Content uploaded by Scott Gates
Author content
All content in this area was uploaded by Scott Gates on Aug 13, 2015
Content may be subject to copyright.
WORLD DEVELOPMENT REPORT 2011
BACKGROUND PAPER
CONSEQUENCES OF CIVIL CONFLICT
Scott Gates
Håvard Hegre
Håvard Mokleiv Nygård
Håvard Strand
October 26, 2010
The findings, interpretations, and conclusions expressed in this paper are entirely those of the
authors. They do not necessarily represent the views of the World Development Report 2011
team, the World Bank and its affiliated organizations, or those of the Executive Directors of the
World Bank or the governments they represent.
WDR Background Paper October 26, 2010
Consequences of Civil Conflict∗
Scott Gates, H˚avard Hegre, H˚avard Mokleiv Nyg˚ard & H˚avard Strand
October 26, 2010
Executive Summary
This paper reviews the literature on the development consequences of internal armed conflict
and state fragility and analyzes the relationship using data from World Development Indicators,
UCDP/PRIO Armed Conflict Data, and World Bank state fragility assessments. Our main focus
is on a set of development indicators that capture seven of the Millenium Development Goals,
but we also look briefly into the effect of conflict and fragility on growth, human rights abuses,
and democratization. We analyze these relationships using a variety of methods – averages by
conflict and fragility status; cross-sectional regression analyses of change in each indicator over
the time frame for which we have data; fixed-effects regression analyses of the impact on each
indicator for each five-year period 1965-2009; as well as occasional panel time series models and
matching techniques.
The analyses leave no doubt that conflict, fragility and poor development outcomes are
closely related – these problems largely occur in the same set of developing countries, most of
which are located in Asia and Sub-Saharan Africa. Acknowledging the difficulty of analyzing
the effect of conflict on a set of indicators that we know are also causally related to the onset of
conflict, we still conclude that conflict and fragility at least exacerbate these pre-existing con-
ditions. Conflict and fragility are indeed major obstacles to development for several indicators.
The table summarizes our findings, indicator by indicator.
MDG Label Indicator Effect of Effect of
conflict fragility
Cross-section Fixed-effects Cross-section Fixed-effects
MDG 1 Ending Poverty Undernourishment Detrimental Detrimental Detrimental Unclear
MDG 1 and Hunger Poverty Headcount Detrimental Detrimental Detrimental No effect
MDG 1 Life expectancy Detrimental Detrimental Detrimental Detrimental
MDG 1 GDP per capita Detrimental Detrimental Detrimental Detrimental∗
MDG 2 Universal Prim. Sch. Enrollment Detrimental Detrimental Detrimental Beneficial?
MDG 2 Education Sec. Sch. Attainment Detrimental Unclear Detrimental Unclear
MDG 3 Gender Parity Primary School ratio Detrimental Beneficial? Detrimental No effect
MDG 3 Life expect. ratio No effect Unclear No effect No effect
MDG 4 Child Mortality Infant Mortality Detrimental Detrimental Detrimental Detrimental
MDG 4 Under-5 Mortality Detrimental Detrimental Detrimental Detrimental
MDG 5 Maternal Mort. Birth Attendance No effect Unclear Detrimental Beneficial?
MDG 6 Combat AIDS % HIV positive Beneficial? Beneficial? Beneficial? No effect
MDG 7 Environmental Access to Water Detrimental Detrimental Detrimental Detrimental
MDG 7 Sustainability Access to Sanitation No effect Unclear No effect No effect
*:Estimated on growth in GDP per capita using OLS with panel-corrected standard errors.
∗We thank the World Bank and the Norwegian Ministry of Foreign Affairs for support. We especially thank
Gary Milante, Sarah Cliffe, Colin Scott, Lene Lind, and Nadia Piffaretti at the World Bank, and Olaf DeGroot and
Tilman Br¨uck at DIW, as well participants at a World Bank Brownbag Seminar, Households in Conflict Network
Workshops (in Berlin and Lisbon), a Norwegian Ministry of Foreign Affairs workshop, and the Tinbergen meetings
in Amsterdam.
1
WDR Background Paper October 26, 2010
Contents
1 Introduction 4
2 Methodology 5
2.1 Countries included in analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 Data............................................. 5
2.3 Conflict Country Categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.4 ModelSpecification..................................... 7
2.4.1 Cross-sectionalmodels............................... 9
2.4.2 Country Fixed-effects models . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.4.3 Autocorrelation................................... 11
2.4.4 Matching ...................................... 11
3 Overview of Effects of Conflict 12
3.1 Conflict, fragility, and gaps in development outcomes . . . . . . . . . . . . . . . . . . 12
3.2 Is the gap caused by conflict and fragility? . . . . . . . . . . . . . . . . . . . . . . . . 21
3.3 Summary of results from our statistical analysis . . . . . . . . . . . . . . . . . . . . . 24
3.4 Conflict and the attainment of the Millennium Development Goals . . . . . . . . . . 28
4 Analysis of Individual Indicators 28
4.1 MDG 1: Ending Poverty and Hunger . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.1.1 GlobalTrends.................................... 28
4.1.2 Literature on Effects of Conflict on Poverty and Hunger . . . . . . . . . . . . 29
4.1.3 EmpiricalAnalysis................................. 32
4.2 MDG 2: Universal Education . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.2.1 GlobalTrends.................................... 40
4.2.2 Literature on Effects of Conflict on Education . . . . . . . . . . . . . . . . . . 41
4.2.3 EmpiricalAnalysis................................. 42
4.3 MDG3:GenderParity .................................. 44
4.3.1 GlobalTrends.................................... 44
4.3.2 Literature on Effects of Conflict on Gender Equality . . . . . . . . . . . . . . 45
4.3.3 EmpiricalAnalysis................................. 46
4.4 MDG 4: Infant Mortality Rates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.4.1 GlobalTrends.................................... 48
4.4.2 Literature on Effects of Conflict on Infant Mortality . . . . . . . . . . . . . . 49
4.4.3 EmpiricalAnalysis................................. 49
4.5 MDG 5: Maternal Mortality/Birth Attendance . . . . . . . . . . . . . . . . . . . . . 51
4.5.1 GlobalTrends.................................... 51
4.5.2 Literature on Effects of Conflict on Maternal Mortality . . . . . . . . . . . . 52
4.5.3 EmpiricalAnalysis................................. 53
4.6 MDG6:CombatHIV/AIDS ............................... 54
4.6.1 GlobalTrends.................................... 54
4.6.2 Literature on Effects of Conflict . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.6.3 EmpiricalAnalysis................................. 55
4.7 MDG 7: Environmental Sustainability . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2
WDR Background Paper October 26, 2010
4.7.1 GlobalTrends.................................... 56
4.7.2 Literature on Effects of Conflict . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.7.3 EmpiricalAnalysis................................. 59
A Appendix 64
A.1 Listofcountries....................................... 64
A.2 List of conflict country matches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
A.3 RegressionResults ..................................... 66
A.3.1 MDG 1: Ending Poverty and Hunger . . . . . . . . . . . . . . . . . . . . . . . 66
A.3.2 MDG 2: Universal Education . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
A.3.3 MDG3:GenderParity .............................. 73
A.3.4 MDG 4: Child Mortality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
A.3.5 MDG 5: Maternal Mortality . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
A.3.6 MDG 6: Combat HIV/AIDS . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
A.3.7 MDG 7: Environmental Sustainability . . . . . . . . . . . . . . . . . . . . . . 81
3
WDR Background Paper October 26, 2010
1 Introduction
War is a development issue. War kills, but its consequences extend far beyond direct deaths. In
addition to battlefield casualties, armed conflict often leads to forced migration, refugee flows, and
the destruction of societies’ infrastructure. Social, political, and economic institutions are indelibly
harmed. The consequences of war, especially civil war, for development are profound. This paper is
a statistical analysis of the consequences of conflict. The effects of armed conflict are evaluated with
respect to the achievement of the Millennium Development Goals; economic growth; the political
institutions of a state; and human rights. The direct and indirect mechanisms through which violent
conflict degrades population health are also evaluated.
In Section 2, we summarize our methodological choices and present our conflict data. Section
3 summarizes the results of our analysis. Section 4 analyzes the effects of internal armed conflict
on the attainment of the individual Millennium Development Goals.
Not all the consequences of armed conflict have ever been measured, and some are not even
measurable. Indeed, many consequences of armed conflict are not incorporated in our analysis, such
as the increased number of young males with war experience; the accumulation of light weapons
subsequently used in violent crime; traumatic experiences (Ringdal, Ringdal and Simkus 2008);
erosion of trust and emergence of ethnic prejudice (Strabac and Ringdal 2008); and so on. Another
burden difficult to measure is the environmental impact. Few indicators allow a systematic com-
parison of this burden. We show the detrimental effect of conflict on the accessibility of water and
adequate sanitation facilities, which are indicators with a considerable environmental component.
But other environmental outcomes are difficult to assess because the impact of war differs from
one place to another. In some countries, such as Cambodia and Liberia, conflict sets the stage for
large-scale illegal logging; in other places, other aspects of environmental regulation break down;
and elsewhere, unexploded ordinances is a major problem caused by armed conflict. Such problems
(missing data or unmeasurable variables) make it especially difficult to systematically assess the
economic, political, social, environmental, and health effects of conflict.
We show how civil war harms the achievement of most of the indicators for which we have data.
The results are consistent across most of our indicators. This suggests that war is also detrimental
4
WDR Background Paper October 26, 2010
for development outcomes for which we have not been able to do any quantitative analysis, but
that are highly correlated with the indicators that we do look into.
2 Methodology
2.1 Countries included in analysis
In most of our analyses, we link conflict or fragility status to improvements in development indi-
cators. We expect conflict and fragile states to have less improvement than countries that avoid
these political problems. However, many of the indicators have a natural maximum: Primary
education attainment cannot exceed 100%; infant mortality rates (IMR) can hardly go below 5
per 1,000; and measures such as our democracy index have a fixed maximum. Many industrial
countries have reached the maximum values for many indicators, and do not improve much beyond
that level. Also, these countries have no armed conflict (or relatively limited conflicts such as the
one in Northern Ireland). To avoid our analysis being affected by the non-improvement in these
countries, we remove all (but one: South Africa) of the countries classified as industrialized in the
first World Bank Development Report (World Bank 1978, p.77) and a few other countries that we
regard as industrialized by the 1970s. 1
2.2 Data
We alternate between three datasets in our analysis. Most of the outcome indicators are measured
in five-year intervals, so most analyses are based on a dataset containing one observation for each
country for each five-year period. However, for the growth and democracy indicators, we use a
country-year design with one observation for each country for each year. For our cross-sectional
analysis, we use a dataset with one observation per country.
The conflict data come from the Uppsala Conflict Data Program (UCDP), the most comprehen-
sive, accurate, and widely used data source on global armed conflicts. The versions of these data we
1The industrial countries we exclude are Austria, Australia, Belgium, Canada, Denmark, Finland, France, Ger-
many, Greece, Iceland, Ireland, Italy, Japan, Luxembourg, Netherlands, New Zealand, Norway, Portugal, Spain,
Sweden, Switzerland, United Kingdom, and United States. We retain South Africa because only parts of it can be
said to be fully industrialized. A complete list of countries included is found in Table A-1.
5
WDR Background Paper October 26, 2010
used were backdated and adapted for statistical use in collaboration with PRIO and is referred to
as the UCDP/PRIO Armed Conflict Data (Gleditsch et al. 2002; Harbom and Wallensteen 2009).
UCDP defines armed conflict as a contested incompatibility that concerns a 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. A civil (or intrastate) conflict occurs between
a government and a non-government party. This definition of armed conflict is becoming a standard
in how conflicts are systematically defined and studied. In the gap table presented in Section 3, we
restrict the definition to conflict that have accumulated 1,000 deaths over a multi-year period.
Updates to these data have been published annually in the report series States in Armed Conflict
since 1987, in the SIPRI Yearbook since 1988, the Journal of Peace Research since 1993, and in
the Human Security Report since 2005. The data were also used in the World Bank PRR Breaking
The Conflict Trap (Collier et al. 2003). The World Bank co-funded the backdating of these data
from 1946 to 1989.
We use three measures of amount of conflict in the preceding five-year period. The first, we
call conflict, measures the number of years within the preceding five-year period with conflict in
the country as recorded in the UCDP/PRIO dataset (Gleditsch et al. 2002). A country without
conflict the previous period receives a score of 0; a country with only a one-year minor conflict, a
score of 1; and a country with minor conflict in each of the five years is assigned a 5. If the conflict
was recorded as major (more than 1,000 battle-related deaths within a year), each year of conflict
is counted twice. Thus, a country with five years of major conflict receives the maximum score of
10.
The second conflict measure we call battle deaths: the log of the count of battle-related deaths
caused by fighting in the five years preceding the observation period. About 20% of the country-
periods in our dataset have conflicts. The median conflict period led to about 2,500 deaths. The
most destructive conflict periods (in Afghanistan and Cambodia) caused over 200,000 deaths each.
In the cross-sectional analyses, we count the total number of years the country has been in
either minor or major conflict over the time period analyzed.
We also estimate the effect of state fragility as coded in the IDA Fragile States Dynamic List.
6
WDR Background Paper October 26, 2010
Countries are coded as fragile if they either host peacekeeping missions or have a low score on the
World Bank’s Country Policy and Institutional Assessment (CPIA) rate. Countries have low CPIA
scores if their policies and/or institutions are weak in terms of economic management, structural
policies, policies for social inclusion and equity, and public sector management and institutions.2
The variable has the value 1 if the country is regarded as fragile in at least one of the preceding
five years, and the value 0 if it was not coded as fragile in any of these years.
2.3 Conflict Country Categories
In a number of figures, we present information classified by conflict country category. We group
countries into three categories: countries that have had no conflict between 1980 and 2008 (non-
conflict), countries that had at least one year of conflict in the 1981–1990 period but no conflicts
thereafter (post-conflict), and countries that had conflicts during the 1991–2008 period (conflict).
For the gap tables (Tables 1 and 4) we classify countries into four mutually exclusive categories.
2.4 Model Specification
We present three types of analyses of the relationships among conflict, fragility, and our outcome
variables. First, we look at simple comparisons between countries within each conflict country cat-
egory. Most earlier studies of the effects of conflict on development outcomes use this methodology.
We also compare indicators such as the percentage of the population that suffers from undernour-
ishment for countries with conflict with the same indicators for countries without conflict. Figure 1
2More precisely, states are coded as fragile in years where they:
1. For observations for years t= 2004 −2008: have CPIAt<3.2
2. For years t= 1978 −2003: have CPIAt<(Standardized CPIA Cutoff for year t). The cutoff is calculated
as the Average CP I At+ (Z-score*Standard Deviation (CPIA)) where Z-score = (3.24–(Sample Averages for
2005 −2009 /(Sample Standard Dev. 2005 −2009 ))
3. For all years: Have an ongoing international or regional peacekeeping or political (non-border) mission, in-
cluding special SRSG friends of political missions.
4. Are low-income countries without a CPIA score.
We use the harmonized list of fragile situations. A country is eligible to graduate out of fragility when
CPIAt>Cutoff and it has not qualified in any other way for the previous three years. This rule works in re-
verse as well: A country only relapses into fragility if it has a CP I A <Cutoff for three years or meets other criteria
above or below. Countries which are non-IDA before they join IDA but are fragile once they join IDA (or receive a
CPIA score) are coded as fragile in the years preceding.
7
WDR Background Paper October 26, 2010
exemplifies this analysis by showing the distribution of undernourishment for conflict, post-conflict,
and non-conflict countries (Undernourishment by Conflict) and the corresponding distribution for
fragile versus non-fragile states (Undernourishment by Fragility).3
Figure 1: The Percentage of Population Suffering from Undernourishment, By Conflict Status and
State Fragility in 2005
The box plots show clearly that populations in conflict countries on average suffer more from
undernourishment than those in non-conflict countries, and that post-conflict countries are located
between the two. However, this method is problematic methodologically, since it is not certain that
these differences are caused by conflicts. Undernourishment is closely associated with other aspects
of underdevelopment. Most conflict studies confirm that development, as measured by GDP per
capita or energy consumption per capita, is among the most robust predictors of civil war (Hibbs
1973; Hegre et al. 2001; Fearon 2003; Collier et al. 2003; Hegre and Sambanis 2006). Underdevelop-
ment in a general sense clearly facilitates both the occurrence of conflict and of undernourishment.
It is necessary to account for these factors to avoid attributing development effects to factors that
tend to cause conflicts in the first place.
We estimate three types of statistical models that attempt to account for this problem. The
first type is a set of cross-sectional models where the dependent variable is the improvement from
the first to the last observation for each country. The second is a set of fixed-effects models. We
3The median value in each group is given by the vertical line inside the box in the center of each box-whisker
combination. The outer values of this box are the 25th and 75th percentiles. The whiskers represent the ‘adjacent
values’ (Tukey 1977) – the upper adjacent value is the largest value smaller than x75 +3
2(x75 −x25). The dots are
extreme observations outside the adjacent values.
8
WDR Background Paper October 26, 2010
also estimate some models that take auto-correlation into account. Finally, we have estimated most
models using matching techniques. We do not report the results from the matching models here.
In most cases, they yield results that are consistent with the fixed-effects models.
2.4.1 Cross-sectional models
The dependent variable in our cross-sectional models (for example, Table A-3) is the improvement
within each country between the first and the last non-missing observation for each of the MDG
indicators. In other words, we examine changes in the gap between countries that are in conflict
and that not in conflict, as well as between fragile and non-fragile states. The models include
three control variables. The first is exposure – the number of years over which the improvement
is measured. For some indicators, we have data from the 1960s to 2007 or 2008; for others, only
a decade or so. Most countries have reduced problems, such as undernourishment and infant
mortality, over time, and this improvement is likely to be larger the longer the period we analyze.
The second control variable, ‘(firstnm) ...’, is the value for the indicator for the first year. Countries
that were very poor at the outset of the period may improve more than countries that were rich.
The third control variable is log population size. Our models also include a set of region indicator
variables.
We measure conflict with three different operationalizations. The first column in each of these
tables reports the results from the first, which consists of the variables war and minor. These
two variables count the number of years that the country in question has seen either war (1,000+
persons killed per year in battle-related situations) or minor armed conflict (more than 25 persons
killed per year but less than 1,000). The second reports the total number of battle-related fatalities
over the period. The third column is the log of battle deaths accumulated over the observation
period.
In the fourth column of each cross-sectional analysis table we estimate a model with the fragility
indicator as well as for the CPIA score alone.
9
WDR Background Paper October 26, 2010
2.4.2 Country Fixed-effects models
Cross-sectional analysis of improvement (that is, closing the gap) accounts for some of the factors
that affect both the development outcome and the risk of internal armed conflict. To handle
this problem more systematically, we follow Iqbal (2010), arguably the most comprehensive and
sophisticated study of the health consequences of conflict, in using fixed-effects regression models.
These models remove between-countries differences in the outcome variables and concentrate on
the within-country effects. If conflicts increase undernourishment, we should observe an increase
relative to the country’s average levels in the indicator during the conflict or in the period following
the conflict. The fixed-effects models estimate the systematic within-country effect of conflicts.
Fixed-effects models may overprotect against such omitted-variable bias. In particular, countries
that have had conflicts constantly over the entire period for which we have data will not contribute
much to the estimated effect of conflict – conflict is then largely part of the fixed effect itself. Since
these countries are also likely to be the most severely affected by conflict, a fixed-effects model
may yield too-conservative estimates. This is accentuated by having data only for relatively short
periods. Some countries may be poor when our data series start (typically at some year from 1970
to 1990) because of conflicts they have had before then. Our models will also ignore this effect.
Still, we choose to present a set of conservative estimates. For some indicators, this probably
prevents us from identifying an effect of conflict. For other indicators we find substantial detrimental
effects of conflict despite these limitations. This is particularly true for indicators for which we have
long time series.
Time trends Most indicators have trends that show improvement in the MDG indicators. Given
these strong trends, conflict countries may also improve the general situation in the country. To
account for such trends, we include dummy variables for each five-year period in the fixed-effects
models.
10
WDR Background Paper October 26, 2010
2.4.3 Autocorrelation
Subsequent observations for the same country may be dependent on each other. To account for
this, we also estimate a set of population-averaged models with an AR1correlation model for the
error terms within each country.
2.4.4 Matching
While the fixed-effects approach effectively eliminates any omitted-variable bias from country-
specific variables such as culture and geography, fixed-effects models cannot account for unobserved
variables that vary over time. An alternative approach is matching (Ho et al. 2007). In a fully
randomized experiment, the effect of a given variable can be gauged directly. By matching ob-
servations, we approach the experimental situation with real-world data. In our case, we want to
match particular observations of countries that have had conflict with otherwise similar countries
that have not had conflict, and then observe how these countries differ as an effect of the conflict.
The literature on matching has not converged towards a single best practice for a given problem
(Hill 2008). Given the relative complexity in the assessment of the underlying assumptions, we rely
on the R MatchIt package (Ho et al. 2007). This package alternates our dataset to create a better
balance between our conflict and non-conflict samples, and provides accessible verification of the
post-hoc compliance with our model’s assumptions. Following Iacus, King and Porro (2009), we
employ a Coarsened Exact Matching (CEM) model. In exact-matching models, balance is sought by
matching all units with the exact same value on explanatory variables, thereby forming subclasses
that, in turn, should differ only with respect to the treatment variable (here, the treatment is the
occurrence of conflict). Exact matching lends itself poorly to continuous explanatory variables such
as the number of battle deaths over the last five years. CEM solves this problem by recoding the
continuous variables into categorical variables and then applying the exact matching formula. This
method is superior to distance-oriented models (Abadie and Imbens 2006) since it effectively limits
how different a matched pair can be.
Having produced a fairly balanced set of matches, the need for control variables is effectively
removed. Ideally, all observations within a match strata, except for the treatment, should be
11
WDR Background Paper October 26, 2010
similar. Although this would be the case with an exact matching design, the recoding of the CEM
procedure retains some imbalance, justifying the inclusion of control variables. Following Blackwell
et al. (2009) we estimate the sample average treatment effect on the treated unit through an OLS
regression with CEM weights applied.
Using this technique, we have created a set of matched pairs, which allows us to compare
countries that experience conflict and those that did not. The matching is done on the basis of
the countries’ values for population, GDP per capita, education levels, and ethnic fractionalization.
(See Table A-2. Note that the list does not include all conflict countries – the CEM routine excludes
cases where no match can be located.
3 Overview of Effects of Conflict
3.1 Conflict, fragility, and gaps in development outcomes
We analyze the effect of conflict on seven Millennium Development Goals, represented by 14 different
indicators. There is no doubt that conflict countries and fragile states have less-favorable scores
for all of these indicators.4To illustrate these differences, we have grouped the world’s developing
countries into five groups according to their status over the 2003–2008 period: countries with armed
conflicts causing at least 1,000 battle deaths (conflict countries); fragile countries without ongoing
conflict (fragile states); countries that do not have ongoing conflicts and are not fragile, but have
been so in at least one of the preceding 10 years (post-fragile/post-conflict); India and China as a
separate category; and countries that have neither had conflicts nor been fragile in any of the 11
years from 10 years ago to the current year (other countries).5The world map in Figure 2 shows
the classification. Conflict and fragile states are most common in Sub-Saharan Africa and in an
Asian belt ranging from Turkey in the west to Papua New Guinea in the east.
Table 1 shows the inter-group gaps by summarizing the 14 MDG indicators separately for each
of the six groups. Among the 146 developing countries in our dataset in 2008, 19 were in conflict,
22 were fragile but not in conflict, 17 post-conflict or post-fragile, 85 neither of these, and India,
4With the possible exception of prevalence of HIV/AIDS.
5See Section 2.2 for precise definitions of all categories.
12
WDR Background Paper October 26, 2010
Figure 2: The Countries in the World as Classified in Table 1
The Gap Map - Categories
Legend
Industrialized
Conflict
Fragile
Post-conflict/fragile
Other Developing Countries
China/India/Russia
China, and Russia were treated as a separate group. We report the MDG indicators for the latest
year for which we have sufficient data coverage.6The first line in the table reports mean and median
population size in each group. Notice that conflict countries are, on average, larger than those in
other categories, and that fragile states are smaller. This is probably due to the strong correlations
between fragility and risk of conflict on the one hand and population size and risk of war on the
other. Most large, fragile states are likely to also have conflict and are therefore classified in the
first group.
In Table 2 we show the similar gaps summarized over another set of groups: fragile states, post-
fragile states, other developing countries, and India/China/Russia. Among the 146 developing
countries, 30 were fragile (some of them also in conflict), 14 post-conflict or post-fragile, and 99
neither of these.
It is clear from Tables 1 and 2 that conflict countries and fragile states perform worse than
6For several of the indicators, some countries have more recent data than the year indicated here. We have
chosen to use a less recent time point if that gave us a considerably larger number of countries to analyze. For the
demographic variables (infant mortality, population) we use UN estimates for 2008.
13
WDR Background Paper October 26, 2010
Table 1: Gap Table – Averages and medians by Comparison Group. Conflict and Fragility Classification
Year All Conflict Fragile Post-conflict/ India/China/ Other deve-
countries countries states fragile Russia loping countries
Mean Median Mean Median Mean Median Mean Median Mean Median Mean Median
Population 2008 39.9 9.4 52.3 31.9 9.7 6.2 25.1 12.7 888.1 1,186.2 18.9 6.1
Countries 2008 146 146 19 19 22 22 17 17 3 3 85 85
Undernourishment 2005 15.6 11.0 20.9 17.0 29.1 32.0 23.0 22.0 14.0 9.0 11.4 5.0
Poverty 2003 29.1 18.4 26.7 31.4 37.0 37.5 51.2 42.5 31.1 24.2 19.8 10.0
Life exp 2007 67.1 68.8 65.8 65.5 57.3 57.6 55.5 64.8 69.1 67.6 69.2 72.4
GDPcap 2007 1932 1675 1552 1033 623 431 657 532 1369 1811 3642 3090
Primedu 2005 88.3 89.4 79.8 88.9 72.2 61.4 70.8 84.8 93.8 97.9 90.8 91.6
Secedu 2008 0.70 0.68 0.64 0.64 0.51 0.50 0.56 0.47 0.78 0.87 0.68 0.76
PrimeduRatio 2005 95.6 97.2 89.7 91.2 87.5 84.7 87.7 96.8 98.2 99.6 98.2 98.1
LifeexpRatio 2007 1.06 1.06 1.05 1.05 1.07 1.06 1.05 1.07 1.06 1.05 1.07 1.07
IMR 2008 42.5 30.4 49.3 64.9 79.8 68.2 82.4 62.7 36.9 23.0 31.3 19.4
Underfivemort 2007 55.0 34.7 64.5 54.7 111.0 100.3 125.7 70.2 43.6 21.9 41.4 23.8
Attendance 2003 70.4 88.5 63.6 55.5 52.0 52.7 43.4 56.8 73.9 95.5 78.2 96.9
HIVprev 2007 1.0 0.8 0.7 0.6 3.1 1.8 2.5 0.7 0.2 0.3 1.9 0.6
Water 2006 84.4 88.0 80.9 80.0 70.8 70.0 58.9 77.0 88.9 89.0 87.7 93.0
Sanitation 2006 55.5 68.5 60.4 46.5 50.3 41.0 37.9 48.0 49.9 65.0 68.6 83.0
14
WDR Background Paper October 26, 2010
the other countries for most of the MDG indicators. Among fragile states, the mean proportion of
populations undernourished is 29%, as contrasted to 11% in other countries. In conflict countries,
the average proportion undernourished is lower than in fragile states (about 19%), which might
be because state fragility is more detrimental to nourishment than overt armed conflicts are, but
it might also be due to a size effect. Conflict countries are, on average, larger than non-conflict
countries, and conflicts are often partly local and rarely affect the entire population in large countries
(Buhaug and Gates 2002; Buhaug and Rød 2006; Raleigh et al. 2010). Measuring the effect of
conflict using country-level indicators underestimates the local effect of conflict, in cases. On the
other hand, the fragile states are predominantly small countries, and their fragility is likely to affect
their entire population.
Comparing the mean and median values for the indicators is also instructive. For many indica-
tors (for example, undernourishment, poverty, and HIV prevalence), the median value is consider-
ably lower than the mean value in the other-countries group, suggesting a right-skewed distribution
in this group (this is evident in the box plots in Figure 1). There are much smaller differences
between the mean and median for the conflict country and fragile state groups. A minority of the
non-conflict countries performs poorly in terms of the MDGs, but most of the conflict and fragile
countries have poor development outcomes. Comparing median values is more appropriate in this
case, because the median ‘other country’ is more typical than the mean within the category. Table
4 shows an even larger distance between this group of countries and the fragile states and conflict
countries.
In Table 4, we have calculated the number of individuals affected for a subset of the MDG
indicators. The first line reports the total population in 2008 in each of the five categories of
developing countries. About 1.35 billion people live in conflict or fragile countries – about 20%
of the population we study here. Among the 1.1 billion inhabitants in conflict countries, 19%
are estimated to be undernourished, or 210 million people. About 30% of the developing world’s
undernourished inhabitants live in fragile states or conflict countries.
In the tables that present the number of people affected, we present some of the indicators
differently from the more traditional way we look at them other places in this paper. For MDG 2,
15
WDR Background Paper October 26, 2010
Table 2: Gap Table – Averages and medians by Comparison Group. Fragility Classification
Year All Fragile Post- Other deve- India/China/
states fragile states loping countries Russia
Mean Median Mean Median Mean Median Mean Median Mean Median
Population 2008 39.9 9.4 13.6 7.8 27.0 12.4 24.0 8.5 888.0 1,186.2
Countries 2008 146 146 30 30 14 14 99 99 3 3
Undernourishment 2005 15.6 11.0 36.4 32.0 23.8 24.0 13.0 5.0 14.0 9.0
Poverty 2003 29.1 18.4 47.6 48.8 56.4 45.7 19.6 10.5 31.1 24.2
Life exp 2007 67.1 68.8 54.7 57.4 53.9 62.1 69.1 72.3 69.1 67.6
GDPcap 2007 1932.1 1674.9 437.1 384.3 466.2 494.7 2987.6 2608.6 1369.3 1811.2
Primedu 2005 88.3 89.4 59.2 58.3 68.1 82.7 89.7 92.0 93.8 97.9
Secedu 2008 0.70 0.68 0.49 0.45 0.53 0.44 0.68 0.75 0.78 0.87
PrimeduRatio 2005 95.6 97.2 83.4 84.7 86.1 95.2 96.2 98.1 98.2 99.6
LifeexpRatio 2007 1.06 1.06 1.06 1.06 1.04 1.07 1.07 1.07 1.06 1.05
IMR 2008 42.5 30.4 90.0 79.7 87.7 58.8 33.4 19.8 36.9 23.0
Underfivemort 2007 55.0 34.7 129.6 108.6 136.0 80.5 42.4 24.1 43.6 21.9
Attendance 2003 70.4 88.5 46.3 49.0 39.6 48.0 75.0 95.4 73.9 95.5
HIVprev 2007 1.0 0.8 2.5 1.7 2.8 0.8 1.4 0.5 0.2 0.3
Water 2006 84.4 88.0 63.3 67.0 56.0 65.0 87.4 93.0 88.9 89.0
Sanitation 2006 55.5 68.5 40.8 33.0 34.6 48.0 68.1 81.5 49.9 65.0
16
WDR Background Paper October 26, 2010
Table 3: Gap Table – Averages and medians by Comparison Group. With Homicide Classification
Year All Conflict Fragile Post-conflict/ High-homicide Other deve-
countries countries states fragile countries loping countries
Mean Median Mean Median Mean Median Mean Median Mean Median Mean Median
Population 2008 39.9 9.4 52.3 31.9 9.7 6.2 25.1 12.7 26.6 6.2 55.1 6.3
Countries 2008 146 146 19 19 22 22 17 17 23 23 65 65
Undernourishment 2005 15.6 11.0 20.9 17.0 29.1 32.0 23.0 22.0 6.9 11.0 14.1 5.0
Poverty 2003 29.1 18.4 26.7 31.4 37.0 37.5 51.2 42.5 9.3 14.6 30.3 5.0
Life exp 2007 67.1 68.8 65.8 65.5 57.3 57.6 55.5 64.8 69.4 70.2 69.0 72.9
GDPcap 2007 1932.1 1674.9 1551.9 1033.1 622.7 431.3 657.3 532.3 4056.9 2665.3 1870.5 3083.5
Primedu 2005 88.3 89.4 79.8 88.9 72.2 61.4 70.8 84.8 94.7 91.0 92.4 92.0
Secedu 2008 0.70 0.68 0.64 0.64 0.51 0.50 0.56 0.47 0.75 0.69 0.74 0.82
PrimeduRatio 2005 95.6 97.2 89.7 91.2 87.5 84.7 87.7 96.8 97.0 98.6 98.4 98.1
LifeexpRatio 2007 1.06 1.06 1.05 1.05 1.07 1.06 1.05 1.07 1.12 1.08 1.05 1.06
IMR 2008 42.5 30.4 49.3 64.9 79.8 68.2 82.4 62.7 23.1 24.1 36.9 18.0
Underfivemort 2007 55.0 34.7 64.5 54.7 111.0 100.3 125.7 70.2 27.3 31.6 45.5 16.6
Attendance 2003 70.4 88.5 63.6 55.5 52.0 52.7 43.4 56.8 92.7 92.0 72.4 98.0
HIVprev 2007 1.0 0.8 0.7 0.6 3.1 1.8 2.5 0.7 2.4 1.1 0.4 0.3
Water 2006 84.4 88.0 80.9 80.0 70.8 70.0 58.9 77.0 93.1 93.0 87.7 94.5
Sanitation 2006 55.5 68.5 60.4 46.5 50.3 41.0 37.9 48.0 78.7 77.0 52.4 87.0
17
WDR Background Paper October 26, 2010
Table 4: Gap Table – Millions of people affected. Conflict and Fragility Classification
Year All Conflict Fragile Post-conflict/ India/China/ Other deve-
countries countries states fragile Russia loping countries
Population 2008 5827.3 994.0 214.5 426.9 2664.3 1527.7
Countries 2008 146 19 22 17 3 85
Undernourishment 2005 909.2 207.8 62.4 98.0 373.8 173.9
Poverty 2003 1694.8 265.2 79.2 218.5 828.4 302.6
No primary education 2005 115.4 38.0 12.3 25.7 25.6 23.3
No secondar education 2008 97.3 21.0 6.6 11.1 32.2 27.3
Infant mortality 2008 4.8 1.1 0.4 0.9 1.6 0.9
Under-five mortality 2007 31.0 7.3 3.1 7.2 9.6 6.0
No birth attendance 2003 1723.7 361.4 102.9 241.6 696.5 332.5
HIV positive 2007 56.2 7.1 6.6 10.7 6.5 29.4
Without water 2006 908.4 189.5 62.6 175.4 294.9 187.3
Without sanitation 2006 2594.6 393.9 106.5 265.2 1334.4 480.1
Table 5: Gap Table – Millions of people affected. Fragility Only Classification
Year All Fragile Post- Other deve- India/China/
countries states fragile loping countries Russia
Population 2008 5827.3 408.4 377.4 2377.2 2664.3
Countries 2008 146 30 14 99 3
Undernourishment 2005 909.2 148.5 89.7 309.1 373.8
Poverty 2003 1694.8 194.3 212.8 467.0 828.4
No primary education 2005 115.4 35.6 25.0 41.9 25.6
No secondary education 2008 97.3 12.8 10.2 43.0 32.2
Infant mortality 2008 4.8 1.1 0.9 1.6 1.6
Under-five mortality 2007 31.0 7.6 7.1 9.9 9.6
No birth attendance 2003 1723.7 219.2 228.0 593.4 696.5
HIV positive 2007 56.2 10.2 10.5 34.5 6.5
Without water 2006 908.4 149.8 166.2 300.1 294.9
Without sanitation 2006 2594.6 241.9 246.8 758.3 1334.4
18
WDR Background Paper October 26, 2010
for example, we present the percentage of children not enrolled in primary education, rather than
the percentage that is. We do the same for secondary school non-attainment, births non-attendance,
and lack of access to water and sanitation.
We calculate the number of children that do not attain primary education by first computing
the total population in each age group for each conflict/fragility category and multiplying with the
proportions affected (reported in Table 1).7We estimate that 38 million out of about 230 million
children in conflict countries and fragile states fail to attain primary education, or about 4 million
children per year. About 30% of the children that fail to complete primary education live in these
countries.
In Table 5, we show the number of people affected in each of the five-category variables that
only classify countries in terms of fragility.
Table 6: Gap Table – Millions of people affected. Fragility Only Classification - including Pakistan
and Kenya
Year All Fragile Post- Other deve- India/China/
countries states fragile loping countries Russia
Population 2008 5827.3 613.9 377.4 2171.7 2664.3
Countries 2008 146.0 32.0 14.0 97.0 3.0
Undernourishment 2005 909.2 197.2 89.7 258.5 373.8
Poverty 2003 1694.8 242.2 212.8 403.6 828.4
Primedu 2005 115.4 49.3 25.0 30.2 25.6
Secedu 2008 97.3 18.8 10.2 37.4 32.2
IMR 2008 4.8 1.4 0.9 1.3 1.6
Underfivemort 2007 31.0 9.9 7.1 7.8 9.6
Attendance 2003 1723.7 330.0 228.0 484.2 696.5
HIVprev 2007 56.2 9.9 10.5 34.5 6.5
Water 2006 908.4 182.7 166.2 266.9 294.9
Sanitation 2006 2594.6 334.2 246.8 664.9 1334.4
7We base these estimates on data from United Nations (2007) that give countries’ populations grouped in five-year
intervals, e.g. 0–4 years, 5–9 years, etc. To calculate the population of primary school age, we add the 10–14 year
population and 80% of the 5–9 year population. For secondary school enrollment, we use 60% of the 15–19 year
population. For infant mortality, we use the population in the 0–4 year category divided by 5.
19
WDR Background Paper October 26, 2010
Table 7: Gap Table – Millions of people affected. With-Homicide Classification
Year All Conflict Fragile Post-conflict/ High-homicide Other deve-
countries countries states fragile countries loping countries
Population 2008 5827.3 994.0 214.5 426.9 612.3 3579.7
Countries 2008 146 19 22 17 23 65
Undernourishment 2005 909.2 207.8 62.4 98.0 42.2 506.5
Poverty 2003 1694.8 265.2 79.2 218.5 57.2 1086.4
No primary education 2005 115.4 38.0 12.3 25.7 5.0 43.6
No secondary education 2008 97.3 21.0 6.6 11.1 8.5 50.9
Infant mortality 2008 4.8 1.1 0.4 0.9 0.2 2.3
Under-five mortality 2007 31.0 7.3 3.1 7.2 1.5 14.1
No birth attendance 2003 1723.7 361.4 102.9 241.6 44.5 987.0
HIV positive 2007 56.2 7.1 6.6 10.7 14.6 15.8
Without water 2006 908.4 189.5 62.6 175.4 42.5 439.7
Without sanitation 2006 2594.6 393.9 106.5 265.2 130.2 1703.2
The relationship between conflict, fragility, and the failure to attain development goals is il-
lustrated in Figure 3. In the Undernourishment map (on the left), countries that have rates of
undernourishment above the average for the developing countries are highlighted – these countries
can be said to be the main contributors to undernourishment in the world. The Infant Mortality
map (on the right) shows the main contributors to infant mortality according to the same definition.
The relationship to conflict and fragility appears from comparing these maps to Figure 2 – as for
conflict and fragility, the main contributors are found in Sub-Saharan Africa and in a belt of Asian
countries from Pakistan to Papua New Guinea.
Figure 3: Contributors to MDG deficit: Undernourishment and infant mortality rates
Legend
Non-contributors
Contributors
Missing
Undernourishment
Legend
Non-contributors
Contributors
Missing
Infant Mortality
In Figure 4, we aggregate over all 14 indicators the information shown for the two indicators
in Figure 3. This map shows the proportion of the 14 indicators for which the country has a
20
WDR Background Paper October 26, 2010
worse than the average country. If a country is missing data for one or more of the indicators, the
maps shows the proportion of indicators with data for which the country performs worse than the
average. DRC, for instance, is worse than the average for all indicators for which we have data for
that country.
Figure 4: Contributors to MDG deficit: All MDG indicators
The Gap Map
Legend
0.00 - 0.10
0.11 - 0.24
0.25 - 0.49
0.50 - 0.74
0.75 - 0.89
0.90 - 1.00
Industrialized
Although the overlap with the map of conflict and fragile states is not perfect, there are clear
resemblances. Poor development outcomes are most common in Sub-Saharan Africa and in an
Asian belt ranging from Iraq in the west to Papua New Guinea in the east.
3.2 Is the gap caused by conflict and fragility?
There are a number of good reasons to think that conflicts do have a causal, detrimental effect
on these indicators. Ghobarah, Huth and Russett (2003, 191–192) suggest a useful theoretical
framework for analyzing the effect of conflict and fragility on the development outcomes summarized
in the MDGs. Noting that “health conditions are shaped by the interplay of exposure to conditions
21
WDR Background Paper October 26, 2010
that create varying risks of death and disease for different groups in society and the ability of
groups in society to gain access to health care and receive the full range of benefits produced by
the health-care system”, they first list four sources of differences in health outcomes:
1. The extent to which populations are exposed to conditions that increase the risk of death,
disease, and disability
2. The financial and human resources available to address the public health needs of populations
3. The level of resources actually allocated to public health needs by the private and public
sectors
4. The degree to which resources actually allocated to public health are efficiently used
The first source mainly affects the health-related MDGs (MDG 1, 4, 5, and 6), whereas the
other three are equally relevant to the other outcomes we have analyzed.
Civil wars directly expose populations to conditions that increase mortality and disability, source
1 above. The most obvious is, of course, battle deaths. Mortality increases and life expectancy
decreases through deaths incurred as a direct consequence of fighting. The effects of this mechanism
on aggregate life expectancy and mortality levels depend on the technology of war used by the
warring parties. Conflict characterized by low-scale guerilla warfare will produce many fewer battle
deaths than conflict in which artillery shelling and aerial bombardment is used.
By and large, however, the indirect effects of conflict are likely to be much greater than the
direct effects. This is not only true for intra-state wars. “For instance the influenza epidemic that
spread in 1918 and 1919 killed more people than the deaths [that] resulted directly from military
activity in World War I (...) some of the causes of the magnitude that epidemic reached included
the mass movement of armed forces, the conditions in which soldiers lived in the trenches, and the
effects of mustard gas and fumes generated by some weapons” (Iqbal 2010, 40). Civil wars also
often displace large populations, and the temporary accommodations often expose them to new risk
factors. As noted by Ghobarah, Huth and Russett (2003, 192), “epidemic diseases – tuberculosis,
measles, pneumonia, cholera, typhoid, paratyphoid, and dysentery – are likely to emerge from
22
WDR Background Paper October 26, 2010
crowding, bad water, and poor sanitation in camps, while malnutrition and stress compromise
people’s immune systems”.
Even without displacement, conflict can destroy pre-existing local health facilities, as well as
block access to proximate facilities because of the risks involved in traveling through conflict zones.
This is particularly true for infant and under-five mortality and for birth-related maternal mortality.
Epidemiological research argues that disease, especially diarrhea, has a greater effect on mortality
rates than direct battle deaths. Degomme and Guha-Sapir (2010, 297) study Darfur and argue that
“more than 80% of excess deaths were not a result of [the] violence. (...) but the main cause of
mortality during the stabilization period were diseases such as diarrhoea”. Such excess deaths result
from an increased spread of disease, which drives up infant mortality rates. The increased spread
of disease may be caused by the inability of states to provide health services for their population
during wartime, or to conditions, such as in refugee camps, that increase the transmission of disease.
Ghobarah, Huth and Russett (2003, 192) further note that other forms of violence often escalate
in the aftermath of war, adding to the mortality and disability rates. Civil wars also affect the
second and third sources. Widespread violence and physical destruction disrupts transportation,
cutting rural populations off from health and education facilities. Civil war leads to diversion
and dissaving effects (Collier 1999): Military expenditures invariably increase during war, reducing
funds available to promote public health, education, poverty alleviation, and so on. (Gleditsch
1996; Knight, Loayza and Villanueva 1996). Local economies may be disrupted, partly because of
disincentives to invest at all, partly because of capital flight (Collier 1999); the effect is a reduction
of public spending. Local effects can be much more severe than national effects: When distinct
population groups are perceived as the opposition, the government will often be tempted to cut
off public spending in their territory at the same time as the military contest is likely to be most
intense in the opposition’s home region.
State fragility is, by our definition, partly associated with the same point. Countries receive
low CPIA scores if they are weak in terms of inter alia economic management; policies for social
inclusion and equity; and public sector management. Such weakness is likely to be manifested in
inadequate investment in public health infrastructure.
23
WDR Background Paper October 26, 2010
Finally, conflict reduces the efficiency of the public health resources that are allocated. Again
in the words of Ghobarah, Huth and Russett (2003, 193), “ Wartime destruction and disruption of
the transportation infrastructure (roads, bridges, railroad systems; communications and electricity)
weakens the ability to distribute clean water, food, medicine, and relief supplies, both to refugees
and to others who stay in place.” Medical personnel tend to leave conflict zones if they can, leaving
the poorest and most immobile behind. Ghobarah et al. note that military forces often deliberately
target health facilities and transportation infrastructure to weaken the opposition.
Fragility, defined as weakness in institutions, is obviously also associated with low efficiency in
the use of public health expenditures.
To what extent can these mechanisms be traced empirically?
3.3 Summary of results from our statistical analysis
Table 8 lists the MDGs and the various indicators we analyze to gauge the causal effect of conflict
on goal attainment. The detailed results are discussed in Section 4 and reported in Appendix A.
Two columns in the table summarize the effect of conflict – as estimated in cross-sectional analyses
(column 4) and in fixed-effects regression models (column 5). The two right-most columns (Effect
of Fragility) report the same for the fragility variable.
Table 8: Summary of Regression Results, Millenium Development Goals
MDG Label Indicator Effect of Effect of
conflict fragility
Cross-section Fixed-effects Cross-section Fixed-effects
MDG 1 Ending Poverty Undernourishment Detrimental Detrimental Detrimental Unclear
MDG 1 and Hunger Poverty Headcount Detrimental Detrimental Detrimental No effect
MDG 1 Life expectancy Detrimental Detrimental Detrimental Detrimental
MDG 1 GDP per capita Detrimental Detrimental Detrimental Detrimental∗
MDG 2 Universal Prim. Sch. Enrollment Detrimental Detrimental Detrimental Beneficial?
MDG 2 Education Sec. Sch. Attainment Detrimental Unclear Detrimental Unclear
MDG 3 Gender Parity Primary School ratio Detrimental Beneficial? Detrimental No effect
MDG 3 Life expect. ratio No effect Unclear No effect No effect
MDG 4 Child Mortality Infant Mortality Detrimental Detrimental Detrimental Detrimental
MDG 4 Under-5 Mortality Detrimental Detrimental Detrimental Detrimental
MDG 5 Maternal Mort. Birth Attendance No effect Unclear Detrimental Beneficial?
MDG 6 Combat AIDS % HIV positive Beneficial? Beneficial? Beneficial? No effect
MDG 7 Environmental Access to Water Detrimental Detrimental Detrimental Detrimental
MDG 7 Sustainability Access to Sanitation No effect Unclear No effect No effect
*:Estimated on growth in GDP per capita using OLS with panel-corrected standard errors.
24
WDR Background Paper October 26, 2010
As the table indicates, we find clear detrimental effects of conflict on the reduction of poverty
and hunger, on primary education, on the reduction of child mortality, and on access to water. In
the cross-section analyses we also find a detrimental effect on secondary education and on gender
parity measured in the female-to-male primary education attainment ratio.
We find clear detrimental effects of conflict on the reduction of poverty and hunger, on primary
education, on the reduction of child mortality, and on access to water. In the cross-sectional
analyses we also find a detrimental effect on secondary education and on gender parity measured
in the female-to-male primary education attainment ratio. As we will discuss, these effects are
quite strong. For example, five years of sustained conflict with only a moderate amount of direct
fatalities, on average, push 3%–4% of the population into undernourishment. We also find that
conflicts generate surplus infant mortality at the same level as the direct deaths – for every soldier
killed in battle, one infant dies that would otherwise have survived the indirect effects of conflict.
We find very limited evidence that conflict affects gender parity, measured as the female-to-male
life expectancy ratio. Internal conflicts seem to harm males and females in equal measures. We
find no effect of conflict on access to sanitation facilities. Finally, our empirical analysis indicates,
if anything, that conflicts tend to limit the spread of HIV/AIDS, adding to the inconclusiveness
of other studies (for example, Spiegel et al. 2007). This is possibly because of the relative short
time horizon in our study. In the short term, populations may be less mobile and interact less with
each other, limiting contamination (and certainly diagnosis). We do not assess the long-term nor
the spatial effects, however. The HIV virus has a long incubation period, and conflict may delay
reporting or detection of HIV positive individuals. A recent study (Iqbal and Zorn 2010) finds clear
detrimental effects of conflict on HIV/AIDS prevalence.
The effects of fragility are somewhat less clear. The cross-sectional analyses indicate a detri-
mental effect on most of the MDG indicators, except for female-to-male life expectancy ratio,
HIV/AIDS prevalence, and access to sanitation. In the more conservative fixed-effects models,
however, fragility has a clear detrimental effect only on life expectancy, growth, infant and under-
five mortalities, and access to safe water. We have also estimated the relationship between the
CPIA score and the development outcomes in a set of fixed-effects models. These analyses yield
25
WDR Background Paper October 26, 2010
clear results only for one indicator – good governance, as measured by the World Bank, increases life
expectancy. For other indicators, results are inconclusive. The inconclusiveness of the CPIA com-
ponent of the fragile-state indicator suggests that the conflict component of the fragility indicator
is more important than the governance component.
Section 4 also presents a number of trend graphs, some of which indicate beneficial effects of
conflict. This probably reflects conflict countries and fragile states being among the most underde-
veloped in 1990. Together with others of the poorest countries, they have partly closed the gap to
the more developed countries. The cross-sectional analyses show, however, that conflict countries
improve at a slower pace than other countries that started at the same level in 1990.
Because of having fairly short time series and available data only for every five years, we have
not been able to assess the extent to which the detrimental effects linger on after the conflict
or fragility period. This has been possible for GDP per capita, however, and given the strong
correlation between economic production and most of the MDG outcomes, this should indicate the
extent to which the effect of conflict becomes permanent.
Figure 5: Simulated change in GDP per capita 1970–2000, for conflict and non-conflict country,
long war (1974–86) and short war (1974–78)
Figure 5 shows simulated GDP per capita for the 1970–2000 period for a country that started
out at 1,100 dollars per capita, about the level of Algeria in 1970. The dotted lines in the two
sub-figures show the average growth trajectory for non-conflict developing countries. By regression
models we have estimated the difference in growth rates from this average for countries that were
at conflict at t, t−1, t−2, and so on. The results from the regression model are given in Table A-31.
26
WDR Background Paper October 26, 2010
The dashed lines indicate the growth trajectory for the same countries if they experience conflict.
The left panel shows expected GDP per capita for a country with war (more than 1,000 battle
deaths per year) that broke out in 1974 and lasted for five years until 1978, with peace thereafter.
This figure shows that the growth loss over the first five years of the conflict is very large – about
20% relative to the non-conflict country. The estimates indicate that countries see an immediate
pick-up in growth after conflicts of this duration. The right panel simulates a country that had an
outbreak of war in 1974 that lasted for 13 years up to 1986. After 10 years of conflict, there are
some signs of conflict countries recovering parts of the war losses, and this continues in the first
five years of the post-conflict period. Five years after the conflict ended, we cannot discern further
pick-up growth in either of the scenarios. The aggregate pick-up growth up to then is, on average,
not sufficient to close the gap caused by the conflict. The median-conflict country is almost 10%
under the trajectory it would have followed without the conflict. There are some uncertainties in
these estimates – the probability that the conflict country closes the gap to the non-conflict country
is larger than 10%. But the probability that the aggregate growth loss is as large as 20% is also
larger than 10%.
In a series of analyses not shown here, we have attempted to estimate the time it takes for
countries to recover from the effects of civil war or fragility. Figure 5 shows results for GDP per
capita, an indicator for which we have annual data and long time series. Our analysis shows that,
on average, countries recover partly over a five-year period, but we are not able to find evidence for
much recovery beyond this point.8For other indicators we have not been able to estimate recovery
in a similar fashion, partly because we only have data for five-year periods, and, for most indicators,
data for shorter periods. This problem is compounded by the fact that variables such as education
enrollment and infant mortality rates take longer to respond to changed environments than GDP
per capita.
8This is partly because of the methodological problems involved in tracing effects over long time spans with data
for only four decades, with most post-conflict periods occurring in the last 15 years.
27
WDR Background Paper October 26, 2010
3.4 Conflict and the attainment of the Millennium Development Goals
It is evident from this analysis that internal armed conflict and state fragility hamper development.
Nourishment, infant and maternal mortality, education attainment, and other goals improve at a
considerably higher rate in countries that avoid conflict. Unfortunately, conflict is most frequent
in poor, underdeveloped countries – this is the conflict trap (Collier et al. 2003) in which conflict
creates underdevelopment that in turn, engenders conflict. If the conflict trap cannot be broken,
the world will continue to diverge between successful developers and ‘the bottom billion’ (Collier
2007).
We then turn to a more detailed discussion of the impact of conflict on the individual millenium
development goals.
4 Analysis of Individual Indicators
4.1 MDG 1: Ending Poverty and Hunger
4.1.1 Global Trends
Figure 6 shows the trends in prevalence of undernourishment during the years 1990–2008 for the
three conflict categories and the two state-fragility categories. The countries that had no conflicts
(non-conflict: the dashed line) in the period 1980–2008 have a slowly decreasing undernourishment
rate. The countries that had conflict after 1990 (conflict: the solid line) start with a much higher
rate of undernourishment in 1990 (at about 28% as compared to 12%) but decrease steadily. The
countries that had conflict during the 1980s but not after (post-conflict: -the dotted line) had
undernourishment rates at close to 20%, but this rate rapidly decreases toward the non-conflict
countries over the next 18 years.
Figure 7 shows the distribution in 2005 for the countries in each category. In addition to
showing that most non-fragile states are better nourished than the fragile ones, this shows that
the fragile group is more homogenous than the non-fragile group – a handful of states that are not
coded as fragile have about 40% of the population without proper nourishment. The difference
between the median non-fragile state and the median fragile state is considerably larger than the
28
WDR Background Paper October 26, 2010
Figure 6: Trends in percentage of population that is undernourished, by conflict type and fragility
difference between the means shown in Figure 6. This is also seen to a lesser degree in the sub-figure
comparing conflict, post-conflict and non-conflict countries.
Figure 7: The Percentage of Population Suffering from Undernourishment, By Conflict Status and
State Fragility in 2005
4.1.2 Literature on Effects of Conflict on Poverty and Hunger
Systematic studies of the effect of conflict on poverty and hunger are scarce, mostly consisting of
NGO reports. Several studies confirm the popular perception that conflicts exacerbate poverty and
hunger. Messer and Cohen (2004, 3) argue that “conflict causes food insecurity” and that civil
conflicts in Africa from the mid-1960s until 2000 cost the region more than “$120 billion worth of
agricultural production”. Country studies carried out in post-conflict countries also find a marked
increase in poverty and hunger during war. For Angola, Sapir and Gomez (2006, 13) find that
29
WDR Background Paper October 26, 2010
malnutrition rates were severely affected by conflict, but that “one year after the cease-fire, Angola
had been able to leave behind the high rates of crude mortality and malnutrition that field surveys
had recorded during conflict”. For Mozambique, Br¨uck (2006, 33) finds a more lasting effect of
conflict. In 1997, five years after the civil war ended, he reported that in the northern part of the
country “39 percent of all children under 3 years of age [were] moderately or severely underweight”.
Mozambique had a prevalence of undernourishment among the population of 52% in 1997. For
comparison, in Burkina Faso, which has similar GDP per capita but has largely avoided conflict,
undernourishment affected only 12% of the population.9According to the World Bank (2007,
17), Sub-Saharan Africa (SSA) “alone remains seriously off-track to achieve the poverty reduction
MDG”. SSA is also the region with the most civil conflict in the world.
We are not aware of any cross-national studies of conflict’s effect on undernourishment beyond
those reported here, nor of any systematic cross-national studies of the relationship between con-
flict and the poverty-headcount variables referred to in Table 8. Poverty and undernourishment,
however, are to a large extent determined by economic development broadly defined and captured
by the GDP (gross domestic product) measure (Collier and Dollar 2002). Several studies show
that civil conflicts wreak havoc on the economies of the countries they take place in. According
to Collier (1999) this happens through five processes: destruction of resources, disruption of social
order, diversion of public expenditure, dissaving, and the shifting of assets out of the country. In
Breaking the Conflict Trap, Collier et al. (2003) consequently describe civil war as development
in reverse: “After a typical civil war of seven years duration, incomes would be about 15 percent
lower than had the war not happened”. There is substantial literature on the effect of conflict on
economic factors, both directly on issues such as GDP growth, but also on the composition of a
country’s economy and on the effect on spending (for example, military expenditure).10 Here, we
focus on recent articles that deal directly with the impact of civil conflict on the economy.11
Collier (1999) argues that civil wars are “liable to be more damaging than international wars”
and finds that during “civil war the annual [GDP] growth rate is reduced by 2.2%”. These results
9Data from the World Development Indicators
10See Collier et al. (2003), Koubi (2005), and Chen, Loayza and Reynal-Querol (2008) for reviews.
11Refer to these studies for additional reviews.
30
WDR Background Paper October 26, 2010
are very close to the results we present in Figure 5 (Table A-31). Collier finds a difference between
long and short wars. While short wars “cause continued post-war [GDP] decline, [...] sufficiently
long wars give rise to a phase of rapid growth” (Collier 1999, p. 175–176) – a ”Phoenix effect”
(Organski and Kugler 1980). Collier attributes the continued decline in GDP after short wars to
post-war environments being less capital-friendly than pre-war capital environments.12
Koubi (2005) studies the effect of both inter- and intranational wars on average growth in per
capita real output. She finds that a war’s severity, measured in number of battle deaths, has a
significant negative impact on growth. When she repeats the analysis for only interstate wars,
the statistical significance disappears, indicating that the “association between war and economic
growth is due to civil wars” (Koubi 2005, 76–77).
Koubi (2005, 78) also finds, in accordance with Collier (1999), that “the more severe or longer
the war, the higher the subsequent, medium-term economic growth”. Similarly, Chen, Loayza and
Reynal-Querol (2008, p. 71) find that the “average level of per capita GDP is significantly lower
after the war than before it”, and they argue that this is “undoubtedly a direct reflection of the
cost of war”. They, too, find that after “the destruction from war, recovery is achieved through
above-average growth“ but this growth follows the pattern of “an inverted U, with the strongest
results achieved in the fourth or fifth year after the onset of peace”. (Chen, Loayza and Reynal-
Querol 2008, p. 72, 79). Likewise, Blomberg, Hess and Thacker (2000) find a strong negative effect
of both external and internal conflict on growth.
In addition to the impact on the domestic economy, the effects of civil wars tend to spill
over into neighboring countries (Buhaug and Gleditsch 2008; Gleditsch and Ward 2000; Salehyan
and Gleditsch 2006). This spill-over takes the form of increased risk of civil war for countries
with neighbors in a state a civil war, but the influence can also be substantial for the neighbors’
economic growth. Murdoch and Sandler (2002, p. 96) argue that a neighboring civil war may
affect a country’s GDP through collateral damage where battles close to the border may destroy
infrastructure and capital, and by increasing “perceived risk to would-be investors and divert[ing]
foreign direct investment away from neighbors at peace”. Investigating this argument, they find
12See Davies (2008) for a detailed study of post-conflict capital flight.
31
WDR Background Paper October 26, 2010
that a civil war creates “significant negative influence on short-run growth within the country and
its neighbors” (Murdoch and Sandler 2002, p. 106–07). Murdoch and Sandler (2004) extend the
time frame and approach of their earlier paper. They argue that the spill-over effects may go
beyond a country’s immediate neighbors, “owing to regional economic integration and regional
multiplier effects”. In contrast to the relatively modest effects they found for immediate neighbors
in the earlier paper, they find that, for neighbors within a radius of 800 kilometers “a civil war at
home is associated with a decline in economic growth of 0.1648, while an additional civil war in
a neighbor is associated with a decline of approximately 0.05 or about 30% of the home-country
effect” (Murdoch and Sandler 2004, 145). This implies that “a country in a region with three or
more civil wars may be equally impacted as a country experiencing a civil war” (Murdoch and
Sandler 2004, 150).
4.1.3 Empirical Analysis
The gap table (Table 1) and Figures 6 and 7 show that conflict countries and fragile states are,
on average, more undernourished and poorer, and have shorter life expectancy and lower GDP per
capita than countries that are neither in conflict nor fragile. To what extent is this situation caused
by the conflicts and fragility themselves, rather than by factors that explain conflict, fragility,
and poor development outcomes? To answer this, we look at the cross-sectional and fixed-effects
analyses (see sections 2.4.1 and 2.4.2). The regression tables are reported in Appendix A.3.1.
Undernourishment The cross-sectional analysis (Table A-3) shows how conflict has affected the
reduction in undernourishment from 1990 to 2004. Undernourishment is measured as a percentage
of the total population.13 The combination of the exposure and ‘(firstnm undernourish)’ variables
indicates that the average improvement has been about 10% over the period for a country with
an initial proportion undernourished at about 30%. This analysis shows that countries that have
had a large number of battle deaths in internal conflicts have reduced undernourishment to a much
smaller degree. The conflict variables are statistically significant and strong both in log-transformed
13Variable name in WDI: sn itk defc zs. Improvement in this variable means a reduction in undernourishment, so
coefficients for variables that hamper improvement will have a positive sign in Table A-3.
32
WDR Background Paper October 26, 2010
and non-transformed versions. A country that has gone through a war with 10,000 battle deaths
has reduced undernourishment by 3.5 percentage points less than an otherwise similar non-conflict
country, according to these estimates. This means that, in a country with 10 million inhabitants,
the war has prevented 350,000 people from moving out of the undernourished category.14 The
estimates with the non-transformed battle-deaths variable indicate an even stronger effect, but this
may be driven by a few extremely lethal and detrimental cases. The left-most column indicates
that undernourishment rates that are increased proportionally to the number of years with conflict
reduce the improvement, but the estimates shows lots of variance around the general trend. It is
the severity of the conflict that matters, not the duration.
Column 4 shows that fragile states reduce undernourishment at a much slower pace than non-
fragile states. The ‘(sum) fsida’ variable is the count of years coded as fragile over the 1990–
2004 period. Each year of fragility is associated with an increase in undernourishment in 0.5
percentage points relative to the baseline. The relative increase allows accounting for the general
decrease in global undernourishment levels shown in Figure 6 – a fragile state is estimated to reduce
undernourishment by 2 percentage points over a ten-year period when a non-fragile country with
similar initial conditions decreases it by 7 percentage points.
The right-most column estimates the relationship between improvement in nourishment and the
average value for the CPIA index over the period. This estimate is not significant – controlling for
initial undernourishment rates, the CPIA index explains little change in this variable. This indicates
that the PKO component of the fragility index is more important than the CPIA component, and
further underscores the detrimental effect of conflict.
Table A-7 shows the results from estimating a set of fixed-effects models for this dependent
variable. To account for global changes in the average levels for the indicators, we include dummy
variables for each five-year period. In Model 1, we estimate the effect of the conflict variable.
Conflicts clearly lead to undernourishment in the following five-year period. The estimate for the
conflict variable is 0.78, implying that a single year of minor conflict increases the proportion of
14Obviously, the dichotomous classification between undernourished and not undernourished is crude. Moving out
of undernourishment does not mean an extreme improvement in a population’s diet. In countries where a large
proportion of the population has a marginal income, it may not take much to fall into the undernourished category.
33
WDR Background Paper October 26, 2010
the population that is undernourished by 0.8%. Five years of major conflict is estimated to lead to
an 8% increase in this proportion.
We have tested whether the effect of conflict is contingent on the size of the country – in a large
country, a conflict may be extremely detrimental to the region experiencing warfare, but have little
impact on the country as an entity. We do not find signs of such an interactive term, however. For
this outcome, the effect of conflict seems to be country-wide. The fixed-effects analysis indicates
that the cross-sectional analysis underestimates the effect of conflict in this case. It implies that
for a country of median size of about 10 million inhabitants, a single year of conflict (with up to
1,000 battle deaths) moves a total of 80,000 people into the undernourishment category. Five years
of major conflict affects about 800,000 persons.
In Models 2 and 3, the conflict measure is the count of battle-related deaths due to fighting in
the five years preceding the observation period. Also, this indicator indicates a strong, detrimental
effect of conflict. The estimate of 0.416 implies that a conflict of median severity (2,500 deaths
over 5 years) increases the undernourished proportion of the population by about 3.3% percentage
points.15 This corresponds to about 300,000 persons in the median country. Again, this estimate
is stronger than the corresponding estimate in the cross-sectional analysis.
Figure 8 shows that the fixed-effects model estimates are clearly defined. The line shows how
the undernourishment rate changes over each five-year period as conflict intensity increases.16 The
vertical lines represent the 95% confidence interval around this line.
In Model 4, we estimate the effect of state fragility on undernourishment. Fragile states also
have higher proportions of undernourished populations. Controlling for unobserved country-level
factors and time trends, fragile states on average have 2.4% higher proportion of undernourishment
than non-fragile states.
Figure 9 illustrates the estimated effect of conflict on undernourishment for a hypothetical
country, with a population of about 15 million in 1970, increasing to 35 million in 2005.17 The
initial undernourishment proportion for this country was about 20% in 1970. The dotted line in
15The logarithm of 2,500 is 7.82, which multiplied by the parameter estimates 0.494 is 3.3%.
16The figure is produced using Clarify (King, Tomz and Wittenberg 2000).
17This is close to the size of Tanzania.
34
WDR Background Paper October 26, 2010
Figure 8: Estimated average rate of undernourishment, by ln(battle deaths)
the figure represents the baseline scenario without any conflicts. The estimates of Model 1 in Table
A-7 imply that the poverty rate for the hypothetical country is constant from 1970 to 1995, and
thereafter slowly decreases.18
Figure 9: Simulated effect of conflict on undernourishment.
The dashed line shows estimated poverty rates if this country had five years of minor conflict
starting in 1980: The prevalence of undernourishment then increases to about 25% during those
18The specification of the model underlying Figure 9 does not allow for only a partial recovery as evident for GDP
per capita in Figure 5. For most of the indicators we use, we have data only for every five-year period. This precludes
estimating the same type of model as the one shown in Table A-31.
35
WDR Background Paper October 26, 2010
five years. The solid line shows estimated poverty rates if the country had fifteen years of major
conflict, starting in 1975: Undernourishment then increases to 28% for the entire 15-year period.
Poverty We now turn to the effect of conflict on poverty – the percentage of a country’s popu-
lation that live on less than USD1.25 per day.19 Figure 10 shows the distribution for the poverty
headcount indicator for the year 2000. Just as shown in the gap table, non-conflict and non-fragile
states have much less poverty than conflict and fragile states.
Figure 10: The Percentage of Population Suffering from Poverty, By Conflict Status and State
Fragility in 2005
Results from the cross-sectional analysis are reported in Table A-4. Improvement is measured
from 1991 to 2003. Average improvements for this indicator are very similar to those for undernour-
ishment. Here, too, the count of conflict-years variable is positive but not statistically significant.
The estimate for log battle deaths is significant, though, indicating that a typical country with
a conflict causing 10,000 battle deaths reduces poverty rates 5% less over the 12 years than a
comparable country without conflict.
Fragile states also reduce poverty at a slower pace than non-fragile states – each year of fragility
is associated with a relative increase in poverty of about 1.3 percentage points. This is partly due
to poor governance – improvement is estimated to be about 0.18 percentage points larger for each
additional point-year on the CPIA scale. In other words, over the 13-year period from 1991 to
2003, a country would have reduced poverty by 2.4 percentage points more if it had a CPIA score
19Variable name in WDI: si pov dday.
36
WDR Background Paper October 26, 2010
that was one unit higher.
The results for the fixed-effects analysis of the relationship between conflict, fragility, and
poverty are presented in Table A-8. In contrast to the cross-sectional analysis, we find little trace
of a direct effect of conflict on poverty. Estimates are largely in the expected direction, but not
statistically significant. This is partly due to data sparseness – we have three consecutive obser-
vations for only just above 50 countries and never more than 89 countries for a given year. Data
also tend to be most sparse in conflict countries - for example, we lack data for Afghanistan, DRC,
Algeria and Sudan for the year 2000.
Life expectancy Figures 11 and 12 show the distribution and trends for life expectancy. The
box plots show that fragile and conflict countries have lower average life expectancy than other
countries. The trend figures show the same differences in average life expectancies, but indicate
that conflict countries gradually catch up with non-conflict countries. This is likely because the
potential for improvement is much higher in countries with low initial life expectancies.
Figure 11: Distribution of life expectancy in 2005 by conflict and state fragility status
As for undernourishment, Figure 11 shows that a few non-conflict and non-fragile states have
low life expectancies in 2005. Conflict and fragile states, on the other hand, have consistently lower
values for this indicator.
Life expectancy is a useful measure since it is a function of a wide range of health and mortality
factors, many of which are likely to be affected by conflict. The disadvantage of this variable is
37
WDR Background Paper October 26, 2010
Figure 12: Mean life expectancy over time by conflict and state fragility status
that it is estimated on the basis of demographic data and responds slowly, even to serious shocks
to the health of a population. This may make it difficult to locate a causal effect from fragility and
conflict to changes in the variable.
Table A-5 shows the results from estimating the cross-sectional models with life expectancy as
the MDG1 indicator. For this indicator, we use data from 1967 to 2007. We account for the fact
that countries with low initial life expectancy improve more quickly by adding the initial value as
a control variable. Given this, it is clear that countries with conflict improve much more slowly
than non-conflict countries – ten years of war or a total of 10,000 battle deaths is associated with
a reduction in life expectancy of more than two years relative to the non-conflict counterfactual.
For a median-sized country with 10 million inhabitants, the estimates indicate a loss of about 20
million life years. Only about 500,000 of these life years can be attributed to direct battle-related
deaths.20
Fragile states also have less improvement in life expectancy than non-fragile states – each year
of fragility is associated with a reduction in life expectancy of about 0.2 years. The CPIA indicator
is also significant and has the expected sign. Over the 30 years with CPIA data, a country with
CPIA scores systematically one unit higher than the average developing country, which is estimated
to increase life expectancy by 0.6 years. Table A-5 shows that conflict and fragility variables also
appear as reducing life expectancy in the conservative fixed-effects models. The estimates are
20If the average age of soldiers killed is 20 years and we expect them to live to age 70, 10,000 deaths amounts to
500,000 life years. This estimate is probably too high.
38
WDR Background Paper October 26, 2010
negative and significant in all the models we report. The magnitude of these estimates is quite
similar to those in the cross-sectional analysis.
Growth Undernourishment, poverty headcount, and life expectancy are closely related to the
economic production of a country – low-income countries are poor, and studies such as Collier and
Dollar (2002) show that economic growth on average also reduces poverty. It is therefore useful to
study how conflict and fragility affect economic growth.
Figure 13: Mean GDP per capita over time by conflict and state fragility status
Figure 13 shows the trends in average GDP per capita for countries separately for the three
conflict categories as well as for the two state-fragility categories. The figure shows clearly that non-
conflict and non-failed countries are much richer than the conflict- and failure-affected countries.
Moreover, the non-conflict and non-failed countries have seen substantial economic growth over the
period, whereas the other groups have stagnated.
Figure 14 shows the distribution in GDP per capita for the year 2005 for the various categories.
There are a few high-income conflict countries, but no high-income fragile states. Table A-6 gives
evidence that conflict and fragility hamper growth. Measured as log battle deaths, a country with
a conflict with 10,000 fatalities is estimated to lose 0.45 of log GDP per capita after the war relative
to a non-conflict country. In other words, if initial conditions were the same and the conflict country
ended up at a GDP per capita of USD1,000 after the war, the non-conflict country is likely to have
grown to about USD1,500.
Table A-10 shows the estimates from a fixed-effects model estimation of the relationship between
39
WDR Background Paper October 26, 2010
Figure 14: Distribution of GDP per capita in 2000 by conflict and state fragility status
conflict and economic growth measured as the annual increase in log GDP per capita. Using
a dataset with annual observations, we estimate an OLS model with panel-corrected standard
errors, correcting for autocorrelation.21 Conflict clearly affects economic growth in our estimations,
corroborating the studies reviewed above. One year of minor conflict is estimated to reduce annual
growth by 1% to 2%, major conflicts are assumed to have twice as high an effect. The analyses
using battle deaths as conflict variables indicate effects of the same magnitude – minor conflicts
with 25 battle-related deaths per year lead to an annual 1% growth shortfall, whereas the most
lethal conflicts with more than 100,000 deaths per five-year period cause a growth reduction of
about 4% per year.
4.2 MDG 2: Universal Education
4.2.1 Global Trends
As for the other indicators we analyze, global average education enrollment and attainment levels
have seen a steady improvement from the 1960s to today at the global level. On average, countries
have improved attainment percentages by about 2% in every five-year period. Figure 15 shows
trends in secondary education levels. The Y axis shows the proportion of the population in the
relevant age group that completes secondary education.
The group of conflict countries has attainment rates about 15% lower than the non-conflict
21We assume that the autocorrelation is similar for all panels. Estimated autocorrelation is relatively weak, at
about ˆρ= 0.35.
40
WDR Background Paper October 26, 2010
Figure 15: Trends in secondary education attainment rates 1990–2008, by conflict and fragility
group and shows no signs of closing the gap. The group of fragile states lags behind the non-fragile
states to a similar extent, and this gap seems to widen rather than to narrow. Figure 16 shows the
distribution in attainment rates within the various groups we analyze.
Figure 16: Distribution of secondary education attainment rates in 2000, by conflict and fragility
A similar pattern can be seen from figures depicting the trends and distribution in primary
education attainment rates.22
4.2.2 Literature on Effects of Conflict on Education
The effect of conflict on education is especially interesting given education’s importance for de-
velopment in general. In the 1990 edition of the World Development Report (World Bank 1990),
education is seen as one of four key components for combating, for example, poverty, and education
22The initial gap is wider between conflict and non-conflict countries, but may be narrowing. The initial gap
between fragile and non-fragile states is also wider, but this gap increases from 1990 to 2008.
41
WDR Background Paper October 26, 2010
coupled with economic growth is often seen as the way out of a conflict trap (Collier et al. 2003).
The effect of conflict on education has received more systematic attention from researchers than
poverty and hunger. Several African country studies exist, all reporting significant negative effects
of conflict. In a study on the educational cost of World War II, Ichino and Winter-Ebmer (2004)
find that children of school age during that war received less education than children in neutral
countries, and these individuals experienced a significant earnings gap 40 years later. World War
II is unique in terms of size and severity, but researchers find similar results for civil conflicts too.
On the aggregate level, Lai and Thyne (2007) find that during civil war a state “reduces its edu-
cational expenditures by 3.1% to 3.6% each year”. Perhaps more significant, the authors find that
this reduction in spending is not due to a “guns for butter” tradeoff but that civil wars disrupt a
state’s “general ability to provide social services like education to its citizenry”. Lai & Thyne find
a similar effect of conflict on education enrollment. This is perhaps more disturbing because such
an effect is likely to linger on long after the conflict has ended. In a report on Afghanistan, Human
Rights Watch (2007, 74) document several direct attacks on schools, and argue that schools might
become the target of attacks because they are seen as “symbols of the government or the work of
foreigners”. However, such an argument has not, to the best of our knowledge, been made for other
conflicts.
4.2.3 Empirical Analysis
Table A-11 shows the results from estimating a cross-sectional model of improvements in primary-
school education attainment. Countries that have had conflict have increased school attainment
more slowly than similar countries that have avoided war. Our standard example, a war with 10,000
battle deaths, is associated with a relative decrease in attainment of about 7.5 percentage points.
Fragile states also improve education levels considerably more slowly than non-fragile states. Over
a ten-year period, the estimates indicate that a non-fragile state improves education attainment by
8.7 percentage points more than an initially similar fragile state. Table A-16 indicates that similar
patterns hold for secondary education.
Table A-13 shows the results from estimating fixed-effects regression models of the effect of
42
WDR Background Paper October 26, 2010
conflict on primary education enrollment as a percentage of the relevant age group.23 The first
columns present results for models with our various conflict variables. The estimates indicate that
conflicts adversely affect enrollment rates, but are not statistically significant. Figure 17 shows
how estimated primary school enrollment rates change with conflict in the country. Enrollment
rates decrease with conflict, but the wide confidence band shows that the result is not unclear
statistically.
Figure 17: Estimated average enrollment in primary education, by ln(battle deaths)
The state fragility index seems to be associated with primary-school enrollment rates, but in
the opposite direction of what one would expect. The CPIA index is not significant.
Table A-14 shows the same set of models for male secondary-school attainment rates, measured
as the percentage of the relevant age group. Again, there is no discernible effect of conflict on
education levels in the country – none of the estimates are statistically significant.
Conflicts in the neighborhood do seem to hurt secondary education. Countries that have a neigh-
bor that had five years of minor conflict in the preceding period experience an average reduction
in education attainment of 1.3%. This roughly corresponds to losing 3 to 4 years of development
relative to similar countries located in peaceful neighborhoods.
23Variable name in WB dataset: se prm nenr. Source: WDI.
43
WDR Background Paper October 26, 2010
4.3 MDG 3: Gender Parity
4.3.1 Global Trends
Figure 18 shows trends in the ratio of female to male primary school enrollment levels from 1990
to 2007. The group of conflict countries started as considerably less favorable to female education
than non-conflict countries, but the gap in terms of percentage points difference between girls and
boys is decreasing. This probably overestimates the extent to which this gap is closing, however:
100% is full gender parity, so it takes more for a group to increase this measure by one percentage
point the closer it comes toward this limit. The trends for fragile versus non-fragile states show a
similar gap, with no evidence that it is closing.
Figure 18: Trends in female-to-male primary education enrollment ratio 1990–2008, by conflict and
fragility
Figure 19 shows the distribution for this variable in 2000 for the various groups.
Figure 20 shows trends in the ratio of female to male life expectancy. The global average ratio
in 2000 was about 1.06 (see Figure 21) – women live longer than men for physiological reasons. The
conflict group has a slightly higher ratio than the non-conflict group, whereas the fragile state group
has a lower ratio than the non-fragile states. This may be because fragility is a more important
determinant of this measure – recall that non-conflict fragile states are coded as not in conflict in
all the figures of this type.24 Moreover, it is likely that conflicts are also detrimental to male life
expectancy – after all, most soldiers are male, and this is probably also the case for conflict-related
24The categorization is different from that in Table 1.
44
WDR Background Paper October 26, 2010
Figure 19: Distribution of female-to-male primary education enrollment ratio in 2000, by conflict
and fragility
homicides.
Figure 20: Trends in female-to-male life expectancy ratio 1990–2007, by conflict and fragility
Figure 21 shows the distribution for this indicator for the year 2000.
4.3.2 Literature on Effects of Conflict on Gender Equality
With the adoption by the United Nations Security Council of Resolution 1325 “Women, Peace and
Security” there has been a growing interest in gender and war. There is a burgeoning literature
on gender and conflict, but mainly with conflict as the dependent variable. Caprioli (2000) and
Caprioli and Boyer (2001) study various forms of interstate crisis and find that having more gender
equality at the domestic level leads to states behaving more peacefully in international relations.
Similarly, Melander (2005a,b) finds gender equality to be negatively correlated with human rights
45
WDR Background Paper October 26, 2010
Figure 21: Distribution of female-to-male life expectancy ratio in 2000, by conflict and fragility
abuses as well as intrastate conflict.
Some studies find that conflicts harm women more than men, although most soldiers are men.
The International Committee of the Red Cross (2001, p. 28) argues that “women are particularly
susceptible to marginalization, poverty and the suffering engendered by armed conflict, especially
when they are already victims of discrimination in peacetime”. The United Nations Development
Fund for Women (Rehn and Sirleaf 2002, p. 6) states, “The magnitude of violence suffered by
women before, during and after conflict is overwhelming”. The most systematic study of the
impact of armed conflict on gender differences in conflict effects has been carried out by Neumayer
and Plumper (2006). They find a significant and largely robust negative effect of conflict on female
life expectancy, and conclude that “a civil war reduces [the female to male life expectancy ratio]
by 0.34 percentage points” and that this shows that “the direct and indirect consequences of wars
combined either kill more women than men or that the killed women are younger on average than
the killed men” (Neumayer and Plumper 2006, 744, 747).
4.3.3 Empirical Analysis
Table A-16 presents the results from estimating a cross-sectional regression of improvement in
the female-to-male primary-education ratio. The positive sign for the coefficient for the exposure
variable (number of years between the first and last observations) reflects the general trend toward
increased parity, and the variable labeled ‘(firstnm) femedu’ (initial value for the education gender
ratio variable) that this trend is stronger the more disadvantaged females are at the initial year of
46
WDR Background Paper October 26, 2010
the cross-section. Controlling for these factors, all our conflict and fragility indicators (except the
CPIA score) indicate a reduction in this improvement. The association is strong – 10 years of war
means a relative decrease in parity of about 3.8%, and 10 years of fragility, a relative decrease of
3.4%.
Table A-17 shows the standard set of models run with the ratio of female to male primary
school enrollment (measured as a percentage).25 The results of this analysis are contra-intuitive.
In contrast to the clear negative relationship between conflict and gender parity seen in Figures
18 and 19 and in the cross-sectional analysis, this analysis indicates a positive relationship. This
is puzzling, but may possibly be because our standard model specification is ill suited in this
particular case. Gender parity is difficult to analyze, because the trend within each country shows
a very predictable movement towards gender parity. Some countries move towards parity faster than
other countries, and this unobserved quantity is apparently highly correlated with population size.
When we add country fixed effects, the global time fixed effects are less efficient than population,
which end up being strongly positive. Compare this with the AR(1) model, where the temporal
fixed effects are strong but where population is not. Similarly, the post-conflict indicators are
also more likely to show up late in the dataset, and might therefore pick up this trend. In the
fixed-effects models, conflict appears to level out gender differences, whereas there seems to be no
effect in the AR(1) model. Neither of these models appears very trustworthy. Observing that the
lagged dependent variable accounts for 91% of the variance in gender parity, we use this variable
and population to create a set of matched observations, and based on these matches, we find a
statistically significant effect of -3.10 from the presence of any conflict in the preceding ten years on
gender parity. This latter result throws further doubt on the positive effect from the fixed-effects
models.
25Source: United Nations Educational, Scientific, and Cultural Organization (UNESCO) Institute for Statistics.
Note: the break in the series between 1997 and 1998 because of the change from International Standard Classification
of Education 76 (ISCED76) to ISCED97. Recent data are provisional.
47
WDR Background Paper October 26, 2010
4.4 MDG 4: Infant Mortality Rates
4.4.1 Global Trends
Figure 22 shows the trends in infant mortality rates for non-conflict, post-conflict, and conflict
countries as well as for fragile and non-fragile states.26 Globally, infant mortality rates have de-
creased considerably over the past decades. This is the case for all of groups in the figure – the
group of conflict countries has in fact decreased infant mortality rates by a larger degree than the
group of non-conflict countries. The gap between fragile and non-fragile states, on the other hand,
has been widening over the 18 years for which we have data.
Figure 22: Trends in average infant mortality rates 1990–2008, by conflict and fragility
Figure 23 shows the distribution of infant mortality rates across our groups.
Figure 23: Distribution of average infant mortality rates in 2000, by conflict and fragility
26Variable name in WDI dataset: ssp dyn imrt in.
48
WDR Background Paper October 26, 2010
4.4.2 Literature on Effects of Conflict on Infant Mortality
Infant mortality is highly correlated with other indicators of socio-economic development such as
GDP per capita. Effects of conflict on infant mortality are likely to be similar to those for GDP
per capita. Infant mortality rates (IMR) are often employed as a proxy for a state’s general socio-
economic development as an independent variable (Abouharb and Kimball 2007) because the data
coverage for infant mortality is good in every region of the world.
Davis and Kuritsky (2002); Ammons (1996); Stewart, Humphreys and Lea (1997) all find that
conflict increases infant mortality. For Sub-Saharan Africa, (Davis and Kuritsky 2002, 9) find
that countries that experienced conflict had average infant mortality rates 10% higher than those
without conflict experience.27 In contrast to the fixed-effects analysis we present below, Davis and
Kuritsky (2002, 8) conduct a cross-national time-series regression.
The most recent and comprehensive study on the effect of conflict on infant mortality is Iqbal
(2010). Employing econometric models similar to ours, she finds that infant mortality is increased
by conflict. Her finding, however, is not very robust or highly significant, which she attributes
to “the possibility that during protracted conflicts, populations adjust to societal conditions and
appropriately guard against infant mortality. Consequently, as the conflict perpetuates, the increase
in infant mortality caused by major conflict during the first year is addressed by societies through
capacity building and resource allocation” (Iqbal 2010, p. 88).
4.4.3 Empirical Analysis
Table A-19 presents the results from the cross-sectional regression analyses. The dependent variable
is the improvement from 1969 to 2006 in log infant mortality rates. Annual improvement has on
average been about −0.034, which corresponds to 3.5% annual reduction in (non-logged) infant
mortality. The estimate for the ‘(sum) war’ variable is 0.02 – every year of war implies about 2%
smaller reduction in infant mortality rates than non-war countries. Minor conflicts are not clearly
associated with different trends in infant mortality rates. The estimates for the battle deaths
variables indicate the same relationship: For example, the log battle deaths coefficient means that
27Davis and Kuritsky (2002) use conflict data from R. Sivard’s World Military and Social Expenditures.
49
WDR Background Paper October 26, 2010
countries that have had wars leading to 10,000 battle deaths on average have lost a infant mortality
rate reduction of about 25%.
State fragility also hampers infant mortality rate reduction – on average, the annual reduction is
2% lower in fragile states than in non-fragile states. Table A-20 shows similar results for under-five
mortality rates.
Table A-21 shows results from estimating a set of fixed-effects models with log infant mortality
rates as dependent variable and various operationalizations of conflict as main independent vari-
ables. The analysis indicates that conflicts have a clear detrimental effect on infant mortality rates.
We do not find the effect to be clearly contingent on the size of the country.
The results in Model 1 indicate that one year of minor conflict increases log infant mortality
rates by 0.0111, or approximately 1.11%. For a country with infant mortality rates at 75 per 1,000
live births (typical of SSA in 2005), this translates into an increase to 75.9 per 1,000. The estimates
imply that a 5-year major conflict increases mortality rates to 84-a considerable change. In the
average conflict country, more than 1 million children are born every year.28 Increasing infant
mortality rates from 75 to 75.9 means a surplus mortality of 900 children per year; increasing to 84,
a surplus infant mortality of 9,000 per year. These estimates imply that surplus infant mortality
rates exceed the number of direct battle deaths in the conflict. In Africa South of Sahara, average
infant mortality rates were reduced from about 94 in 1985 to 75 in 2005. The effect of five years
of civil war, then, typically sets a country back 10 years relative to African countries that avoided
conflict.
Model 3 estimates the relationship between infant mortality rates and the battle deaths measure.
Indeed, figure 24 shows that infant mortality increases strongly in conflict. The relationship is
statistically significant as reflected in the narrow confidence bands for the estimates. Table A-22
shows very similar results from a fixed-effects analysis of under-five mortality rates and conflict.
Model 4 estimates the relationship between fragile state status and infant mortality. The
relationship is very clear and of the same magnitude as a major conflict. Fragile states have
infant mortality rates roughly 10% higher than comparable countries according to these estimates.
28The 23 conflict countries listed in A-1 have an average population of 57 million in 2008, and an average population
in the 0-4 years age group of about 6 million, according to United Nations (2007).
50
WDR Background Paper October 26, 2010
Figure 24: Estimated average infant mortality rates, by ln(battle deaths)
In contrast to Iqbal (2010), we find the effect of conflict on IMR to be strong and robust. The
finding holds for different specifications of conflict, and the corresponding results for under-5 child
mortality rates are almost similar (Table A-22).
4.5 MDG 5: Maternal Mortality/Birth Attendance
4.5.1 Global Trends
Data for maternal mortality are too sparse to allow the type of statistical study we use here, so we
have to resort to a proxy. Other studies show that maternal mortality can be effectively reduced
when births take place in the presence of skilled medical personnel. Figure 25 shows the trends
in the percentage of births attended by skilled medical personnel for each of our groups.29 As for
many other indicators analyzed in this paper, improvement in the conflict group is larger than in
the non-conflict group, the post-conflict group improves more strongly than any other groups, and
the fragile states group performs worst of all groups. The gap between the conflict and non-conflict
group is still very large – about 60% compared to about 90% in 2008.
Figure 26 shows the distribution of birth attendance rates across our groups.
29Variable name in WDI dataset: This variable is from the WHO not the WDI dataset!!!.
51
WDR Background Paper October 26, 2010
Figure 25: Trends in average percentage of births attended by skilled medical personnel 1990–2008,
by conflict and fragility
Figure 26: Distribution of average percentage of births attended by skilled medical personnel in
2000, by conflict and fragility
4.5.2 Literature on Effects of Conflict on Maternal Mortality
The literature on the effect of conflict on maternal mortality is scarce, and firm conclusions can
not be drawn from it. Maternal mortality rates, however, are highly correlated with other health
indicators such as infant mortality rates, and a number of studies do find that conflict adversely
affects maternal mortality rates. O’Hare and Southall (2007, 564) find in a cross-sectional anal-
ysis of Sub-Saharan African countries that “the median adjusted maternal mortality in countries
with recent conflict was 1,000/100,000 births versus 690/100,000 births in countries without recent
conflict”.30 Hill et al. (2007) do not study the effect of conflict in particular, but they find that
50% of all global maternal deaths occurred in Sub-Saharan Africa, a region with relatively high
30O’Hare and Southall (2007) use conflict data from the Institute of Development Studies.
52
WDR Background Paper October 26, 2010
prevalence of conflict. As do most of the articles reviewed in this paper, the study may suffer from
omitted variable bias. And other studies report divergent findings. In a case study of the city,
Beira, in Mozambique, Cutts et al. (1996) do not find a strong connection between conflict and
maternal mortality. Murray et al. (2002, 347) do not look explicitly at maternal mortality, but
do find that war increases mortality in general. They argue, however, that “considerably more
research is needed on this question before the global results on the indirect effects of conflict on
mortality can be assessed”.
Ormhaug and Rustad (2010) uses DHS data for 21 sub-Saharan African countries to show
that fewer women received an adequate number (four or more) of antenatal visits and delivery by
skilled health personnel in post-conflict countries than in non-conflict countries. The difference was
discernible up to 18 years after civil war.
4.5.3 Empirical Analysis
Table shows results from a cross-sectional regression analysis of conflict and birth attendance. The
analysis suggests that conflict hinders skilled medical personnel from attending births, but the
results are statistically weak. This may be because of sparse data – we have data for a proper
set of countries for only seven years, from 1997 to 2004. On the other hand, the analysis of the
fragile state indicator variable shows a strong relationship despite the sparse data, as one would
expect from Figure 25. On average, state fragility reduces improvement on this indicator by about
0.5% annually – given an average estimated annual improvement of 0.6%, this means they hardly
improve at all.
Table A-24 shows the results from a set of fixed-effects models to estimate the relationship
between civil conflict and the proportion of births attended by skilled medical personnel. The
data are very sparse, covering only about half of low-income countries in 2000, and considerably
fewer countries in 2005 and 1995. Data before 1995 are not available. Given data sparsity, the
fixed-effects models indicate no relationship between conflict and birth attendance.
53
WDR Background Paper October 26, 2010
4.6 MDG 6: Combat HIV/AIDS
4.6.1 Global Trends
Figure 27 shows the trends in prevalence of HIV/AIDS 1990–2008. Prevalence is measured as the
percentage of the population in the 15–49 age group that are HIV positive.31
Figure 27: Trends in average prevalence of HIV/AIDS 1990–2008, by conflict and fragility
Figure 28 shows the distribution. Note the large number of extreme observations in all cate-
gories. Neither of these figures indicate a clear relationship between conflict and HIV prevalence.
If anything, conflict countries have a lower prevalence of HIV/AIDS than non-conflict countries.
Figure 28: Distribution of prevalence of HIV/AIDS in 2000, by conflict and fragility
31Variable name in WDI dataset: sh dyn aids zs.
54
WDR Background Paper October 26, 2010
4.6.2 Literature on Effects of Conflict
Conflict and HIV/AIDS prevalence relate to each other in a complicated fashion. In contrast to
the other MDG goals, some of the otherwise detrimental effects of conflict may possibly prevent
the spread of this disease, which is the counter-intuitive finding of several medical studies. Using
UCDP/PRIO Conflict data to create a count of the number of years a country is in conflict,
Strand et al. (2007) find a marked negative effect of conflict on HIV/AIDS that they attribute
to “constraints in population mobility and normal civil interactions” and possibly also to the
“disruption of medical services (for example, vaccination campaigns), where contaminated needles
have been reported to account for an important part of HIV transmission during peace time”
(Strand et al. 2007, 470). Similarly, in a study of seven Sub-Saharan African countries, Spiegel
et al. (2007) do not find any evidence for the claim that conflict increases HIV prevalence. On
the other hand, violence often causes large-scale migration (Moore and Shellman 2004; Davenport,
Moore and Poe 2003), and refugees and refugee camps can greatly facilitate the transmission of
HIV/AIDS, as well as other infectious diseases. War can also facilitate the spread of HIV/AIDS
through an increase in sexual violence often seen during conflicts. Elbe (2002) investigates how
HIV/AIDS has changed the way wars are fought, and argues that the epidemic has had an impact on
the “nature and conduct of armed conflict in Africa”; HIV/AIDS has influenced three components
of armed conflicts in Africa: their combatants, “by diminishing the operational efficiency of many
of Africa’s armed forces” (163); how the conflicts are conducted, by providing armed forces with
“a novel psychological and biological weapon of war” (167); and their social significance, “by
significantly increasing the number of eventual war-related casualties” (171). Iqbal and Zorn (2010,
149) study violent conflict and the spread of HIV/AIDS in Africa. In contrast to the weak and often
contradictory findings in earlier studies, they conclude that there is “a clear positive relationship
between both international and domestic conflict and climbing HIV/AIDS prevalence”.
4.6.3 Empirical Analysis
Our cross-sectional analyses (Table A-25) indicate conclusions in the same direction as Figure 27
– the estimates for all our conflict and fragility measures are negative and statistically significant.
55
WDR Background Paper October 26, 2010
The fixed-effects regression analysis is less clear (Table A-26). Model 1 implies that small coun-
tries with conflict have a somewhat higher HIV prevalence than small non-conflict countries. In
large countries, however, the estimates indicate that conflicts reduce HIV prevalence. Conflict in
neighboring countries, moreover, seems to reduce HIV prevalence.
4.7 MDG 7: Environmental Sustainability
4.7.1 Global Trends
We study two proxies for the goal of environmental sustainability. The first is the percentage of the
population that has access to adequate sanitation.32 The second is the percentage of the population
with access to water.33
Figure 29: Trends in the percentage of population with at least adequate access to excreta disposal
facilities, 1990–2006, by conflict and fragility
Figure 29 shows the trends in the percentage of the population with access to sanitation for each
of our groups. The patterns for this outcome variable resemble most other indicators – there is a
large gap at about 20 percentage points between conflict countries and fragile states on the one hand
and non-conflict and non-fragile countries on the other. The gap seems to narrow slightly between
conflict and non-conflict countries, but widens between fragile and non-fragile states. Figure 30
shows the distribution in the percentage of the population with access to sanitation.
32More precisely, the percentage of the population with at least adequate access to excreta disposal facilities.
Improved facilities range from simple pit latrines to flush toilets. The variable is labeled sh sta acs in the WDI.
33The percentage of population with access to an improved water source such as household connection, public
standpipe, borehole, protected well, or rainwater collection. A person must have access to at least 20 liters a day
within one kilometer. The variable is labeled sh h2o safe zs in the WDI.
56
WDR Background Paper October 26, 2010
Figure 30: Distribution of the percentage of population with at least adequate access to excreta
disposal facilities, in 2000, by conflict and fragility
Figure 31 shows the trends in the percentage of the population with access to water. Access
to water may decline during conflict because military fighting makes it inaccessible, (for example,
when the conflict prevents people from traveling safely to water sources). This problem may end
the day a cease-fire is called, but water may continue to be inaccessible because of land-mines and
unexploded ordinance, which can take huge death tolls after the fighting has stopped. A decrease
in water accessibility can also occur through the destruction of infrastructure, especially pipes and
pumping stations. This source of inaccessibility may be easier to remove after conflicts, particularly
when countries receive ample post-conflict official development assistance.
Access to water is also closely related to the two previous mortality measures we analyzed. As
noted above diarrhea is one of the biggest killers in the wake of conflict. The spread of this disease
is closely related to the availability of adequate drinking water.
Figure 32 shows the distribution of the population with access to water. A handful of non-
conflict and some non-fragile states have very poor access to water in 2000 (for example, Ethiopia,
Chad, Equatorial Guinea, and Madagascar). Indeed, this outcome variable may be more dependent
on the physical environment than many other variables. Still, it is clear that most conflict countries
and fragile states have poorer access to water than non-conflict countries. Note the improvement
over time of the post-conflict states.
57
WDR Background Paper October 26, 2010
Figure 31: Trends in the percentage of population with access to an improved water source, 1990–
2006, by conflict and fragility
Figure 32: Distribution of the percentage of population with access to an improved water source,
in 2000, by conflict and fragility
4.7.2 Literature on Effects of Conflict
The literature on the effect of conflict on access to adequate water and sanitation facilities is scarce.
Although there has been much interest in water as an independent variable when explaining conflict
(Klare 2001; Homer-Dixon 1999), almost no attention has been given to the consequences of conflict
on access to water supplies. The percentage of the population that lacks access to adequate water
and sanitation facilities has declined in every region of the world, but the shortfall between the
MDG target and what has actually been achieved is greatest in the region with the most conflict,
Sub-Saharan Africa (World Bank 2007). According to the same report, less than 20% of less-
developed countries are on track to reach the goal for access to water, and less than 35%, the goal
in access to sanitation. Similar findings are reported by the United Nations (2009, 45–46).
58
WDR Background Paper October 26, 2010
4.7.3 Empirical Analysis
Our cross-sectional analysis on the relationship between civil conflict and our environmental vari-
ables is reported in Tables A-27 and A-28. The analysis shows that conflict clearly affects water
accessibility – on average, one year of war is estimated to remove safe access from 0.5% of the
population, compared to the baseline. In contrast to many other indicators, the detrimental effect
of conflict is stronger than the global average annual improvement in access to water. The cross-
sectional analysis of conflict and access to sanitation indicates no clear relationship between the
two variables.
Results from fixed-effects models are shown in Tables A-29 and A-30. These indicate no clear
relationship between conflict and the environmental variables.
59
WDR Background Paper October 26, 2010
References
Abadie, Alberto and Guido W. Imbens. 2006. “Large Sample Properties of Matching Estimators for Average
Treatment Effects.” Econometrica 74:235–267.
Abouharb, M. Rodwan and Anessa L Kimball. 2007. “A New Dataset on Infant Mortality Rates, 1816-2002.”
Journal of Peace Research 44(6):743–754.
Ammons, Lila. 1996. “Consequences of War on African Countries’ Social and Economic Decelopment.”
African Studies Review 1:67 – 82.
Blackwell, Matthew, Stefano Iacus, Gary King and Giuseppe Porro. 2009. “cem: Coarsened exact matching
in Stata.” The Stata Journal 4:524–546.
Blomberg, S. Brock, Gregory D. Hess and Sidharth Thacker. 2000. Is There Evidence of a Poverty-Conflict
Trap? In World Bank (DECRG) - Center of International Studies Workshop on ’The Economics of Civil
Wars’. Princeton, NJ: .
Br¨uck, Tilman. 2006. “War and Reconstruction in Northern Mozambique.” Economics of Peace and Security
Journal 1(1):29–36.
Buhaug, Halvard and Jan K. Rød. 2006. “Local Determinants of African Civil Wars, 1970–2001.” Political
Geography 25(3):315–335.
Buhaug, Halvard and Kristian Skrede Gleditsch. 2008. “Contagion or Confusion? Why Conflicts Cluster in
Space.” International Studies Quarterly 52:215–233.
Buhaug, Halvard and Scott Gates. 2002. “The Geography of Civil War.” Journal of Peace Research
39(4):417–433.
Caprioli, Mary. 2000. “Gendered Conflict.” Journal of Peace Research 37:51–68.
Caprioli, Mary and Mark A Boyer. 2001. “Gender, Violence, and International Crisis.” Journal of Conflict
Resolution 45:503 —-518.
Chen, Siyan, Norman V. Loayza and Marta Reynal-Querol. 2008. “The Aftermath of Civil War.” The World
Bank Economic Review 22(1):63–85.
Collier, Paul. 1999. “On the Economic Consequences of Civil War.” Oxford Economic Papers-New Series
51(1):168–183.
Collier, Paul. 2007. The Bottom Billion. Why the Poorest Countries are Failing and What Can Be Done
About It. Oxford: Oxford University Press.
Collier, Paul and David Dollar. 2002. Globalization, Growth, and Poverty. Oxford: Oxford University Press.
Collier, Paul, Lani Elliot, H˚avard Hegre, Anke Hoeffler, Marta Reynal-Querol and Nicholas Sambanis. 2003.
Breaking the Conflict Trap. Civil War and Development Policy. Oxford: Oxford University Press.
Cutts, F. T., C. Dos Santos, A. Novoa, P. David, G. Macassa and A. C. Soares. 1996. “Child and Maternal
Mortality during a Period of Conflict in Beira City, Mozambique.” International Journal of Epidemiology
25:349 – 356.
Davenport, Christian, Will H Moore and Steven Poe. 2003. “Sometimes You Just Have to Leave: Domestic
Threats and Forced Migration, 1964 - 1989.” International Interactions 29:27 – 55.
Davies, Victor A. B. 2008. “Postwar capital flight and inflation.” Journal of Peace Research 45(4):519–537.
60
WDR Background Paper October 26, 2010
Davis, David R. and Joel N. Kuritsky. 2002. Violent Conflict and Its Impact on Health Indicators in Sub-
Saharan Africa, 1980 to 1997. In Paper presented to the Annual Meeting of the International Studies
Association, New Orleans, LA, March.
Degomme, Olivier and Debarati Guha-Sapir. 2010. “Patterns of Mortality rates in Darfur Conflict.” The
Lancet 375:294–300.
Elbe, Stefan. 2002. “HIV/AIDS and the Changing Landscape of War in Africa.” International Security
27:159–177.
Fearon, James D. 2003. Presentation. In Workshop on Conflict Data. Oslo: .
Ghobarah, Hazem Adam, Paul K. Huth and Bruce M. Russett. 2003. “Civil Wars Kill and Maim People–Long
after the Shooting Stops.” American Political Science Review 97(2):189–202.
Gleditsch, Kristian S. and Michael D. Ward. 2000. “War and Peace in Space and Time: The Role of
Democratization.” International Studies Quarterly 44(1):1–29.
Gleditsch, Kristian Skrede. 1996. Aspects of Democratization: Economic Development, Spatial Autocor-
relation, and Persistence in Time. In Norwegian National Political Science Conference. Geilo: January
8–9.
Gleditsch, Nils Petter, Peter Wallensteen, Mikael Eriksson, Margareta Sollenberg and H˚avard Strand. 2002.
“Armed Conflict 1946–2001: A New Dataset.” Journal of Peace Research 39(5):615–637.
Harbom, Lotta and Peter Wallensteen. 2009. “Armed Conflicts, 1946–2008.” Journal of Peace Research
46(4):577 – 587.
Hegre, H˚avard and Nicholas Sambanis. 2006. “Sensitivity Analysis of Empirical Results on Civil War Onset.”
Journal of Conflict Resolution 50(4):508–535.
Hegre, H˚avard, Tanja Ellingsen, Scott Gates and Nils Petter Gleditsch. 2001. “Toward a Democratic Civil
Peace? Democracy, Political Change, and Civil War, 1816–1992.” American Political Science Review
95(1):33–48.
Hibbs, Douglas A. 1973. Mass Political Violence. A Cross-National Causal Analysis. New York: Wiley.
Hill, Jennifer. 2008. “Discussion of research using propensity-score matching: ‘Comments on A critical
appraisal of propensity-score matching in the medical literature between 1996 and 2003 by Peter Austin’,
Statistics in Medicine.” Statistics in Medicine 27:2055–2061.
Hill, Kenneth, Kevin Thomas, Carla AbouZahr, Neff Walker, Lale Say, Mie Inoue and Emi Suzuki. 2007.
“Estimates of maternal mortality worldwide between 1990 and 2005: an assessment of available data.”
The Lancet 370:1311 – 1319.
Ho, Daniel, Kosuke Imai, Gary King and Elizabeth Stuart. 2007. “Matching as Nonparametric Preprocessing
for Reducing Model Dependence in Parametric Causal Inference.” Political Analysis 15:199–236.
Homer-Dixon, Thomas. 1999. Environment, Scarcity, and Violence. Princeton, NJ: Princeton University
Press.
Human Rights Watch. 2007. “The Human Cost - The Consequences of Insurgent Attacks in Afghanistan.”
New York, report.
Iacus, Stefano M., Gary King and Giuseppe Porro. 2009. “cem: Software for Coarsened Exact Matching.”
Journal of Statistical Software 30.
61
WDR Background Paper October 26, 2010
Ichino, Andrea and Rudolf Winter-Ebmer. 2004. “The Long?Run Educational Cost of World War II.”
Journal of Labor Economics 22:57 – 87.
International Committee of the Red Cross. 2001. “Women Facing War.” Geneva: ICRC.
Iqbal, Zaryab. 2010. War and the Health of Nations. Stanford, CA: Stanford University Press.
Iqbal, Zaryab and Christopher Zorn. 2010. “Violent Conflict and the Spread of HIV/AIDS.” Journal of
Politics 72:149–162.
King, Gary, Michael Tomz and Jason Wittenberg. 2000. “Making the Most of Statistical Analyses: Improving
Interpretation and Presentation.” American Journal of Political Science 44(2):347–361.
Klare, Michael T. 2001. “The New Geography of Conflict.” Foreign Affairs 80(3):49–61.
Knight, M, Norman Loayza and F. Villanueva. 1996. “The Peace Dividend: Military Spending Cuts and
Economic Growth.” IMF Staff Papers 43:1–37.
Koubi, Vally. 2005. “War and Economic Performance.” Journal of Peace Research 42:67–82.
Lai, Brian and Clayton Thyne. 2007. “The Effect of Civil War on Education.” Journal of Peace Research
44:277 – 292.
Melander, Erik. 2005a. “Gender Equality and Intrastate Armed Conflict.” International Studies Quarterly
49(4):593–743.
Melander, Erik. 2005b. “Political Gender Equality and State Human Rights Abuse.” Journal of Peace
Research 42(2):149–166.
Messer, Ellen and Mark J Cohen. 2004. “Breaking the Links Between Conflict and Hunger in Africa.”.
URL: http://www.ifpri.org/sites/default/files/publications/ib26.pdf
Moore, Will H and Stephan M Shellman. 2004. “Fear of Persecution - Forces Migration, 1952 - 1995.”
Journal of Conflict Resolution 40(5):723 – 745.
Murdoch, James C. and Todd Sandler. 2002. “Economic Growth, Civil Wars and Spatial Spillovers.” Journal
of Conflict Resolution 46:91–110.
Murdoch, James C. and Todd Sandler. 2004. “Civil Wars and Economic Growth: Spatial Dispersion.”
American Journal of Political Science 48(1):138–151.
Murray, Christopher, Gary King, A. D. Lopez, N. Tomijima and E.G. Krug. 2002. “Armed Conflict as a
Public Health Problem.” British Medical Journal 324:346–349.
Neumayer, Eric and Thomas Plumper. 2006. “The Unequal Burden of War: The Effect of Armed Conflict
on the Gender Gap in Life Expectancy.” International Organization 60(00):723–754.
O’Hare, Bernadette A. M. and David P. Southall. 2007. “First Do No Harm: the Impact of Recent Armed
Conflict on Maternal and Child Health in Sub-Saharan Africa.” Journal of the Royal Society of Medicine
100:564 – 570.
Organski, A.F.K. and Jacek Kugler. 1980. The War Ledger. Chicago: University of Chicago Press.
Ormhaug, Christin M. and Siri Aas Rustad. 2010. “Impact of Civil Conflict on Maternal Health Care in
Sub-Saharan Africa.” Typescript, NorAgric.
62
WDR Background Paper October 26, 2010
Raleigh, Clionadh, H˚avard Hegre, Joakim Karlsen and Andrew Linke. 2010. “Introducing ACLED: An
Armed Conflict Location and Event Dataset.” Journal of Peace Research 47:In press.
Rehn, Elisabeth and Ellen Johnson Sirleaf. 2002. “Women, War, Peace.” Report of the United Nations
Development Fund for Women.
Ringdal, Gerd Inger, Kristen Ringdal and Albert Simkus. 2008. “War-Related Distress Among Kosovar
Albanians.” Journal of Loss and Trauma 13:59–71.
Salehyan, Idean and Kristian S. Gleditsch. 2006. “Refugees and the Spread of Civil War.” International
Organization 60(2):335–366.
Sapir, Debarati Guha and Vicente Teran Gomez. 2006. “Angola: The Human Impact of War - A data review
of field surveys between 1999-2005.” Report of the Centre for Research on the Epidemiology of Disasters
(CRED), Brussels.
Spiegel, Paul B, Anne R Bennedsen, Johanna Claass, Laurie Bruns, Njogu Patterson, Dieudonne Yiweza
and Marian Schilperoord. 2007. “Prevalence of HIV Infection in Conflict-Affected and Displaced People
in Seven Sub-Saharan African Countries: A Systematic Review.” The Lancet 369:2187–2195.
Stewart, Frances, Frank P. Humphreys and Nick Lea. 1997. “Civil Conflict in Developing Countries over
the Last Quarter of a Century: An Empirical Overview of Economic and Social Consequences.” Oxford
Development Studies 25:11–41.
Strabac, Zan and Kristen Ringdal. 2008. “Individual and Contextual Influences of War on Ethnic Prejudice
in Croatia.” The Sociological Quarterly 49:769–796.
Strand, Roland T., L. Fernandes Dias, S. Bergstr¨om and S. Andersson. 2007. “Unexpected Low Prevalence
of HIV Among Fertile Women in Luanda, Angola. Does War Prevent the Spread of HIV?” International
Journal of STD & AIDS 18:467–471.
Tukey, J. W. 1977. Exploratory Data Analysis. Reading, MA: Addison-Wesley.
United Nations. 2007. World Population Prospects. The 2006 Revision. Number 202.
United Nations. 2009. The Millennium Development Goals Report. New York: United Nations.
World Bank. 1978. World Development Report. Washington, DC: The International Bank for Reconstruction
and Development.
World Bank. 1990. World Development Report 1990: Poverty. Washington, DC: The International Bank for
Reconstruction and Development.
World Bank. 2007. Global Monitoring Report 2007. Washington, DC: The International Bank for Recon-
struction and Development.
63
WDR Background Paper October 26, 2010
A Appendix
A.1 List of countries
Table A-1: Countries included in analysis, classified by category, 2008
Conflict Fragile Post-conflict or post-fragile
Afghanistan Angola Azerbaijan
Algeria Bosnia and Herzegovina Cambodia
Burundi Cameroon Equatorial Guinea
Chad Central African Republic Ethiopia
Colombia Comoros Korea, Dem. Rep.
Congo, Dem. Rep. Congo, Rep. Lao PDR
Iraq Cote d’Ivoire Macedonia, FYR
Israel Djibouti Moldova
Liberia Eritrea Mozambique
Nepal Gambia, The Nicaragua
Pakistan Guinea Niger
Philippines Guinea-Bissau Nigeria
Russian Federation Haiti Peru
Somalia Myanmar Rwanda
Sri Lanka Papua New Guinea Senegal
Sudan Sierra Leone Serbia
Thailand Solomon Islands Uzbekistan
Turkey Tajikistan
Uganda Timor-Leste
Togo
Yemen, Rep.
Zimbabwe
Neither conflict nor fragile
Albania Fiji Namibia
Argentina Gabon
Armenia Ghana Oman
Bahamas, The Guatemala Panama
Bahrain Guyana Paraguay
Bangladesh Honduras Poland
Barbados Hungary Qatar
Belarus Jamaica Romania
Belize Jordan Saudi Arabia
Benin Kazakhstan Singapore
Bhutan Kenya Slovak Republic
Bolivia Korea, Rep. Slovenia
Botswana Kuwait South Africa
Brazil Kyrgyz Republic Suriname
Brunei Darussalam Latvia Swaziland
Bulgaria Lebanon Syrian Arab Republic
Lesotho
Burkina Faso Libya Taiwan, China
Cape Verde Lithuania Tanzania
Chile Madagascar Trinidad and Tobago
Costa Rica Malawi Tunisia
Croatia Malaysia Turkmenistan
Cuba Maldives Ukraine
Cyprus Malta United Arab Emirates
Czech Republic Mauritania Uruguay
Dominican Republic Mauritius Venezuela, RB
Ecuador Mexico Vietnam
Egypt Mongolia Zambia
El Salvador Montenegro
Estonia Morocco
India/China
India China
64
WDR Background Paper October 26, 2010
A.2 List of conflict country matches
Table A-2: Conflict country matches
Year Name of conflict country Name of matching country
1971 Cameroon Madagascar (Malagasy)
1973 Syria Zimbabwe (Rhodesia)
1978 El Salvador Nicaragua
1979 Tunisia Syria
1980 Cote d’Ivoire Tunisia
1981 Chile Malaysia
1989 Burkina Faso (Upper Volta) Chad
1990 Burkina Faso (Upper Volta) Mali
1990 Gabon Trinidad and Tobago
1990 Niger Mali
1991 Burkina Faso (Upper Volta) Burundi
1991 Malawi Burundi
1994 Malawi Burundi
1994 Rwanda Burundi
1996 Togo Central African Republic
1998 France United Kingdom
1998 Gambia Lesotho
1998 Mauritania Lesotho
2007 Burkina Faso (Upper Volta) Mali
2008 Armenia Georgia
2008 Egypt Iran (Persia)
65
WDR Background Paper October 26, 2010
A.3 Regression Results
A.3.1 MDG 1: Ending Poverty and Hunger
Cross-sectional analyses
Table A-3: Improvement in nourishment, Cross-section 1990–2004
(1) (2) (3) (4) (5)
improvement improvement improvement improvement improvement
exposure -0.493 -0.608 -0.519 -0.564 -0.376
(0.510) (0.475) (0.499) (0.487) (0.504)
(firstnm) undernourish -0.379*** -0.397*** -0.409*** -0.461*** -0.379***
(0.0520) (0.0471) (0.0524) (0.0551) (0.0495)
logged total population 0.0602 -0.354 -0.126 0.301 0.343
(0.453) (0.414) (0.438) (0.410) (0.436)
(sum) minor 0.0644
(0.233)
(sum) war 0.140
(0.349)
bd1k 0.151***
(0.0354)
lnbd 0.385**
(0.188)
(sum) fsida 0.498***
(0.152)
(sum) cpia -0.0579
(0.0350)
Constant 8.682 13.78 9.953 8.253 6.188
(9.200) (8.557) (8.936) (8.658) (8.962)
N 141 141 141 141 141
Start 1990.4 1990.4 1990.4 1990.4 1990.4
End 2004.9 2004.9 2004.9 2004.9 2004.9
r2 0.324 0.405 0.343 0.374 0.336
Standard errors in parentheses
*p < 0.10, ** p < 0.05, *** p < 0.01
Table A-4: Improvement in poverty headcount, Cross-section 1991–2003
(1) (2) (3) (4) (5)
improvement improvement improvement improvement improvement
exposure -0.511*** -0.486*** -0.555*** -0.575*** 0.214
(0.167) (0.162) (0.162) (0.151) (0.405)
(firstnm) poverty -0.398*** -0.405*** -0.408*** -0.451*** -0.414***
(0.0565) (0.0549) (0.0537) (0.0519) (0.0540)
logged total population -0.0905 0.0906 -0.186 0.581 -0.00175
(0.851) (0.856) (0.829) (0.774) (0.817)
(sum) minor 0.0327
(0.375)
(sum) war 0.548
(0.592)
bd1k 0.0174
(0.0591)
lnbd 0.558**
(0.279)
(sum) fsida 1.272***
(0.313)
(sum) cpia -0.183*
(0.0985)
Constant 11.75 10.06 10.66 5.452 9.113
(7.949) (8.010) (7.591) (7.224) (7.588)
N 114 114 114 114 114
Start 1991.7 1991.7 1991.7 1991.7 1991.7
End 2003.3 2003.3 2003.3 2003.3 2003.3
r2 0.506 0.499 0.518 0.568 0.515
Standard errors in parentheses
*p < 0.10, ** p < 0.05, *** p < 0.01
66
WDR Background Paper October 26, 2010
Table A-5: Improvement in life expectancy, 1967–2007
(1) (2) (3) (4) (5)
improvement improvement improvement improvement improvement
exposure 0.222*** 0.223*** 0.225*** 0.204*** 0.209***
(0.0513) (0.0493) (0.0513) (0.0497) (0.0523)
(firstnm) lifeexpec -0.606*** -0.621*** -0.617*** -0.681*** -0.574***
(0.0604) (0.0583) (0.0616) (0.0626) (0.0608)
logged total population -0.168 -0.112 -0.0536 -0.498* -0.415
(0.299) (0.270) (0.301) (0.269) (0.281)
(sum) minor 0.0128
(0.0571)
(sum) war -0.221***
(0.0827)
bd1k -0.0228***
(0.00540)
lnbd -0.249**
(0.107)
(sum) fsida -0.197***
(0.0487)
(sum) cpia 0.0207*
(0.0114)
Constant 43.60*** 44.42*** 44.36*** 50.90*** 43.11***
(5.253) (5.018) (5.237) (5.368) (5.260)
N 153 153 153 153 153
Start 1967.6 1967.6 1967.6 1967.6 1967.6
End 2007.2 2007.2 2007.2 2007.2 2007.2
r2 0.733 0.750 0.729 0.748 0.725
Standard errors in parentheses
*p < 0.10, ** p < 0.05, *** p < 0.01
Table A-6: Improvement in log GDP per capita, 1972–2007
(1) (2) (3) (4) (5)
improvement improvement improvement improvement improvement
exposure 0.0138** 0.0127** 0.0171*** 0.0123** 0.0107
(0.00640) (0.00619) (0.00623) (0.00553) (0.00663)
(firstnm) gdp capita -0.429*** -0.430*** -0.446*** -0.501*** -0.412***
(0.0653) (0.0652) (0.0634) (0.0594) (0.0696)
logged total population -0.0943* -0.0986** -0.0512 -0.133*** -0.110**
(0.0498) (0.0494) (0.0481) (0.0407) (0.0455)
(sum) minor -0.00727
(0.00996)
(sum) war 0.000855
(0.0188)
bd1k -0.000713
(0.00201)
lnbd -0.0505***
(0.0181)
(sum) fsida -0.0481***
(0.00830)
(sum) cpia 0.00177
(0.00232)
Constant 4.311*** 4.353*** 4.200*** 5.253*** 4.299***
(0.799) (0.796) (0.753) (0.705) (0.786)
N 148 148 148 148 148
Start 1972.1 1972.1 1972.1 1972.1 1972.1
End 2007.7 2007.7 2007.7 2007.7 2007.7
r2 0.363 0.361 0.394 0.486 0.363
Standard errors in parentheses
*p < 0.10, ** p < 0.05, *** p < 0.01
67
WDR Background Paper October 26, 2010
Fixed-effects models
Table A-7: Fixed-Effects Analysis, Undernourishment, 1992–2005
Conflict Battle Deaths BD*Pop. Fragility CPIA PCSE, AR(1)
1975-79 0 0 0 0 0 0
(0) (0) (0) (0) (0) (0)
1980-84 0 0 0 0 0 0
(0) (0) (0) (0) (0) (0)
1985-89 0 0 0 0 0 0
(0) (0) (0) (0) (0) (0)
1990-94 0 0 0 0 0 0
(0) (0) (0) (0) (0) (0)
1995-99 0 0 0 0 0 15.63∗∗∗
(0) (0) (0) (0) (0) (2.624)
2000-04 -0.894 -0.830 -0.788 -1.078∗-0.993 14.60∗∗∗
(0.578) (0.580) (0.581) (0.508) (0.617) (2.524)
2005-08 -2.324∗∗ -2.209∗∗ -2.158∗-2.520∗∗∗ -2.713∗∗∗ 13.54∗∗∗
(0.834) (0.838) (0.839) (0.681) (0.803) (2.533)
Log of Population -2.299 -3.036 -3.579 -2.269 -2.971 0.0581
(3.750) (3.749) (3.788) (3.371) (4.019) (0.323)
Conflict t-1 0.697∗∗∗ 1.114∗∗
(0.167) (0.385)
Conflict in Neighbourhood -0.0923 -0.128 -0.121 0.257
(0.111) (0.111) (0.111) (0.215)
Log of Battle Deaths t-1 0.416∗∗∗ 0.365∗∗
(0.106) (0.117)
Log of Battle Deaths t-1 X Log of Population 0.0822
(0.0816)
Fragility 0.608
(1.237)
Cumulative CPIA Score 0.566
(0.605)
Constant 39.50 46.26 51.00 39.21 46.15 0
(33.41) (33.38) (33.71) (30.08) (36.82) (0)
Observations 395 395 395 414 343 395
Log likelihood -939.5 -941.0 -940.2 -1016.5 -857.4
χ23242.8
Standard errors in parentheses
∗p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
Table A-8: Fixed-Effects Analysis, Poverty Rates, 1980–2005
Conflict Battle Deaths BD*Pop. Fragility CPIA PCSE, AR(1)
1975-79 0 0 0 0 0 0
(0) (0) (0) (0) (0) (0)
1980-84 0 0 0 -0.0156 -0.984 0
(0) (0) (0) (4.098) (4.197) (0)
1985-89 -0.0476 -0.801 -1.548 0 0 -9.473
(4.184) (4.156) (4.154) (0) (0) (5.911)
1990-94 -0.815 -1.725 -2.332 -0.633 -0.829 -11.58
(4.348) (4.273) (4.263) (1.855) (2.032) (7.312)
1995-99 -2.363 -3.425 -4.143 -2.167 -1.931 -11.87
(4.511) (4.422) (4.415) (1.979) (2.404) (6.868)
2000-04 -3.019 -4.201 -4.899 -2.773 -2.199 -13.20
(4.802) (4.697) (4.688) (2.240) (2.869) (6.900)
2005-08 -6.063 -7.168 -7.978 -6.122∗-5.486 -17.59∗
(5.050) (4.920) (4.914) (2.605) (3.394) (7.506)
Log of Population -10.60 -8.822 -8.278 -10.49 -15.16∗0.259
(5.893) (5.754) (5.730) (5.378) (7.519) (1.408)
Conflict t-1 -0.156 0.691∗
(0.271) (0.294)
Conflict in Neighbourhood 0.205 0.221 0.250 0.649
(0.184) (0.184) (0.184) (0.363)
Log of Battle Deaths t-1 0.0982 -0.129
(0.191) (0.231)
Log of Battle Deaths t-1 X Log of Population 0.198
(0.114)
Fragility 0.987
(2.234)
Cumulative CPIA Score 0.257
(0.854)
Constant 130.8∗114.1∗109.1∗129.4∗174.0∗32.71∗
(55.07) (53.87) (53.63) (50.82) (70.74) (14.62)
Observations 278 278 278 290 278 278
Log likelihood -827.9 -827.9 -825.5 -864.0 -828.5
χ211.99
Standard errors in parentheses
∗p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
68
WDR Background Paper October 26, 2010
Table A-9: Fixed-Effects Analysis, Life Expectancy, 1970–2005
Conflict Battle Deaths BD*Pop. Fragility CPIA PCSE, AR(1)
1975-79 1.654∗∗∗ 1.714∗∗∗ 1.713∗∗∗ 1.854∗∗∗ 0 2.149∗∗∗
(0.395) (0.393) (0.392) (0.379) (0) (0.0901)
1980-84 3.106∗∗∗ 3.171∗∗∗ 3.113∗∗∗ 3.398∗∗∗ -3.194∗∗∗ 4.370∗∗∗
(0.415) (0.413) (0.413) (0.397) (0.540) (0.132)
1985-89 4.388∗∗∗ 4.447∗∗∗ 4.362∗∗∗ 4.643∗∗∗ -1.236∗∗ 6.427∗∗∗
(0.463) (0.460) (0.460) (0.434) (0.433) (0.190)
1990-94 5.143∗∗∗ 5.202∗∗∗ 5.114∗∗∗ 5.412∗∗∗ 0 7.908∗∗∗
(0.517) (0.512) (0.512) (0.478) (0) (0.218)
1995-99 5.646∗∗∗ 5.734∗∗∗ 5.635∗∗∗ 5.681∗∗∗ 0.905∗8.954∗∗∗
(0.570) (0.566) (0.566) (0.521) (0.421) (0.256)
2000-04 5.885∗∗∗ 5.977∗∗∗ 5.856∗∗∗ 6.125∗∗∗ 1.982∗∗∗ 9.809∗∗∗
(0.621) (0.617) (0.617) (0.565) (0.491) (0.237)
2005-08 6.584∗∗∗ 6.658∗∗∗ 6.500∗∗∗ 6.848∗∗∗ 2.897∗∗∗ 10.86∗∗∗
(0.662) (0.657) (0.659) (0.607) (0.565) (0.266)
Log of Population 5.435∗∗∗ 5.292∗∗∗ 5.377∗∗∗ 5.355∗∗∗ 0.00203 -0.197
(0.732) (0.729) (0.728) (0.671) (1.462) (0.229)
Conflict t-1 -0.107 -0.154∗∗
(0.0559) (0.0522)
Conflict in Neighbourhood 0.0179 0.0229 0.0263 -0.0203
(0.0420) (0.0418) (0.0417) (0.0418)
Log of Battle Deaths t-1 -0.124∗∗∗ -0.165∗∗∗
(0.0359) (0.0399)
Log of Battle Deaths t-1 X Log of Population 0.0576∗
(0.0243)
Fragility -1.405∗∗∗
(0.329)
Cumulative CPIA Score 0.610∗∗
(0.195)
Constant 9.024 10.34 9.620 10.14 58.19∗∗∗ 56.82∗∗∗
(6.158) (6.136) (6.127) (5.614) (13.00) (2.114)
Observations 1029 1029 1029 1068 651 1029
Log likelihood -2436.1 -2431.3 -2428.0 -2515.0 -1502.7
χ22930.1
Standard errors in parentheses
∗p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
69
WDR Background Paper October 26, 2010
Table A-10: Effect of Conflict on Annual Growth in GDP per Capita (PPP, logged), 1960–2005,
OLS with Panel-Corrected Standard Errors
(1) (2) (3) (4) (5) (6)
C C with lags C, lags, FE BD BD with lags est6
conflict -0.0160∗∗∗ -0.0178∗∗∗ -0.0168∗∗∗
(0.00303) (0.00423) (0.00419)
td65 0.0361∗∗ 0.0560∗∗ 0.0549∗∗ 0.0466∗∗ 0.0358∗∗
(0.0126) (0.0181) (0.0180) (0.0166) (0.0126)
td70 0.0187 0.0211 0.0204 0.0163 0.0186
(0.0132) (0.0175) (0.0174) (0.0159) (0.0132)
td75 0.00353 0.00517 0.00379 0.00236 0.00290
(0.0131) (0.0172) (0.0171) (0.0157) (0.0132)
td80 -0.0101 -0.00662 -0.00800 -0.0119 -0.0113
(0.0128) (0.0169) (0.0168) (0.0156) (0.0128)
td85 -0.00727 -0.00553 -0.00678 -0.00981 -0.00839
(0.0125) (0.0166) (0.0165) (0.0154) (0.0126)
td90 -0.0206 -0.00998 -0.0111 -0.0156 -0.0221
(0.0124) (0.0165) (0.0164) (0.0155) (0.0124)
td95 0.00807 0.00834 0.00738 0.00237 0.00695
(0.0122) (0.0164) (0.0163) (0.0154) (0.0122)
td00 0.0132 0.0152 0.0147 0.00538 0.0127
(0.0122) (0.0163) (0.0162) (0.0154) (0.0122)
td05 0.0181 0.0193 0.0197 0.0105 0.0186
(0.0124) (0.0166) (0.0165) (0.0156) (0.0124)
conflict 1 -0.00571 -0.0103∗-0.00473
(0.00459) (0.00440) (0.00466)
conflict 2 0.00236 -0.000982 0.00300
(0.00469) (0.00474) (0.00473)
conflict 3 -0.000136 -0.00272 -0.000121
(0.00482) (0.00487) (0.00482)
conflict 4 0.00406 0.00292 0.00469
(0.00477) (0.00486) (0.00477)
conflict 5 0.00358 0.00287 0.00435
(0.00480) (0.00488) (0.00484)
conflict 6 0.00296 0.00223 0.00398
(0.00480) (0.00486) (0.00483)
conflict 7 -0.00122 -0.00160 -0.00130
(0.00483) (0.00488) (0.00486)
conflict 8 0.00528 0.00431 0.00483
(0.00485) (0.00489) (0.00487)
conflict 9 -0.000564 -0.0000236 -0.000683
(0.00475) (0.00481) (0.00480)
conflict 10 0.00236 0.00235 0.00193
(0.00435) (0.00440) (0.00430)
bd -0.00220∗∗∗ -0.00237∗
(0.000656) (0.00101)
bd 1 -0.00111
(0.00108)
bd 2 0.00215∗
(0.00109)
bd 3 -0.00103
(0.00109)
bd 4 -0.000803
(0.00107)
bd 5 0.000377
(0.00108)
bd 6 -0.0000260
(0.00108)
bd 7 -0.000445
(0.00109)
bd 8 0.00163
(0.00110)
bd 9 0.00122
(0.00107)
bd 10 -0.000501
(0.00100)
Constant 0.0172 0.0141 0.0143 0.0162 0.0165 0.0225∗∗∗
(0.0106) (0.0152) (0.0151) (0.0135) (0.0106) (0.00394)
Observations 5956 5235 5235 5235 5960 5235
Log likelihood
χ281.81 91.30 59.60 32611.0 61.81 21.05
Standard errors in parentheses
∗p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
C: measure. BD: Battle Deaths measure
70
WDR Background Paper October 26, 2010
A.3.2 MDG 2: Universal Education
Cross-sectional analyses
Table A-11: Improvement in primary school education attainment, Cross-section 1992–2005
(1) (2) (3) (4) (5)
improvement improvement improvement improvement improvement
exposure 1.000*** 0.992*** 1.018*** 1.115*** 0.836***
(0.173) (0.164) (0.160) (0.157) (0.214)
(firstnm) primaryschool -0.476*** -0.474*** -0.518*** -0.539*** -0.482***
(0.0518) (0.0498) (0.0500) (0.0488) (0.0505)
logged total population 1.488*** 1.691*** 2.041*** 1.180** 1.230**
(0.547) (0.515) (0.523) (0.474) (0.529)
(sum) minor 0.0187
(0.325)
(sum) war -0.416
(0.563)
bd1k -0.153**
(0.0681)
lnbd -0.812***
(0.239)
(sum) fsida -0.876***
(0.187)
(sum) cpia 0.0514
(0.0456)
Constant 18.33** 17.51** 18.57*** 24.74*** 22.07***
(7.313) (7.101) (6.909) (6.785) (7.621)
N 147 147 147 147 147
Start 1992.5 1992.5 1992.5 1992.5 1992.5
End 2005.2 2005.2 2005.2 2005.2 2005.2
r2 0.553 0.567 0.586 0.613 0.555
Standard errors in parentheses
*p < 0.10, ** p < 0.05, *** p < 0.01
Table A-12: Improvement in secondary school education enrollment, Cross-section 1973–2008
(1) (2) (3) (4) (5)
improvement improvement improvement improvement improvement
exposure 0.00472** 0.00460** 0.00468** 0.00442** 0.00407**
(0.00187) (0.00186) (0.00190) (0.00186) (0.00190)
(firstnm) edu -0.317*** -0.328*** -0.320*** -0.361*** -0.284***
(0.0584) (0.0582) (0.0615) (0.0608) (0.0585)
logged total population 0.00450 0.00530 0.00436 0.000265 -0.00153
(0.00688) (0.00634) (0.00707) (0.00618) (0.00639)
(sum) minor 0.00125
(0.00166)
(sum) war -0.00592**
(0.00234)
bd1k -0.000396***
(0.000149)
lnbd -0.00240
(0.00276)
(sum) fsida -0.00340***
(0.00120)
(sum) cpia 0.000636**
(0.000286)
Constant 0.239** 0.254*** 0.257*** 0.308*** 0.270***
(0.0935) (0.0924) (0.0944) (0.0938) (0.0933)
N 153 153 153 153 153
Start 1973.6 1973.6 1973.6 1973.6 1973.6
End 2008.9 2008.9 2008.9 2008.9 2008.9
r2 0.488 0.490 0.467 0.493 0.483
Standard errors in parentheses
*p < 0.10, ** p < 0.05, *** p < 0.01
71
WDR Background Paper October 26, 2010
Fixed-effects models
Table A-13: Fixed-Effects Analysis, Primary Education Attainment, 1995–2005
Conflict Battle Deaths BD*Pop. Fragility CPIA PCSE, AR(1)
1975-79 0 0 0 0 0 0
(0) (0) (0) (0) (0) (0)
1980-84 0 0 0 0 0 0
(0) (0) (0) (0) (0) (0)
1985-89 0 0 0 0 0 0
(0) (0) (0) (0) (0) (0)
1990-94 0 0 0 0 0 0
(0) (0) (0) (0) (0) (0)
1995-99 0 0 0 -0.857 -0.566 77.23
(0) (0) (0) (1.314) (1.585) .
2000-04 -0.126 -0.193 -0.131 -0.750 -0.823 80.85
(1.212) (1.212) (1.219) (0.869) (1.039) .
2005-08 0.766 0.632 0.709 0 0 83.50
(1.769) (1.773) (1.782) (0) (0) .
Log of Population 35.48∗∗∗ 35.31∗∗∗ 34.91∗∗∗ 32.63∗∗∗ 42.24∗∗∗ 0.636
(7.804) (7.797) (7.845) (6.325) (7.733) .
Conflict t-1 0.103 -1.033
(0.345) (0)
Conflict in Neighbourhood 0.240 0.215 0.225 -0.743
(0.217) (0.217) (0.218) (0)
Log of Battle Deaths t-1 -0.137 -0.180
(0.226) (0.240)
Log of Battle Deaths t-1 X Log of Population 0.0904
(0.168)
Fragility 6.620∗∗
(2.170)
Cumulative CPIA Score -0.800
(1.212)
Constant -234.1∗∗∗ -232.0∗∗∗ -228.7∗∗∗ -207.6∗∗∗ -301.1∗∗∗ 0
(68.04) (67.97) (68.36) (56.42) (70.82) (0)
Observations 348 348 348 361 291 348
Log likelihood -1019.4 -1019.1 -1018.9 -1049.0 -851.5
χ2.
Standard errors in parentheses
∗p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
Table A-14: Fixed-Effects Analysis, Secondary Education Attainment 1970–2005
Conflict Battle Deaths BD*Pop. Fragility CPIA PCSE, AR(1)
1975-79 0.0422∗∗∗ 0.0423∗∗∗ 0.0423∗∗∗ 0.0444∗∗∗ 0 0.0511∗∗∗
(0.00808) (0.00809) (0.00809) (0.00783) (0) (0.00564)
1980-84 0.0800∗∗∗ 0.0801∗∗∗ 0.0796∗∗∗ 0.0766∗∗∗ 0 0.0970∗∗∗
(0.00841) (0.00843) (0.00844) (0.00825) (0) (0.00813)
1985-89 0.121∗∗∗ 0.121∗∗∗ 0.120∗∗∗ 0.115∗∗∗ 0.0436∗∗∗ 0.149∗∗∗
(0.00929) (0.00928) (0.00930) (0.00891) (0.00739) (0.0109)
1990-94 0.150∗∗∗ 0.150∗∗∗ 0.149∗∗∗ 0.143∗∗∗ 0.0767∗∗∗ 0.189∗∗∗
(0.0103) (0.0103) (0.0103) (0.00973) (0.00905) (0.0137)
1995-99 0.174∗∗∗ 0.174∗∗∗ 0.173∗∗∗ 0.167∗∗∗ 0.105∗∗∗ 0.223∗∗∗
(0.0112) (0.0112) (0.0112) (0.0104) (0.0112) (0.0171)
2000-04 0.195∗∗∗ 0.195∗∗∗ 0.194∗∗∗ 0.186∗∗∗ 0.129∗∗∗ 0.255∗∗∗
(0.0121) (0.0121) (0.0121) (0.0113) (0.0130) (0.0140)
2005-08 0.213∗∗∗ 0.213∗∗∗ 0.212∗∗∗ 0.208∗∗∗ 0.157∗∗∗ 0.279∗∗∗
(0.0128) (0.0128) (0.0129) (0.0121) (0.0146) (0.0141)
Log of Population 0.0645∗∗∗ 0.0641∗∗∗ 0.0648∗∗∗ 0.0675∗∗∗ 0.0477 -0.00498
(0.0140) (0.0140) (0.0140) (0.0131) (0.0245) (0.00560)
Conflict t-1 -0.0000687 -0.00235∗
(0.00116) (0.00112)
Conflict in Neighbourhood -0.00250∗∗ -0.00249∗∗ -0.00245∗∗ -0.00184
(0.000866) (0.000867) (0.000867) (0.00132)
Log of Battle Deaths t-1 -0.000160 -0.000599
(0.000747) (0.000831)
Log of Battle Deaths t-1 X Log of Population 0.000608
(0.000506)
Fragility 0.000516
(0.00684)
Cumulative CPIA Score -0.00121
(0.00326)
Constant -0.168 -0.165 -0.170 -0.194 0.00257 0.425∗∗∗
(0.118) (0.119) (0.119) (0.110) (0.212) (0.0546)
Observations 1035 1035 1035 1081 652 1035
Log likelihood 1561.4 1561.4 1562.3 1624.1 1162.1
χ2821.4
Standard errors in parentheses
∗p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
72
WDR Background Paper October 26, 2010
A.3.3 MDG 3: Gender Parity
Cross-sectional analyses
Table A-15: Improvement in female-to-male primary school attainment ratio, cross-section 1975–
2006
(1) (2) (3) (4) (5)
improvement improvement improvement improvement improvement
exposure 0.246*** 0.257*** 0.260*** 0.235*** 0.192**
(0.0838) (0.0801) (0.0841) (0.0804) (0.0921)
(firstnm) femedu -0.826*** -0.846*** -0.834*** -0.873*** -0.825***
(0.0338) (0.0329) (0.0347) (0.0349) (0.0341)
logged total population -0.0180 0.122 0.146 -0.320 -0.357
(0.403) (0.356) (0.410) (0.350) (0.382)
(sum) minor 0.0768
(0.109)
(sum) war -0.369**
(0.156)
bd1k -0.0384***
(0.00935)
lnbd -0.293*
(0.164)
(sum) fsida -0.336***
(0.0838)
(sum) cpia 0.0271
(0.0184)
Constant 76.72*** 77.55*** 77.06*** 83.08*** 80.03***
(5.166) (4.895) (5.135) (5.088) (5.384)
N 151 151 151 151 151
Start 1975.7 1975.7 1975.7 1975.7 1975.7
End 2006.3 2006.3 2006.3 2006.3 2006.3
r2 0.889 0.897 0.887 0.896 0.886
Standard errors in parentheses
*p < 0.10, ** p < 0.05, *** p < 0.01
Table A-16: Improvement in female-to-male life expectancy ratio, cross-section 1967–2007
(1) (2) (3) (4) (5)
improvement improvement improvement improvement improvement
exposure -0.0146 -0.0115 -0.00935 -0.00772 -0.0134
(0.0287) (0.0288) (0.0289) (0.0286) (0.0292)
(firstnm) ratio -0.749*** -0.746*** -0.742*** -0.731*** -0.745***
(0.0686) (0.0688) (0.0690) (0.0688) (0.0688)
logged total population 0.191 0.293* 0.342** 0.288* 0.286*
(0.164) (0.155) (0.169) (0.149) (0.153)
(sum) minor 0.0319
(0.0322)
(sum) war 0.0454
(0.0465)
bd1k 0.000533
(0.00315)
lnbd -0.0338
(0.0596)
(sum) fsida -0.0420
(0.0265)
(sum) cpia 0.00281
(0.00641)
Constant 78.22*** 77.26*** 76.59*** 75.68*** 77.16***
(7.636) (7.652) (7.702) (7.631) (7.634)
N 153 153 153 153 153
Start 1967.6 1967.6 1967.6 1967.6 1967.6
End 2007.2 2007.2 2007.2 2007.2 2007.2
r2 0.600 0.593 0.593 0.600 0.593
Standard errors in parentheses
*p < 0.10, ** p < 0.05, *** p < 0.01
73
WDR Background Paper October 26, 2010
Fixed-effects models
Table A-17: Fixed-Effects Analysis, Female-to-Male Primary School Attainment Ratio 1975–2006
Conflict Battle Deaths BD*Pop. Fragility CPIA PCSE, AR(1)
1975-79 -0.210 -0.0572 -0.0462 -1.285 0 2.125
(1.569) (1.574) (1.574) (1.511) (0) (1.287)
1980-84 0.0263 0.304 0.251 0.381 1.637 6.171∗∗∗
(1.632) (1.634) (1.634) (1.574) (1.177) (1.451)
1985-89 -0.944 -0.512 -0.581 0.419 0.429 8.720∗∗∗
(1.763) (1.761) (1.761) (1.682) (0.940) (1.583)
1990-94 -1.314 -0.785 -0.837 0.566 0 11.16∗∗∗
(1.907) (1.902) (1.902) (1.809) (0) (1.675)
1995-99 -2.386 -1.805 -1.862 0.0975 -1.604 12.51∗∗∗
(2.051) (2.045) (2.044) (1.941) (0.911) (1.737)
2000-04 -2.615 -2.008 -2.089 -0.311 -2.004 14.21∗∗∗
(2.205) (2.199) (2.199) (2.075) (1.078) (1.731)
2005-08 -1.215 -0.574 -0.706 0.513 -1.680 16.57∗∗∗
(2.331) (2.325) (2.326) (2.202) (1.242) (1.753)
Log of Population 21.56∗∗∗ 21.07∗∗∗ 21.08∗∗∗ 18.51∗∗∗ 25.53∗∗∗ -0.855∗
(2.295) (2.296) (2.294) (2.158) (3.220) (0.407)
Conflict t-1 0.406∗∗ -0.186
(0.148) (0.193)
Conflict in Neighbourhood 0.553∗∗∗ 0.566∗∗∗ 0.569∗∗∗ -0.00739
(0.116) (0.116) (0.116) (0.160)
Log of Battle Deaths t-1 0.121 0.0554
(0.0972) (0.108)
Log of Battle Deaths t-1 X Log of Population 0.0903
(0.0655)
Fragility 1.243
(0.941)
Cumulative CPIA Score -0.756
(0.424)
Constant -103.1∗∗∗ -99.13∗∗∗ -99.18∗∗∗ -73.64∗∗∗ -137.2∗∗∗ 85.97∗∗∗
(19.13) (19.14) (19.12) (17.91) (28.71) (3.511)
Observations 907 907 907 939 631 907
Log likelihood -2969.1 -2972.7 -2971.5 -3101.3 -1938.5
χ21237.3
Standard errors in parentheses
∗p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
Table A-18: Fixed-Effects Analysis, Female-to-Male Life Expectancy Ratio 1967–2007
Conflict Battle Deaths BD*Pop. Fragility CPIA PCSE, AR(1)
1975-79 0.629∗∗ 0.621∗∗ 0.620∗∗ 0.498∗0 0.350∗
(0.205) (0.206) (0.205) (0.202) (0) (0.138)
1980-84 1.134∗∗∗ 1.133∗∗∗ 1.104∗∗∗ 1.104∗∗∗ -0.634∗0.758∗∗∗
(0.216) (0.216) (0.216) (0.213) (0.252) (0.210)
1985-89 1.548∗∗∗ 1.558∗∗∗ 1.514∗∗∗ 1.542∗∗∗ -0.406∗1.039∗∗∗
(0.241) (0.240) (0.240) (0.232) (0.202) (0.268)
1990-94 1.995∗∗∗ 2.009∗∗∗ 1.964∗∗∗ 2.050∗∗∗ 0 1.366∗∗∗
(0.269) (0.268) (0.268) (0.255) (0) (0.322)
1995-99 2.340∗∗∗ 2.353∗∗∗ 2.302∗∗∗ 2.483∗∗∗ 0.325 1.627∗∗∗
(0.297) (0.296) (0.296) (0.279) (0.196) (0.380)
2000-04 2.107∗∗∗ 2.119∗∗∗ 2.056∗∗∗ 2.117∗∗∗ -0.119 1.277∗∗∗
(0.323) (0.323) (0.323) (0.302) (0.229) (0.296)
2005-08 1.936∗∗∗ 1.951∗∗∗ 1.869∗∗∗ 1.861∗∗∗ -0.475 0.848∗∗
(0.344) (0.344) (0.345) (0.325) (0.263) (0.289)
Log of Population -2.016∗∗∗ -2.020∗∗∗ -1.976∗∗∗ -2.047∗∗∗ -1.121 -0.0387
(0.381) (0.381) (0.381) (0.359) (0.681) (0.109)
Conflict t-1 0.0404 0.0155
(0.0291) (0.0371)
Conflict in Neighbourhood 0.0609∗∗ 0.0610∗∗ 0.0627∗∗ -0.00344
(0.0218) (0.0219) (0.0218) (0.0230)
Log of Battle Deaths t-1 0.0233 0.00176
(0.0188) (0.0209)
Log of Battle Deaths t-1 X Log of Population 0.0298∗
(0.0127)
Fragility 0.0904
(0.176)
Cumulative CPIA Score -0.141
(0.0908)
Constant 122.9∗∗∗ 122.9∗∗∗ 122.6∗∗∗ 123.6∗∗∗ 118.2∗∗∗ 106.8∗∗∗
(3.205) (3.210) (3.205) (3.002) (6.055) (1.125)
Observations 1029 1029 1029 1068 651 1029
Log likelihood -1764.3 -1764.6 -1761.3 -1846.6 -1005.3
χ225.59
Standard errors in parentheses
∗p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
74
WDR Background Paper October 26, 2010
A.3.4 MDG 4: Child Mortality
Cross-sectional analyses
Table A-19: Improvement in log infant mortality rates, Cross-section 1969–2005
(1) (2) (3) (4) (5)
improvement improvement improvement improvement improvement
exposure -0.0343*** -0.0345*** -0.0345*** -0.0307*** -0.0322***
(0.00431) (0.00417) (0.00428) (0.00416) (0.00443)
(firstnm) limr -0.0510 -0.0525 -0.0733 -0.154** -0.0179
(0.0709) (0.0686) (0.0713) (0.0722) (0.0731)
logged total population 0.0321 0.0272 0.0165 0.0597** 0.0543**
(0.0275) (0.0251) (0.0277) (0.0242) (0.0257)
(sum) minor 0.000496
(0.00534)
(sum) war 0.0195**
(0.00768)
bd1k 0.00208***
(0.000528)
lnbd 0.0277***
(0.00993)
(sum) fsida 0.0202***
(0.00454)
(sum) cpia -0.00164
(0.00107)
Constant -0.238 -0.219 -0.136 -0.137 -0.536*
(0.315) (0.302) (0.321) (0.300) (0.321)
N 153 153 153 153 153
Start 1969.9 1969.9 1969.9 1969.9 1969.9
End 2006.9 2006.9 2006.9 2006.9 2006.9
r2 0.652 0.671 0.654 0.680 0.641
Standard errors in parentheses
*p < 0.10, ** p < 0.05, *** p < 0.01
Table A-20: Improvement in under-five mortality rate, Cross-section 1991–2003
(1) (2) (3) (4) (5)
improvement improvement improvement improvement improvement
exposure -0.948*** -0.974*** -0.935*** -0.674*** -0.629**
(0.279) (0.261) (0.279) (0.255) (0.280)
(firstnm) underfivemort -0.683*** -0.689*** -0.699*** -0.752*** -0.663***
(0.0360) (0.0337) (0.0369) (0.0348) (0.0352)
logged total population -0.733 -1.627 -1.939 1.005 1.238
(1.826) (1.592) (1.845) (1.509) (1.636)
(sum) minor -0.0866
(0.376)
(sum) war 1.458***
(0.514)
bd1k 0.185***
(0.0345)
lnbd 1.874***
(0.679)
(sum) fsida 1.808***
(0.288)
(sum) cpia -0.260***
(0.0677)
Constant 18.54 23.15 20.87 4.084 -0.0447
(15.63) (14.29) (15.62) (13.65) (14.83)
N 153 153 153 153 153
Start 1970.2 1970.2 1970.2 1970.2 1970.2
End 2006.9 2006.9 2006.9 2006.9 2006.9
r2 0.846 0.864 0.845 0.872 0.852
Standard errors in parentheses
*p < 0.10, ** p < 0.05, *** p < 0.01
75
WDR Background Paper October 26, 2010
Fixed-effects models
Table A-21: Fixed-Effects Analysis, Log Infant Mortality Rates 1970–2005
Conflict Battle Deaths BD*Pop. Fragility CPIA PCSE, AR(1)
1975-79 -0.197∗∗∗ -0.205∗∗∗ -0.206∗∗∗ -0.212∗∗∗ 0 -0.160∗∗∗
(0.0318) (0.0316) (0.0315) (0.0311) (0) (0.00703)
1980-84 -0.430∗∗∗ -0.436∗∗∗ -0.431∗∗∗ -0.433∗∗∗ 0 -0.351∗∗∗
(0.0330) (0.0328) (0.0328) (0.0325) (0) (0.0124)
1985-89 -0.651∗∗∗ -0.656∗∗∗ -0.649∗∗∗ -0.635∗∗∗ -0.249∗∗∗ -0.536∗∗∗
(0.0369) (0.0366) (0.0366) (0.0356) (0.0265) (0.0216)
1990-94 -0.835∗∗∗ -0.841∗∗∗ -0.834∗∗∗ -0.809∗∗∗ -0.458∗∗∗ -0.689∗∗∗
(0.0410) (0.0405) (0.0405) (0.0388) (0.0322) (0.0265)
1995-99 -1.005∗∗∗ -1.014∗∗∗ -1.005∗∗∗ -0.951∗∗∗ -0.629∗∗∗ -0.832∗∗∗
(0.0452) (0.0447) (0.0448) (0.0424) (0.0398) (0.0318)
2000-04 -1.172∗∗∗ -1.182∗∗∗ -1.172∗∗∗ -1.132∗∗∗ -0.840∗∗∗ -0.975∗∗∗
(0.0492) (0.0487) (0.0488) (0.0459) (0.0462) (0.0277)
2005-08 -1.352∗∗∗ -1.359∗∗∗ -1.346∗∗∗ -1.315∗∗∗ -1.048∗∗∗ -1.130∗∗∗
(0.0524) (0.0519) (0.0521) (0.0493) (0.0520) (0.0293)
Log of Population 0.369∗∗∗ 0.384∗∗∗ 0.376∗∗∗ 0.289∗∗∗ 0.725∗∗∗ 0.0767∗∗∗
(0.0582) (0.0578) (0.0577) (0.0544) (0.0875) (0.0200)
Conflict t-1 0.0111∗0.00863∗∗
(0.00441) (0.00277)
Conflict in Neighbourhood 0.00511 0.00455 0.00424 0.00534
(0.00334) (0.00331) (0.00330) (0.00311)
Log of Battle Deaths t-1 0.0131∗∗∗ 0.0163∗∗∗
(0.00284) (0.00321)
Log of Battle Deaths t-1 X Log of Population -0.00427∗
(0.00195)
Fragility 0.0997∗∗∗
(0.0270)
Cumulative CPIA Score -0.0121
(0.0116)
Constant 1.314∗∗ 1.174∗1.239∗2.024∗∗∗ -1.959∗3.711∗∗∗
(0.492) (0.488) (0.488) (0.457) (0.759) (0.194)
Observations 1006 1006 1006 1043 642 1006
Log likelihood 181.8 190.5 193.3 186.1 340.0
χ24304.6
Standard errors in parentheses
∗p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
Table A-22: Fixed-Effects Analysis, Under-Five Year Mortality Rates 1970–2005
Conflict Battle Deaths BD*Pop. Fragility CPIA PCSE, AR(1)
1975-79 -14.37∗∗∗ -14.10∗∗∗ -14.12∗∗∗ -16.26∗∗∗ 0 -20.09∗∗∗
(2.724) (2.723) (2.711) (2.657) (0) (0.652)
1980-84 -26.68∗∗∗ -26.43∗∗∗ -25.84∗∗∗ -29.07∗∗∗ 19.07∗∗∗ -40.28∗∗∗
(2.844) (2.841) (2.835) (2.793) (3.014) (0.974)
1985-89 -34.60∗∗∗ -34.32∗∗∗ -33.56∗∗∗ -38.00∗∗∗ 6.763∗∗ -56.00∗∗∗
(3.162) (3.149) (3.145) (3.036) (2.397) (1.579)
1990-94 -38.63∗∗∗ -38.35∗∗∗ -37.55∗∗∗ -42.80∗∗∗ 0 -67.12∗∗∗
(3.520) (3.502) (3.496) (3.331) (0) (2.127)
1995-99 -41.17∗∗∗ -40.76∗∗∗ -39.84∗∗∗ -45.19∗∗∗ -4.938∗-75.71∗∗∗
(3.883) (3.866) (3.860) (3.636) (2.327) (2.536)
2000-04 -44.22∗∗∗ -43.80∗∗∗ -42.68∗∗∗ -49.43∗∗∗ -12.21∗∗∗ -84.47∗∗∗
(4.225) (4.210) (4.208) (3.940) (2.722) (2.071)
2005-08 -48.86∗∗∗ -48.51∗∗∗ -47.07∗∗∗ -54.18∗∗∗ -18.56∗∗∗ -92.33∗∗∗
(4.500) (4.482) (4.487) (4.228) (3.136) (2.137)
Log of Population -53.88∗∗∗ -54.54∗∗∗ -55.36∗∗∗ -48.02∗∗∗ -34.09∗∗∗ 1.603
(4.990) (4.985) (4.970) (4.659) (8.198) (1.671)
Conflict t-1 -0.393 0.923∗
(0.382) (0.419)
Conflict in Neighbourhood -0.251 -0.229 -0.261 0.381
(0.287) (0.287) (0.286) (0.351)
Log of Battle Deaths t-1 -0.502∗-0.124
(0.246) (0.276)
Log of Battle Deaths t-1 X Log of Population -0.499∗∗
(0.168)
Fragility 3.516
(2.318)
Cumulative CPIA Score -2.390∗
(1.081)
Constant 609.6∗∗∗ 615.7∗∗∗ 622.7∗∗∗ 556.1∗∗∗ 415.4∗∗∗ 138.0∗∗∗
(42.13) (42.11) (41.98) (39.07) (73.07) (14.62)
Observations 1014 1014 1014 1051 646 1014
Log likelihood -4336.1 -4334.3 -4329.1 -4494.2 -2592.4
χ25283.8
Standard errors in parentheses
∗p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
76
WDR Background Paper October 26, 2010
A.3.5 MDG 5: Maternal Mortality
Cross-sectional analyses
Table A-23: Improvement in proportion of births attended, Cross-section analysis 1997–2004
(1) (2) (3) (4) (5)
improvement improvement improvement improvement improvement
exposure 0.531*** 0.547*** 0.547*** 0.609*** 0.322
(0.169) (0.165) (0.166) (0.165) (0.223)
(firstnm) attended -0.100*** -0.105*** -0.106*** -0.126*** -0.104***
(0.0332) (0.0322) (0.0329) (0.0328) (0.0319)
logged total population 0.358 0.412 0.364 0.240 0.189
(0.430) (0.432) (0.440) (0.407) (0.423)
(sum) minor 0.293
(0.441)
(sum) war -0.413
(0.698)
bd1k -0.0340
(0.0700)
lnbd -0.0230
(0.217)
(sum) fsida -0.552**
(0.241)
(sum) cpia 0.0871
(0.0584)
Constant 6.619 6.528 7.032 9.362* 9.160
(5.520) (5.529) (5.432) (5.408) (5.552)
N 147 147 147 147 147
Start 1997.9 1997.9 1997.9 1997.9 1997.9
End 2004.4 2004.4 2004.4 2004.4 2004.4
r2 0.255 0.253 0.252 0.280 0.264
Standard errors in parentheses
*p < 0.10, ** p < 0.05, *** p < 0.01
77
WDR Background Paper October 26, 2010
Fixed-effects models
Table A-24: Fixed-Effects Analysis, Births Attended by Skilled Health Workers 1995–2005
Conflict Battle Deaths BD*Pop. Fragility CPIA PCSE, AR(1)
1975-79 0 0 0 0 0 0
(0) (0) (0) (0) (0) (0)
1980-84 0 0 0 0 0 0
(0) (0) (0) (0) (0) (0)
1985-89 0 0 0 0 0 0
(0) (0) (0) (0) (0) (0)
1990-94 0 0 0 0 0 0
(0) (0) (0) (0) (0) (0)
1995-99 -3.564∗∗ -3.698∗∗ -3.732∗∗ 0 0 98.07
(1.264) (1.259) (1.252) (0) (0) .
2000-04 -1.476∗-1.583∗-1.584∗1.323∗1.056 100.1
(0.719) (0.710) (0.706) (0.644) (0.771) .
2005-08 0 0 0 2.481∗∗ 2.873∗∗ 103.3
(0) (0) (0) (0.809) (0.935) .
Log of Population 13.38∗12.69∗12.88∗17.00∗∗∗ 20.43∗∗∗ -2.481
(5.768) (5.755) (5.720) (4.438) (5.529) (0)
Conflict t-1 0.289 -2.043
(0.338) (0)
Conflict in Neighbourhood 0.0617 0.0606 0.0616 -0.389
(0.168) (0.169) (0.168) (0)
Log of Battle Deaths t-1 -0.00541 0.159
(0.178) (0.206)
Log of Battle Deaths t-1 X Log of Population -0.230
(0.147)
Fragility 3.466∗
(1.509)
Cumulative CPIA Score -1.589
(0.804)
Constant -44.98 -38.30 -39.80 -79.23∗-111.2∗0
(52.52) (52.37) (52.04) (39.89) (51.29) (0)
Observations 244 244 244 258 215 244
Log likelihood -554.6 -555.4 -552.8 -579.4 -489.3
χ2.
Standard errors in parentheses
∗p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
78
WDR Background Paper October 26, 2010
A.3.6 MDG 6: Combat HIV/AIDS
Cross-sectional analyses
Table A-25: Improvement in HIV/AIDS prevalence, Cross-section analysis 1991–2003
(1) (2) (3) (4) (5)
improvement improvement improvement improvement improvement
exposure 0.146 0.110 0.160 0.147 0.0326
(0.104) (0.106) (0.106) (0.104) (0.111)
(firstnm) hivprev -0.345* -0.397** -0.402** -0.422** -0.387**
(0.175) (0.179) (0.175) (0.174) (0.180)
logged total population -0.0757 -0.169 -0.0881 -0.343 -0.396
(0.240) (0.244) (0.240) (0.226) (0.244)
(sum) minor -0.280**
(0.118)
(sum) war -0.181
(0.214)
bd1k -0.0551*
(0.0326)
lnbd -0.287***
(0.105)
(sum) fsida -0.212***
(0.0681)
(sum) cpia 0.0272
(0.0198)
Constant 0.411 0.548 0.406 1.852 2.594
(2.792) (2.876) (2.765) (2.688) (2.828)
N 123 123 123 123 123
Start 1993.1 1993.1 1993.1 1993.1 1993.1
End 2007.0 2007.0 2007.0 2007.0 2007.0
r2 0.254 0.211 0.242 0.256 0.205
Standard errors in parentheses
*p < 0.10, ** p < 0.05, *** p < 0.01
79
WDR Background Paper October 26, 2010
Fixed-effects models
Table A-26: Fixed-Effects Analysis, HIV/AIDS Sero-Prevalence 1990–2005
Conflict Battle Deaths BD*Pop. Fragility CPIA PCSE, AR(1)
1975-79 0 0 0 0 0 0
(0) (0) (0) (0) (0) (0)
1980-84 0 0 0 0 0 0
(0) (0) (0) (0) (0) (0)
1985-89 0 0 0 0 0 0
(0) (0) (0) (0) (0) (0)
1990-94 0 0 0 -1.533∗∗∗ 0 2.223
(0) (0) (0) (0.428) (0) (1.329)
1995-99 1.441∗∗∗ 1.440∗∗∗ 1.427∗∗∗ 0 2.008∗∗∗ 3.709∗∗
(0.409) (0.411) (0.404) (0) (0.543) (1.308)
2000-04 1.721∗∗ 1.720∗∗ 1.690∗∗ 0.557 2.790∗∗∗ 4.285∗∗∗
(0.519) (0.522) (0.513) (0.407) (0.792) (1.297)
2005-08 1.199 1.183 1.217 0.372 2.863∗∗ 4.117∗∗
(0.641) (0.642) (0.631) (0.491) (0.996) (1.263)
Log of Population 1.970 2.090 2.446 1.742 -1.083 -0.129
(1.832) (1.828) (1.801) (1.860) (2.939) (0.149)
Conflict t-1 -0.0586 -0.0370
(0.0900) (0.0609)
Conflict in Neighbourhood -0.274∗∗∗ -0.277∗∗∗ -0.284∗∗∗ -0.0849
(0.0642) (0.0643) (0.0633) (0.0776)
Log of Battle Deaths t-1 -0.0164 0.0188
(0.0615) (0.0615)
Log of Battle Deaths t-1 X Log of Population -0.172∗∗
(0.0544)
Fragility -0.867
(0.609)
Cumulative CPIA Score -0.234
(0.327)
Constant -14.73 -15.82 -18.71 -12.82 11.11 0
(15.95) (15.91) (15.68) (16.37) (25.97) (0)
Observations 386 386 386 389 334 386
Log likelihood -803.0 -803.3 -796.1 -827.0 -712.7
χ2179.8
Standard errors in parentheses
∗p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
80
WDR Background Paper October 26, 2010
A.3.7 MDG 7: Environmental Sustainability
Cross-sectional analyses
Table A-27: Improvement in percentage with access to safe water, Cross-section analysis 1991–2003
(1) (2) (3) (4) (5)
improvement improvement improvement improvement improvement
exposure 0.430** 0.441** 0.431** 0.434** 0.221
(0.194) (0.192) (0.192) (0.187) (0.216)
(firstnm) water -0.299*** -0.302*** -0.312*** -0.339*** -0.301***
(0.0392) (0.0386) (0.0394) (0.0396) (0.0391)
logged total population 0.889** 1.026** 1.037** 0.551 0.535
(0.432) (0.417) (0.425) (0.395) (0.418)
(sum) minor 0.125
(0.208)
(sum) war -0.618**
(0.298)
bd1k -0.0830**
(0.0322)
lnbd -0.398**
(0.176)
(sum) fsida -0.459***
(0.127)
(sum) cpia 0.0633*
(0.0333)
Constant 13.96** 13.61** 14.92** 20.62*** 18.71***
(6.216) (6.125) (6.129) (6.153) (6.440)
N 147 147 147 147 147
Start 1991.7 1991.7 1991.7 1991.7 1991.7
End 2005.5 2005.5 2005.5 2005.5 2005.5
r2 0.468 0.477 0.471 0.499 0.466
Standard errors in parentheses
*p < 0.10, ** p < 0.05, *** p < 0.01
Table A-28: Improvement in percentage with access to sanitation, Cross-section analysis 1991–2003
(1) (2) (3) (4) (5)
improvement improvement improvement improvement improvement
exposure 0.495** 0.485** 0.498** 0.500** 0.435**
(0.195) (0.193) (0.193) (0.193) (0.205)
(firstnm) sanitation -0.178*** -0.177*** -0.180*** -0.187*** -0.175***
(0.0357) (0.0347) (0.0356) (0.0370) (0.0347)
logged total population 0.866* 0.770* 0.912** 0.805* 0.763*
(0.445) (0.433) (0.439) (0.418) (0.427)
(sum) minor -0.0226
(0.227)
(sum) war -0.00160
(0.300)
bd1k 0.0210
(0.0335)
lnbd -0.0733
(0.179)
(sum) fsida -0.101
(0.135)
(sum) cpia 0.0269
(0.0329)
Constant 4.810 5.472 4.738 6.016 5.645
(5.774) (5.675) (5.615) (5.793) (5.668)
N 143 143 143 143 143
Start 1991.8 1991.8 1991.8 1991.8 1991.8
End 2005.6 2005.6 2005.6 2005.6 2005.6
r2 0.394 0.396 0.395 0.397 0.397
Standard errors in parentheses
*p < 0.10, ** p < 0.05, *** p < 0.01
81
WDR Background Paper October 26, 2010
Fixed-effects models
Table A-29: Fixed-Effects Analysis, Access to Water 1990–2005
Conflict Battle Deaths BD*Pop. Fragility CPIA PCSE, AR(1)
1975-79 0 0 0 0 0 0
(0) (0) (0) (0) (0) (0)
1980-84 0 0 0 0 0 0
(0) (0) (0) (0) (0) (0)
1985-89 0 0 0 0 0 0
(0) (0) (0) (0) (0) (0)
1990-94 0 0 0 0 0 -7.097
(0) (0) (0) (0) (0) (0)
1995-99 2.051∗∗ 2.008∗∗ 2.009∗∗ 2.085∗∗ 1.658∗-4.580
(0.637) (0.638) (0.638) (0.643) (0.800) (0)
2000-04 4.150∗∗∗ 4.054∗∗∗ 4.073∗∗∗ 3.894∗∗∗ 3.384∗∗ -2.005
(0.817) (0.817) (0.818) (0.787) (1.047) (0)
2005-08 6.424∗∗∗ 6.229∗∗∗ 6.283∗∗∗ 5.826∗∗∗ 5.409∗∗∗ 0
(1.023) (1.021) (1.023) (0.955) (1.277) (0)
Log of Population 4.503 5.263 5.245 7.485∗∗ 11.82∗∗ -1.026
(2.980) (2.955) (2.956) (2.784) (3.828) (0)
Conflict t-1 -0.398∗∗ -0.733
(0.139) (0)
Conflict in Neighbourhood 0.0978 0.0927 0.0873 -0.147
(0.101) (0.101) (0.101) (0)
Log of Battle Deaths t-1 -0.233∗-0.185
(0.0941) (0.110)
Log of Battle Deaths t-1 X Log of Population -0.0557
(0.0661)
Fragility 1.881
(1.100)
Cumulative CPIA Score -0.287
(0.502)
Constant 33. 47 26.72 26.92 6.768 -34.86 92.02
(26.38) (26.14) (26.15) (24.62) (34.41) .
Observations 500 500 500 518 420 500
Log likelihood -1329.6 -1331.1 -1330.6 -1391.4 -1121.8
χ2.
Standard errors in parentheses
∗p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
Table A-30: Fixed-Effects Analysis, Access to Sanitation Facilities 1990–2005
Conflict Battle Deaths BD*Pop. Fragility CPIA PCSE, AR(1)
1975-79 0 0 0 0 0 0
(0) (0) (0) (0) (0) (0)
1980-84 0 0 0 0 0 0
(0) (0) (0) (0) (0) (0)
1985-89 0 0 0 0 0 0
(0) (0) (0) (0) (0) (0)
1990-94 0 0 0 0 0 75.92∗∗∗
(0) (0) (0) (0) (0) (11.41)
1995-99 2.286∗∗∗ 2.217∗∗∗ 2.215∗∗∗ 2.260∗∗∗ 2.390∗∗∗ 78.48∗∗∗
(0.616) (0.619) (0.619) (0.605) (0.671) (11.54)
2000-04 4.493∗∗∗ 4.381∗∗∗ 4.376∗∗∗ 3.967∗∗∗ 4.124∗∗∗ 81.10∗∗∗
(0.787) (0.789) (0.790) (0.738) (0.881) (11.43)
2005-08 6.341∗∗∗ 6.173∗∗∗ 6.155∗∗∗ 5.706∗∗∗ 5.914∗∗∗ 82.86∗∗∗
(0.986) (0.986) (0.989) (0.895) (1.075) (11.27)
Log of Population 2.106 2.888 2.893 4.777 6.261 -1.951
(2.861) (2.843) (2.847) (2.610) (3.245) (1.153)
Conflict t-1 -0.267∗-0.503∗∗
(0.133) (0.174)
Conflict in Neighbourhood -0.0625 -0.0664 -0.0640 -0.183
(0.0987) (0.0991) (0.0995) (0.172)
Log of Battle Deaths t-1 -0.0815 -0.102
(0.0916) (0.111)
Log of Battle Deaths t-1 X Log of Population 0.0213
(0.0666)
Fragility 0.679
(1.017)
Cumulative CPIA Score 0.0118
(0.423)
Constant 37. 58 30.44 30.38 13.77 -5.769 0
(25.41) (25.24) (25.27) (23.17) (29.23) (0)
Observations 485 485 485 502 412 485
Log likelihood -1267.3 -1269.6 -1269.6 -1310.1 -1022.7
χ21396.4
Standard errors in parentheses
∗p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
82
WDR Background Paper October 26, 2010
Regression results underlying Figure 5
83
WDR Background Paper October 26, 2010
Table A-31: Effect of Conflict on Annual Growth in GDP per Capita (PPP, logged),
(1)
Growth
lgdpcp -0.00267∗
(0.00109)
conflict -0.0192∗∗∗
(0.00270)
conflict 1 -0.00685∗
(0.00321)
conflict 2 0.00502
(0.00325)
conflict 3 0.00191
(0.00329)
conflict 4 0.00691∗
(0.00325)
conflict 5 0.00902∗∗∗
(0.00273)
td75 -0.0113∗∗∗
(0.00325)
td80 -0.0316∗∗∗
(0.00321)
td85 -0.0227∗∗∗
(0.00323)
td90 -0.0304∗∗∗
(0.00326)
td95 -0.0148∗∗∗
(0.00329)
td00 -0.0187∗∗∗
(0.00334)
td05 -0.00586
(0.00356)
regionb==ECA 0.0139
(0.00969)
regionb==LAC -0.0197∗∗∗
(0.00344)
regionb==MNA -0.0164∗∗∗
(0.00382)
regionb==OECD -0.0155∗∗∗
(0.00413)
regionb==SAR -0.00861
(0.00487)
regionb==SSA -0.0229∗∗∗
(0.00363)
ethfra -0.0161∗∗∗
(0.00367)
fraction of pop with attained secondary education 0.0215∗∗∗
(0.00571)
lnpop 0.00138∗
(0.000575)
Constant 0.0542∗∗∗
(0.0103)
Observations 4401
Log likelihood 6860.6
χ2
Standard errors in parentheses
∗p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
84
