 2 include the standard of living. It

shows that this variable is not statistically signi�cant by itself but it is very signi�cant interacted

with civil wars. The estimation shows that, conditional on having a civil war, countries with a

high standard of living generate less migrants. The regression in column 3 includes the proportion

of urban population and the interaction of urban population and civil war. As predicted by the

theoretical model the proportion of urban population, conditional on the country su¤ering a civil

war, has a negative e¤ect on migration. Since urban population is not immune to the malaria

parasite, it is more dangerous for these people to walk around infected areas and, therefore, their

propensity to migrate is lower than the propensity among rural population. The e¤ect of adding

irrigated land to the variables of the model is also the one predicted by the theoretical model. People

living in countries with a large proportion of irrigated land have less propensity to migrate than the

rest, conditional on su¤ering a civil war. When there is no civil war the proportion of land irrigated

has no e¤ect on migration since people do not have to walk through unpaved areas or close to water

dumps. Column 5 shows that the results are not a¤ected by the inclusion of the openness variables,

that measures the porosity of the national boundaries, nor by the inclusion of the e¤ect of droughts.

14

The basic results are maintained when we include dummy variables for the di¤erent natural disasters

consider in the previous subsection. The results are basically the same if we use a direct proxy of

the distribution of income/consumption instead of the standard of living. One such a proxy could

be the proportion of income accruing to the richest quintile, Q5. The data come from Deninger and

Squire (1996). The correlation of the Gini coe¢ cient with Q5 is very high: 0.89 in 1960, 0.92 in

1970, 0.95 in 1980 and 0.98 in 1990. One explanation for such a high correlation is the fact that Q5

has a larger standard deviation than other quintile shares. The estimation shows that the higher is

the proportion of income accruing to the richest quintile the smaller is the �ow of refugees.

Table 2 analyzes the same speci�cations but using a somehow di¤erent de�nition of civil war.

As previously discussed, we include the data on civil war from PRIO to test the consistency of our

results under alternative de�nitions of con�ict. The sign of the coe¢ cients are consistent with the

theoretical model of the previous section. A civil war generates an important proportion of refugees

from the country that su¤ers it, as in table 1. The coe¢ cient is statistically signi�cant although the

impact of civil wars on the proportion of refugees over total population is a little bit smaller than

the e¤ect in the case of the dataset of Doyle and Sambanis (2000). The proportion of refugees is

negatively related with the standard of living, the proportion of urban population, and the extension

of land irrigation, conditional on the country su¤ering a civil war. However, once the civil war is

taken into account, natural disasters have no e¤ect on the intensity of refugees��ows. As expected

the degree of openness has a positive sign but it is not statistically signi�cant16 . Therefore the

empirical results seem to agree with the theoretical model even when we use the dataset on civil

wars of PRIO.

Table 1 and 2 use the event of a civil war to capture the impact of it on refugees and the

conditioning e¤ect of malaria. However, we also want to analyze the e¤ect of the intensity of civil

wars on the �ow of refugees conditional on the likelihood of being infected with malaria if migration

occurs. It is true that the de�nitions of civil war used previously have some components of intensity:

in fact a civil war is consider as such depending on the number of casualties produced every year, the

accumulated number of casualties over the course of the con�ict or both conditions together. In any

case the civil war dummy transforms a continuous variables (number of casualties) in a 0-1 variable.

We can obtain a more precise description of the intensity of the con�ict if we use the number of

casualties of the civil war. We use the Uppsala/PRIO Con�ict Year Data File, which provides

16Since the endogenous variable is a proportion we run also several regressions using a grouped logit speci�cation.

The qualitative results are basically the same as the ones in table 1.

15

information on annual battle deaths in con�icts (Lacina and Gleditsch 2005) to obtain the number

of casualties. The variable DEATHS is constructed as the ratio of casualties over urban population.

The Lacina and Gleditsch dataset de�nes battle deaths as "deaths resulting directly from violence

in�icted through the use of armed force by a party to an armed con�ict during contested combat".

We use speci�cally what the authors called "best estimate of annual battle fatalities". In the basic

regression we interact this new variable with the variables related with the social conditions for

malaria transmission as we did before with the dummy of civil war.

REFPCit = �0LNGDPit + �1DEATHSit

+�2STLIVit + �3STLIVit �DEATHSit

+�4IRRIGit + �5IRRIGit �DEATHSit

+�6URBPOPit + �7URBPOPit �DEATHSit

+�8OPENit +X

�kNATDISkit + ui + vit

Table 3 present the results of the estimation. The intensity of the civil war has a positive and

signi�cant impact on the number of refugees. This is analogous to the impact of civil wars in the

previous tables. The product of the number of casualties by the standard of living is negative and

signi�cant as in tables 1 and 2. Therefore, given a level of casualties the higher is the standard

of living the lower is the number of refugees generated by a civil war. The same is true for the

interaction between the proportion of irrigated land over the total area of the country. The only

di¤erence with respect to the results in tables 1 and 2 is the not signi�cant e¤ect of the interaction

between casualties and the proportion of urban population.

4 Conclusions

Civil wars have many negative consequences. Among them the destruction of civil infrastructures

and the lost of human lives are two of the most important. However, there are many other important

consequences related with the health status of the surviving victims of the civil war which can have

very long lasting e¤ects on the productivity of the economy and the health conditions of the country.

The infection with the malaria parasite is one of these circumstances. The massive movement of

non-immune people across unpaved areas infested with the malaria vector is one of the collateral

16

consequences of civil wars. In this paper we investigate how the likelihood of being infected if people

migrate a¤ects the �ows of refugees during a civil war. We construct a theoretical model to capture

the basic conditions that can explain the size of the �ows of refugees. Malariology treaties insist

on some social conditions that favor the incidence of malaria. Civil wars and force migration are

the most important reasons. Some other men-related conditions, as the proportion of irrigated land

where mosquito larvae can develop, and the proportion of population outside of urban areas, are also

important for the explanation of the incidence of malaria. The model generates clear predictions

about the e¤ect of the intensity of civil wars, the standard of living of the country and the conditions

that favor the incidence of malaria (proportion of irrigated land and proportion of urban population).

The empirical results con�rm the theoretical predictions of the model. The intensity of the civil war

increase the �ow of refugees although in a lower proportion the higher is the standard of living, the

proportion of the irrigated land and the proportion of urban population.

17

References

[1] Ahituv, A., Hotz, J. and T. Philipson (1996), "The responsiveness of the demand of condoms

to the local prevalence of AIDS," Journal of Human Resources, 31 (4), 869-897.

[2] Anderson, R. and R. May (1992), Infectious Diseases of Humans: Dynamics and Control,

Oxford University Press.

[3] Bloland, P. and H. Williams (2003), Malaria control during mass population movements and

natural disasters, The National Academy Press, Washington.

[4] Bruce-Chwatt, L. (1985), Essential Malariology, John Wiley and Sons, New York.

[5] Collier, P., and A. Hoe­ er (2004) Greed and Grievances in Civil Wars, Oxford Economic Papers,

56: 663-595.

[6] Collier, P., (1999), �On the Economic Consequences of civil war�, Oxford Economic Papers, 51:

168:83.

[7] Collier, P., Elliott, V.L., Hegre, H., Hoe­ er, A., Reynal-Querol, M., and Sambanis, N. (2003),

Breaking The Con�ict Trap, Oxford University Press.

[8] Cruz Marques, (1987), �Human Migration and the spread of Malaria in Brasil,�Parasitology

Today, 3: 166-170.

[9] Deininger, K. and L. Squire (1996), "Measuring inequality: a new data-base," World Bank

Economic Review, 10 (3), 565-91.

[10] Doyle, Michael W., and Nicholas Sambanis (2000) �International Peacebuilding: A Theoretical

and Quantitative Analysis.�American Political Science Review 94:4 (December).

[11] EM-DAT: The OFDA/CRED International Disaster Database, www.cred.be/emdat-Universite

Catholique de Louvain-Brussels-Belgium.

[12] Fearon, J. and Laitin, D. (2003). �Ethnicity, Insurgency, and Civil War,�American Political

Science Review, 97 (February )

[13] Geo¤ard, P. and T. Philipson (1996), �Rational Epidemics and their Public Control,�Interna-

tional Economic Review, 37, 3, 603-624.

18

[14] Gersovitz, M. and J. Hammer (2004), "The Economical Control of Infectious Diseases," Eco-

nomic Journal, 114, 1-27.

[15] Gersovitz, M. and J. Hammer (2003), "Infectious Diseases, Public Policy and the Marriage of

Economics and Epidemiology," The World Bank Research Observer, 18 (2), 129-157.

[16] Ghobarah, H. A., Huth, P., anf Russett, B. (2003), �Civil War Kill and Maim People-Long

after Shooting Stops�, American Political Science Review, 97(2).

[17] Global Development Network Growth Database, GDNG, World Bank.

[18] Kremer, M. (2000a), Creating Markets for New Vaccines part I: Rationale, mimeo.

[19] _________ (2000b), Creating Markets for New Vaccines part II: Design Issues, mimeo.

[20] Kermer, M. (1996), �Integrating Behavioral Choice into Epidemiological Models of AIDS,�

Quarterly Journal of Economics, 549-573.

[21] Lacina, B. and N. P. Gleditsch (2005), "Monitoring Trends in Global Combat: A New Dataset

of Battle Deaths," European Journal of Population, 21, 145-166.

[22] Montalvo, J. G. and M. Reynal-Querol (2005), Ethnic polarization, potential con�ict and civil

wars, American Economic Review, 95(3), 796-816.

[23] Montalvo, J. G. and M. Reynal-Querol (2006), �Fighting against malaria: prevent wars while

waiting for the �miraculous�vaccine,�Review of Economics and Statistics, forthcoming.

[24] Murdoch, J., and Sandler T. (2002), �Civil Wars and economic Growth: A Regional Compari-

son�, Defense and Peace Economics, 13(6): 451-64

[25] Najera, J., Liese, B. and J. Hammer (1992), �Malaria: New Patterns and Perspectives,�World

Bank Technical Paper number 183.

[26] Salehyan, I. and K. Skrede-Gleditsch (2006), "Refugees and the Spread of Civil War," Interna-

tional Organization, 60 (2), 335-366..

[27] Schmeidl, S. and J. Jenkins (2001), �Global Refugee and Displaced dataset,�Mershon Center

for International Security, Ohio State University.

[28] Singer, J.D. and M. Small (1982) Resort to arms: International and civil war, 1816-1980.

Beverly Hills, CA: Stage.

19

[29] Singer, J. D. and M. Small (1994), Correlates of war project: International and civil war data,

1816-1992 (data �le; April; ICPSR 9905); Ann Arbor, University of Michigan.

[30] Soares, R. (2006), "The welfare cost of violence across countries," Journal of Health Economics,

forthcoming.

20

Table 1 Determinants of refugees: Fixed effects panel data estimation. 1 2 3 4 5 6 LNGDP 0.00

(0.36) 0.0002(1.59)

0.0002(1.45)

0.0002(1.44)

0.0002 (1.47)

0.0002 (1.49)

DSCW 0.004 (5.75)

0.104(20.53)

0.12(23.27)

0.127(23.80)

0.13 (23.47)

0.13 (23.51)

DSCW* STLIV

-0.0012(-20.12)

-0.0012(-21.65)

-0.0013(-22.04)

-0.0013 (-21.63)

-0.0013 (-21.68)

STLIV 0.00(0.92)

0.00(0.51)

0.00(0.76)

0.00 (0.98)

0.00 (1.08)

DSCW* URBPOP

-0.0004(-9.36)

-0.0003(-5.97)

-0.0003 (-5.50)

-0.0003 (-5.51)

URBPOP 0.00(1.04)

0.00(1.08)

0.00 (0.80)

0.000 (0.84)

DSCW* IRRIG

-0.17(-5.50)

-0.17 (-5.20)

-0.167 (-5.10)

IRRIG -0.02(-0.80)

-0.03 (-0.81)

-0.023 (-0.72)

OPEN 0.00 (0.37)

0.00 (0.41)

NATDIS Drought All constant 0.00

(0.09) -0.003(-1.51)

-0.004(-1.47)

-0.004(-1.34)

-0.004 (-1.33)

-0.005 (-1.45)

N 3366 1136 1136 1075 1006 1006 Numbers in parenthesis are t-statistics

Table 2 Determinants of refugees: Fixed effects panel data estimation. 1 2 3 4 5 6 LNGDP 0.000

(0.51) 0.00

(1.00)0.000(0.76)

0.000(0.43)

0.00 (0.35)

0.00(0.34)

PRIOCW 0.004 (5.34)

0.056(13.70)

0.07(15.68)

0.07(15.88)

0.07 (15.37)

0.073(15.49)

PRIOCW* STLIV

-0.0006(-13.29)

-0.00(-14.00)

-0.0007(-14.09)

-0.007 (-13.58)

-0.00(-13.67)

STLIV -0.000(-1.19)

-0.00(-1.78)

-0.00(-1.62)

-0.00 (-1.25)

-0.00(-1.23)

PRIOCW* URBPOP

-0.00(-7.06)

-0.000(-4.51)

-0.00 (-4.30)

-0.00(-4.32)

URBPOP 0.00(2.34)

0.000(2.29)

0.00 (2.12)

0.00(2.36)

PRIOCW* IRRIG

-0.133(-3.74)

-0.13 (-3.50)

-0.13(-3.58)

IRRIG -0.03(-0.93)

-0.04 (-0.97)

-0.03(-0.94)

OPEN 0.00 (0.66)

0.00(0.71)

NATDIS Drought Allconstant -0.00

(-0.13) 0.001(0.66)

-0.001(-0.37)

-0.00(-0.02)

-0.00 (-0.25)

-0.002(-0.40)

N 3366 1136 1136 1075 1006 1006Numbers in parenthesis are t-statistics

Table 3 Determinants of refugees: Fixed effects panel data estimation. 1 2 3 4 5 6 LNGDP 0.00

(0.55) 0.0001(0.75)

0.0001(0.79)

0.0001 (0.74)

0.0001(0.73)

0.0001 (0.74)

DEATHS 0.397 (4.30)

5.272(5.59)

5.622(5.84)

5.993 (5.96)

6.411(5.97)

6.41 (5.94)

DEATHS * STLIV

-0.06(-5.19)

-0.056(-4.96)

-0.0652 (-5.23)

-0.074(-5.27)

-0.074 (-5.25)

STLIV -0.00(-3.38)

-0.00(-3.48)

-0.0001 (-3.24)

-0.000(-2.42)

-0.00 (-2.41)

DEATHS * URBPOP

-0.035(-1.47)

0.010 (0.30)

0.024(0.65)

0.024 (0.65)

URBPOP 0.00(1.11)

0.000 (0.82)

0.00(0.62)

0.00 (0.67)

DEATHS * IRRIG

-0.082 (-1.85)

-0.096(-2.07)

-0.096 (-2.06)

IRRIG -0.005 (-0.13)

-0.007(-0.19)

-0.007 (-0.19)

OPEN 0.00(1.16)

0.00 (1.17)

NATDIS Drought All constant 0.00

(0.36) 0.008(2.87)

0.006(1.69)

0.0066 (1.68)

0.004(0.89)

0.004 (0.86)

N 3366 1136 1136 1075 1006 1006 Numbers in parenthesis are t-statistics