State Failure Task Force Report: Phase II Findings

Abstract

In response to a request from Vice President Al Gore in 1994, the CIA established "The State Failure Task Force," a group of independent researchers to examine comprehensively the factors and forces that have affected the stability of the post-Cold War world. The Task Force's goal was to identify the factors or combinations of factors that distinguish states that failed from those, which averted crises over the last 40 years. The study represents the first empirical effort to identify factors associated with state failure by examining a broad range of demographic, societal, economic, environmental, and political indicators influencing state stability. The Task Force found that three clusters of variables had significant correlation with subsequent state failures: (1) quality of life; (2) openness to international trade; and (3) the level of democracy. However, it is the interaction among these variables that provided the most important insights. Following are excepts from Phase II of the State Failure Task Force findings.

Full text

Author ⋅ Title
49ENVIRONMENTAL CHANGE & SECURITY PROJECT REPORT, ISSUE 5 (SUMMER 1999)
State Failure Task Force Report: Phase II Findings
Prepared by Daniel C. Esty, Jack A. Goldstone,
Ted Robert Gurr, Barbara Harff, Marc Levy, Geoffrey D. Dabelko,
Pamela T. Surko, and Alan N. Unger
Abstract: In response to a request from Vice President Al Gore in 1994, the CIA established “The State Failure Task Force,” a group of
independent researchers to examine comprehensively the factors and forces that have affected the stability of the post-Cold War world. The
Task Force’s goal was to identify the factors or combinations of factors that distinguish states that failed from those, which averted crises over
the last 40 years. The study represents the first empirical effort to identify factors associated with state failure by examining a broad range of
demographic, societal, economic, environmental, and political indicators influencing state stability. The Task Force found that three clusters
of variables had significant correlation with subsequent state failures: (1) quality of life; (2) openness to international trade; and (3) the level of
democracy. However, it is the interaction among these variables that provided the most important insights. Following are excepts from Phase
II of the State Failure Task Force findings.
INTRODUCTION
The initial report of the State Failure Task Force
1
developed a global model of the factors that contributed to serious
political crises over the last four decades. In this report, we describe the progress of the Task Force on four additional
research issues:
• Confirmation and refinement of the global model. This work included testing the model on an updated problem set,
varying the set of control cases, and testing new or refined variables. In particular, we refined the level-of-democracy variable
to examine partial democracies—countries that combine democratic and autocratic features—and their risks of state failure.
• Fitting a model for Sub-Saharan Africa. We also examined how the global model might best be modified to apply to the
countries of Sub-Saharan Africa. To improve the accuracy of prediction, the Task Force undertook a pilot study of event
sequences in a limited number of Sub-Saharan African cases of state failure and state stability to identify factors that could
be precipitators or “accelerators” of crises.
• Transitions to democracy and autocracy. The initial study only examined cases of adverse or disruptive regime transitions.
Because of the great interest in transitions to democracy, and the conditions that provide for stable or unstable democracy,
the Task Force applied its methodology for analyzing risks of state failure to transitions toward and away from democracy.
This report explores the preliminary findings of these analyses of the emergence and decay of democratic regimes.
• The role of environmental factors in state failure. It appeared from the Phase I results that environmental factors did not
directly contribute to the risks of state failure. The Task Force believes that this finding was due, in part, to the paucity, poor
quality, and lack of comparability of the national-level environmental data and, in part, to the impact of environmental
factors on political conflicts being mediated by other economic, social, and political conditions. We, therefore, undertook
special initiatives to assess the state of global environmental data and to develop a mediated, two-stage model of the role of
environmental factors on the risks of state failure. In this model, it appears that environmental hazards—in states with
underlying vulnerabilities and limited governmental or social capacity to respond to environmental deterioration—is associated
with increased risk of state failure.
Daniel C. Esty, Yale University; Jack A. Goldstone, University of California, Davis; Ted Robert Gurr, University of Maryland, College Park;
Barbara Harff, US Naval Academy; Marc Levy, Columbia University; Geoffrey D. Dabelko, Woodrow Wilson Center; Pamela T. Surko,
Science Applications International Corporation (SAIC); and Alan N. Unger, SAIC. Copyright SAIC, 31 July 1998. Although the work of the
Task Force was funded by the CIA’s Directorate of Intelligence, neither the Task Force’s analyses nor the contents of this report are based on
intelligence reporting. The report does not represent the official view of the U.S. Government, the U.S. Intelligence Community, or the
Central Intelligence Agency, but rather the views of the individual authors themselves.

ENVIRONMENTAL CHANGE & SECURITY PROJECT REPORT, ISSUE 5 (SUMMER 1999)
Special Reports
50
I. CONFIRMATION AND REFINEMENT OF THE GLOBAL MODEL
Updating the Problem Set and Revising the Control
Cases
One problem frequently encountered in statistical analyses
such as the one performed in the initial phase of the State Failure
project is that specific results may be highly sensitive to a
particular data set.
2
If the results reflect statistical accidents, rather
than underlying social and political forces, then slight changes
in the data set may greatly shift the results. Adding or subtracting
cases, or changing the particular control cases, could make some
variables newly significant or remove some variables from the
list of significant factors. Our first task in re-examining our
results was to update the problem set to include state failure
cases from 1994-96, and to select new control sets for testing
this new data, to make certain that our initial results proved
robust.
It was reassuring to find that despite significant revisions
and updating of the problem set and analyses using two different
sets of control cases and three distinct analytical techniques,
the same three variables—infant mortality, trade openness, and
level of democracy—emerged as the critical discriminators
between stable states and state failures. Moreover, these analyses
resulted in about the same two-thirds range of accuracy in
discriminating failures and stable cases.
State Failure Cases
3
The set of “state failure cases” in the initial State Failure
Task Force Report was updated and revised by reexamining all
of the cases and consulting area experts to identify recent events
(1994-96) for inclusion.
4
A number of cases in the initial problem
set were dropped as being of insufficient magnitude or not
meeting the precise definitions for failure events. A considerable
number of new cases from recent years were added. However,
none of these changes affected the global model results.
Control Cases
5
The two new sets of control cases were obtained, as before,
by randomly selecting to match every country-year that preceded
a state failure by two years, three countries that were stable
(experienced no crises for the succeeding five years). Changing
the control sets made no difference to any of the global
model results.
The three analytical techniques used were logistic
regression, neural network analysis, and genetic algorithm
modeling.
6
Logistic regression and neural network analysis were
used to estimate the “predictive” accuracy of our models. Genetic
algorithm modeling was used to help identify candidate sets of
variables, as a check on the univariate regression methodology,
and to validate the suggestions of Task Force social science and
area experts. Although each method relies on different
assumptions and methods of estimation, all techniques
converged on identifying the same three-factor model as the
most efficient discriminator between stable and failure cases
and yielded models with accuracy of predicting case outcomes
of about two-thirds.
Table 1: Historical State Crises, by Type
Type of Crisis Initial Phase Phase II
Revolutionary war 41
50
Ethnic war 60
59
Regime transition 80
88
Genocide and politicide 46
36
Total Number of
Consolidated Crises
113 127
Figure 1: Phase I Analytic Process
Identified State Failure and Control Cases
113 failure cases, including:
• Revolutionary wars
• Ethnic wars
• Genocides or politicides
• Adverse or disruptive regime changes
339 control cases, randomly selected
Selected Variables To Test for
Association with State Failure
Some 600 variables evaluated:
• Demographic and societal
• Economic
• Political and leadership
• Environmental
Total of 75 high-priority variables selected:
• Most likely to correlate with state failure
• Based on reasonably complete and reliable data
Applied Analytical Methods
Single variable tests identified…
31 variables best at distinguishing failures from
nonfailures
Analysis of combinations of up to 14 of these using:
• Statistical logistic regression analysis
• Neural networks analysis
Formulated Models
Single-best model relied on three variables:
• Infant mortality-indirect measure of quality of life
• Openness to international trade-value of imports
and exports divided by GDP
• Level of democracy-from information on political
institutions

Author ⋅ Title
51ENVIRONMENTAL CHANGE & SECURITY PROJECT REPORT, ISSUE 5 (SUMMER 1999)
State Failure Task Force
⋅
State Failure Task Force Report: Phase II Findings
Summary of Phase I Findings
The global model developed in the initial phase of the State Failure project and detailed in the task force report
a
had the
following features:
• It considered as “failures” four different kinds of political crisis—revolutionary wars, ethnic wars, adverse or disruptive
regime transitions, and genocides or politicides—of varying magnitudes.
• The model examined all such crises that occurred during the years 1957-94 in countries whose population in 1994 was
greater than 500,000 according to US Census Bureau data.
b
• The model compared conditions in countries that experienced crises at a time two years before the onset of a crisis with
conditions in a matched set of stable—or “control”—countries that did not experience a crisis any time in the succeeding
five years.
The global model was developed after examining hundreds of candidate factors suggested as theoretically relevant to state crises
and rigorously analyzing 75 variables that had been deemed highly relevant by experts and had global data available for most of
the 1955-94 period. The Task Force found that the most efficient discrimination between “failure cases” and stable states was
obtained from a global model with only three factors: the level of infant mortality, the level of trade openness, and the level of
democracy.
For this global model, a country’s infant mortality was measured relative to the world average level of infant mortality in a
given year (to correct for a long-term global decline in infant mortality rates). Trade openness was measured as the total value of
imports plus exports as a percentage of a country’s GDP. Countries were classified as either “More Democratic” or “Less Democratic”
(autocracies) on the basis of their level of institutional democracy.
Using these three variables, roughly two-thirds of historical failure and nonfailure cases could be accurately classified. In
addition, several interesting relationships among these factors were found:
• Although high infant mortality consistently appeared to be linked to state failures, we are certain that there is NO direct
causal connection between infant deaths and ensuing political crises. Instead, infant mortality appears to be acting
primarily as an indicator for the overall quality of material life. Like the canary in a coal mine, whose death indicates
serious health risks to miners, high infant mortality serves as a powerful indicator of more broadly deleterious living
conditions. This was clear since in some models, income level (real GDP per capita) worked almost as well as infant
mortality in predicting state failure. In addition, both infant mortality and GDP per capita could be replaced by a bundle
of health and welfare indicators, such as levels of nutrition, health care, and education with almost the same results. Infant
mortality plays a key role in the global model not because infant deaths per se are a causal factor, but because infant
mortality is the single-most-efficient variable for reflecting a country’s overall quality of material life.
• The effects of trade openness and infant mortality on risks of state failure were separate, not overlapping. Levels of trade
openness and infant mortality showed almost no relationship. They varied independently and operated independently to
affect state failure risks.
• Infant mortality had a much stronger impact on the risk of state failure in democracies, and had a relatively weak effect on
the risk of failure in less democratic countries. Trade openness showed the reverse pattern; that is, trade openness had a
stronger impact on the risk of state failure in less democratic countries and had a weaker, though still significant, impact on
failure risks in more democratic countries.
• Three additional variables were found to be important indicators for specific kinds of political crises, although they did not
emerge as important in the overall model. For adverse or disruptive regime changes, regime duration was a significant
factor. New regimes were found to have substantially higher risks of further adverse or disruptive changes in their earlier
years. For ethnic conflicts, both the ethnic character of the ruling elite and a youth bulge were found to be important
factors. Ethnic wars were most likely when a single ethnic group dominated the ruling elite; this was true whether the
dominant group came from a minority or majority ethnic group. In addition, the risks of ethnic war were greatly increased
by the presence of a “youth bulge”; that is, a large percentage of 15 to 29-year-olds relative to the population age 30-54.
a
See Esty, Daniel C., Jack A. Goldstone, Ted Robert Gurr, Pamela Surko, and Alan Unger. Working Papers: State Failure Task Force Report.
McLean, VA: Science Applications International Corporation, 30 November 1995.
b
Despite being over our population size cutoff, two countries were omitted: Eritrea (because data were not available) and Qatar. Two countries
with populations below 500,000 using US Census Bureau data—Comoros and Luxembourg—were inadvertently included. These deviations
from the rule did not contribute significant error, however, because the number of countries in the study was large.

ENVIRONMENTAL CHANGE & SECURITY PROJECT REPORT, ISSUE 5 (SUMMER 1999)
Special Reports
52
Retesting With a Refined Level of
Democracy Variable
The original global model, using infant
mortality, trade openness, and level of
democracy, measured democracy as a
dichotomous variable, classifying countries as
“more democratic” or “less democratic.”
However, it became apparent that not all
democracies were “equal” in their vulnerability
to state failure. The rich and well-established
democracies were extremely stable. In contrast,
the more recently established and poorer
democracies were at very high risk of failure.
Given this result, and the interests of
policymakers in democratic transitions, it was
clearly important to better differentiate the
democracy variable to examine the risks
associated with “partial democracies.”
Using both the democracy and autocracy
scales of the Polity III Global Data Set
7
, each
country was classified as a full democracy, a
partial democracy, or an autocracy, on the basis
of its political institutions:
8
• Full democracies have all the characteris-
tics of liberal democracy—such as
elections, competitive parties, rule of
law, limits on the power of government
officials, an independent judiciary—
and few or none of the characteristics
of autocracy.
• Partial democracies have some
democratic characteristics—such as
elections—but also have some
autocratic characteristics, such as a
chief executive with almost no
constraints on his/her power, sharp
limits on political competition, a state-
restrained press, or a cowed or
dependent judiciary. Most are countries
that have recently transitioned toward
democracy but have not yet fully
replaced autocratic practices and
institutions; some resemble what
Fareed Zakaria has referred to in a
recent Foreign Affairs essay as “illiberal
democracies.”
9
They are countries that
have adopted some democratic practices
but have not yet fully extinguished
autocratic practices in their government.
• Autocracies have various characteristics
of autocracy and few or none of the
characteristics of democracies.
Guarantees of political rights are essential
to institutionalized democracies, and most such
0
5
10
15
20
25
0
4
8
12
1955 1960 1965 1970 1975 1980 1985 1990 1995
Year
Number of New Failures
Percentage of Countries in Failure
Percentage
Number
Figure 4: Global State Failures: Revolutionary Wars, 1955-96
0 20 40 60 80 100 120 140
Number
Genocide/
politicides
Revolutionary
Wars
Adverse Regime
Transitions
Ethnic Wars
Total Cases
Figure 3: Number of Global State Failures by Type, 1955-96
Figure 2: Global State Failures, 1955-96
Number of New Failures
0
5
10
15
20
25
Percentage of Countries in Failure
0
5
10
15
20
25
30
1955 1960 1965 1970 1975 1980 1985 1990 1995
Percentage
Number
Year
Revolutionary Wars
with
Revolutionary Wars

Author ⋅ Title
53ENVIRONMENTAL CHANGE & SECURITY PROJECT REPORT, ISSUE 5 (SUMMER 1999)
State Failure Task Force
⋅
State Failure Task Force Report: Phase II Findings
polities guarantee civil rights to all citizens.
Therefore, while the democracy index is based
on an analysis of political institutions, it correlates
very closely (+.90) with Freedom House indices
of political rights and civil liberties.
Results
Using the trichotomized measure of
democracy, we discovered that partial
democracies are indeed far more vulnerable to
state failure–type crises than are either full
democracies or autocracies. To be precise, when
using this measure of democracy in the global
state failure model—along with infant
mortality and trade openness—to discriminate
between stable and crisis cases, we find that
partial democracies, other things being equal,
are on average three times more likely to fail.
This refined version of the global model
also confirms and makes more precise our
estimates of the impact of trade openness and
infant mortality (or overall quality of material
life) on failure risks. Using the updated
problem set, revised data, and new control
cases, we find that states with above-average
trade openness, other things being equal, have
one-half the failure risk of countries with
below-average trade openness. In addition,
countries with above-world median levels of
infant mortality have, other things being equal,
three times the risk of state failure as compared
with countries with below-median levels of
infant mortality.
II. FITTING A MODEL FOR SUB-SAHARAN
AFRICA
In the initial work of the Task Force, there
was some concern that grouping advanced
democratic nations and poor autocracies in
one global analysis was like comparing apples
and oranges. We have, therefore, applied our
analytic techniques to testing the model on
those crisis events and a matched set of control
cases, drawn solely from the countries of
Sub-Saharan Africa.
10
In addition to testing
all of the factors that emerged as significant
in the initial report, we also examined a variety
of additional factors that area experts
suggested as specifically relevant to Africa,
including a country’s colonial heritage,
conditions of ethnic discrimination, and
level of urbanization.
The model that most effectively
discriminated between crisis cases and control
Figure 5: Global State Failures: Ethnic Wars, 1955-96
Number of New Failures
0
5
10
15
20
25
Percentage of Countries in Failure
0
5
10
15
20
1955 1960 1965 1970 1975 1980 1985 1990 1995
Percentage
Number
Year
Figure 6: Global State Failures: Genocides and Politicides, 1955-96
0
5
10
15
20
25
0
4
8
1955 1960 1965 1970 1975 1980 1985 1990 1995
Year
Number of New Failures
Percentage of Countries in Failure
Percentage
Number
Figure 7: Global State Failures: Adverse or Disruptive Regime Changes,
1955-96
0
5
10
15
20
25
0
5
10
1955 1960 1965 1970 1975 1980 1985 1990 1995
Year
Number
of New
Failures
Percentage of Countries in Failure
Percentage
Number
with Ethnic Wars
Ethnic Wars
Percentage of Countries with Genocides/Politicides
Genocides/Politicides
Percentage of Countries with Regime Crises
Regime
Crises

ENVIRONMENTAL CHANGE & SECURITY PROJECT REPORT, ISSUE 5 (SUMMER 1999)
Special Reports
54
Table 2: Global Model Results
Changes to the List of Historical State Crises
The set of crises used in the analyses reported here consists of 127
“consolidated” cases of state failures, of a single type, and complex events
involving several different kinds of failure in sequence. This is 14 more
than in the initial study. The differences, as compared with the list in the
initial report, can be summarized as follows:
• Revolutionary wars. Examples of cases added are Islamist revolu-
tionary movements in Egypt (1986 to present) and in Algeria (1991
to present) and the revolutionary war that overthrew Mobutu’s re-
gime in Zaire (now Congo-Kinshasa) in 1996-97.
• Ethnic wars. Some ethnic rebellions from the original list were dropped
because they were of very low magnitude; others were consolidated into
other events. An example of a consolidated case is India, where mul-
tiple autonomy rebellions from 1952 to the present are treated as one
event. Some internal wars meet the criteria of both revolutionary and
ethnic wars, such as the civil war for control of the Afghan Govern-
ment (1992-97) fought by political movements based on the Pashtuns,
Tajiks, Uzbeks, and Hazaris.
• Adverse or disruptive regime transitions. A number of cases were
dropped and others added. Examples of recent failures of democratic
regimes now included in the data set are Albania 1996, Armenia
1994-96, Belarus 1995-96, and The Gambia 1994. Dates and
descriptions of a number of historical cases also were changed on the
basis of new and more detailed information.
• Genocides and politicides. No new cases since 1994 were identi-
fied, although indiscriminate attacks on civilians in Chechnya dur-
ing 1994-96 approached the threshold for politicide. The cases dropped
were ones in which killings of civilians did not, on closer examina-
tion, appear to be part of a systematic and sustained policy. For
example, killings of Kurdish civilians by Kurdish militants and the
Turkish military since 1984 are not numerous or widespread enough
to meet the definitional criteria.
In addition, the three lowest magnitude ethnic wars—Papua New
Guinea (Bougainville, 1988-97), Thailand (Malay Muslims, 1993-
present), and the United Kingdom (Catholics in Northern Ireland, 1969-
94)—were excluded from the global analysis of state failures because they
were considered too small to count as major events. They were, however,
retained in the data set for future study of ethnic conflicts.
a
Other conflicts categorized and counted as both revolutionary and ethnic
wars are Angola 1975-97, Ethiopia 1975-91, and Somalia 1988 to the present.
Trends and Patterns in State Failures
Some types of state failure are particularly likely
to lead to other failures, with several patterns emerging
from the analysis of discrete and complex cases:
• There is a substantial risk that internal wars—
revolutions and ethnic conflicts—will precede
other state failures. Of 50 revolutionary wars, 19
(38 percent) are the first event in a complex case
that subsequently included one or more adverse
regime transitions, ethnic wars, or genocides. The
percentage is higher for ethnic wars—44 percent
(26 of 59) of these are the first event in a complex
case.
• Adverse and disruptive regime transitions are less
likely than revolutionary or ethnic wars to lead to
other kinds of state failures. Nearly half (41 of
88) are discrete events; less than one-fifth (15 of
88) proved to be the first stage in a complex event.
• Genocides and politicides almost always are a con-
sequence of other kinds of state failure. Usually
the connection is clear-cut, for example, when an
authoritarian regime seizes power and sets out to
eliminate political opponents (as in Chile 1973-
76) or when revolutionary or ethnic challenges
prompt a regime to use extreme measures to rees-
tablish security (as in Indonesia against suspected
Communists in 1965-66 and against East Timor
nationalists after 1975). In 1996, the only ongo-
ing episode was in Sudan.
There also are distinctive trends in the onset and
frequency of each type of state failure. In the aggregate,
the number of states in failure increased up to the end
of the Cold War, but in the mid-1990s began to decline.
Revolutionary wars have declined in frequency;
whereas, ethnic wars have tended to increase, most
sharply so in the immediate aftermath of the Cold War.
Adverse and disruptive regime transformations, on the
other hand, have no distinct long-term trend but show
a sharp upward spike in the 1990s, mainly due to
failures of new and partially democratic regimes in
Africa and some of the post-Communist states.
Key Variables Countries at Greater Risk Countries
a
Material Living Standards Infant mortality above median Infant morta
l
Trade Openness
(imports+exports)/ GDP
Below median Above medi
a
Level of Democracy Partial democracies Autocracies;

Author ⋅ Title
55ENVIRONMENTAL CHANGE & SECURITY PROJECT REPORT, ISSUE 5 (SUMMER 1999)
State Failure Task Force
⋅
State Failure Task Force Report: Phase II Findings
cases in the Sub-Saharan Africa model had six significant
elements.
11
Level of Democracy
As with the general model, partial democracies were most
vulnerable to state failure. This result again showed a high degree
of statistical significance. However, while in the global model
full democracies and autocracies were about equally stable, in
Sub-Saharan Africa autocracies were slightly more stable than
even full democracies—presumably because in Africa full
democracies have greater problems managing ethnic conflicts
and fluctuations in material living standards than do the full
democracies of Europe and North America. In addition—and
this is one of our most striking results—we found that the
vulnerability of partial democracies to state failure was especially
great in Sub-Saharan Africa and much higher than in the world
at large. The precise results of this model were that in Sub-
Saharan Africa, other things being equal, partial democracies
Partial
Democracies
Bosnia and Herzegovina
Cambodia
Comoros
Congo, Republic of the
a
Ethiopia
Fiji
Georgia
Ghana
Guinea-Bissau
Guyana
Honduras
Jordan
Kyrgyzstan
Malaysia
Mexica
Moldova
Mozambique
Pakistan
Paraguay
Peru
Russia
Senegal
Sierra Leone
Slovakia
Sri Lanka
Tanzania
Yemen
Zambia
a
Congo (Brazzaville)
Autocracies
Afghanistan Nigeria
Albania North Korea
Algeria Oman
Angola Rwanda
Armenia Saudi Arabia
Azerbaijan Serbia and
Bahrain Montenegro
Belarus Singapore
Bhutan Somalia
Burkina Faso Sudan
Burma Swaziland
Burundi Syria
Cameroon Tajikistan
Chad Togo
China Tunisia
Congo, Democratic Turkmenistan
Republic of the
b
Uganda
Cote dIvoire United Arab Emirates
Croatia Uzbekistan
Cuba Vietnam
Egypt Zimbabwe
Gabon
The Gambia
Guinea
Indonesia
Iran
Iraq
Kazakhstan
Kenya
Kuwait
Laos
Lebanon
Liberia
Libya
Mauritania
Morocco
Niger
b
Congo (Kinshasa)
Full
Democracies
Argentina Lesotho
Australia Lithuania
Austria Madagascar
Bangladesh Malawi
Belgium Mali
Benin Mauritius
Bolivia Mongolia
Botswana Namibia
Brazil Nepal
Bulgaria Netherlands
Canada New Zealand
Central African Nicaragua
Republic Norway
Chile Panama
Columbia Papua New Guinea
Costa Rica Philippines
Cyprus Poland
Czech Republic Portugal
Denmark Romania
Dominican Republic Slovenia
Ecuador South Africa
El Salcador South Korea
Estonia Spain
Finland Sweden
France Switzerland
Germany Taiwan
Greece Thailand
Guatemala The Former Yugoslav
Haiti Republic of
Hungary Macedonia
India Trinidad and Tobago
Ireland Turkey
Israel Ukraine
Italy United Kingdon
Jamaica Uruguay
Japan Venezuela
Latvia
Figure 8: Countries by Level of Democracy, 1996
were on average 11 times more likely to fail than autocracies.
Full democracies were far less vulnerable; other things being
equal, they were on average more than twice as likely to fail
than autocracies.
Trade Openness
Trade openness is also confirmed as a highly statistically
significant correlate of state failure. The greater a country’s trade
openness, the less likely that country is to experience a major
state crisis. As in the global model, other things being equal,
countries in Sub-Saharan Africa that were above the median in
trade openness were on average only about one-half as likely to
fail as countries below the median.
Change in Material Living Standards
In the global model, which compared countries with a huge

ENVIRONMENTAL CHANGE & SECURITY PROJECT REPORT, ISSUE 5 (SUMMER 1999)
Special Reports
56
Key Variables
Key variables measure the following items:
• Infant mortality. Although this variable directly captures reported deaths to infants under one year old per thousand live
births, it also serves as an indirect measure of a host of broad-based material standard of living and quality-of-life indicators.
Infant mortality is strongly correlated with a variety of other variables encompassing economic performance, education,
social welfare, environmental quality, and democratic institutions.
• Trade openness. This variable is a ratio that measures the value of imports plus exports divided by GDP. Of the other
variables analyzed in the first phase of this project, it correlated only with the density of roads—generally accepted as an
indicator of economic development—and population size.
• Level of democracy. This variable is constructed from information on political institutions. Democratic regimes have
competitive political participation, elected chief executives, and significant checks on executives’ exercise of powers. The
variable correlates closely with indicators of civil and political liberties and also with measures of economic well-being.
• Regime durability. This variable is a count of the number of years since the last major, abrupt change in regime. Abrupt
shifts toward or away from democracy count as regime changes and reset the duration count to zero. Regime changes that
follow state breakdown and civil war also reset the count. Nonviolent transitions from one authoritarian regime to another,
or one democratic regime to another, do not register on this variable. Regime duration is correlated with several indicators
of economic development, including per capita GDP.
• Youth bulge. This variable is a ratio of the population in the 15- to 29-year age bracket relative to that in the 30 to 54-year
age group. It correlates with six other variables related to economic development and education.
• Ethnicity of the ruling elite. This is a variable that compares the ethnic composition of the ruling elite to that of the
population at large in an ethnically divided society. It indicates whether the elite demographically represents a minority
group, a majority group, or the population as a whole. It is uncorrelated with other variables in this study.
• Annual change in GDP per capita. This variable indicates the direction of recent changes in material welfare. It is
measured by taking the change in real GDP per capita from the previous year. Positive change indicates growth; negative
change indicates economic decline.
• Level of Urbanization. This variable measures the proportion of total population that is living in cities of 100,000
inhabitants or larger. It captures the degree to which a country’s overall population is urban.
• Colonial heritage. This variable compares the impact of French colonial heritage to the average impact of all other former
colonial powers in Sub-Saharan Africa. It has often been opined that different colonial powers left (or are still active in
creating) different degrees of political stability in their former colonies. While there are not enough cases to support statistical
tests for every past power—Britain, France, Germany, Italy, Belgium, and Portugal—on the recommendation of area
experts, we chose France as a country with many former colonies and a still active role in most of the period covered.
• Ethnic discrimination. This variable is derived from information on ethnic and religious groups that are economically or
politically disadvantaged because of present or past practices of discrimination by dominant groups. The indicator signifies
the existence of at least one politically significant communal group subject to significant discrimination or that sought
greater political autonomy from the state in which it was situated.
• Land burden. This variable is the number of farmers per unit of cropland multiplied by the ratio of farmers to the total
number of workers. It is highest in countries where a large proportion of the population is dependent on agriculture, but
arable land is limited.
• Deforestation rate. This variable measures the annualized rate of change in forest area from 1980 to 1990, using data
provided by the UN Food and Agriculture Organization.
• Telephones per capita. This variable measures the number of telephone lines in a country divided by the total population.
It is used as a rough indicator of a country’s capacity to respond to “shocks” or changes. A country’s ability to install and
operate a major physical infrastructure reflects, we believe, its general ability to “get things done.” In addition, there are
reasons to think that communication capacity is especially important for effective responses to environmental problems.
Telephones per capita correlates highly with per capita GDP level, although the annual changes in the two variables are not
correlated.
• Soil degradation. This variable combines information about the severity and extent of soil degradation within a country,
based on an assessment completed in 1990. The assessment utilized regional experts to estimate degradation over the
previous five to ten years.
• Population in subsistence agriculture. This variable is used to measure the degree to which a country’s population is
vulnerable to either deforestation or soil degradation. Subsistence agriculture is an activity that indicates high poverty and
high dependence on the health of terrestrial ecosystems. The percent of population dependent on subsistence agriculture is
inversely correlated with the level of GDP per capita, although the annual changes in the two variables are not correlated.

Author ⋅ Title
57ENVIRONMENTAL CHANGE & SECURITY PROJECT REPORT, ISSUE 5 (SUMMER 1999)
Number of New Failures
0
5
10
15
20
25
Percentage of Countries in Failure
0
5
10
15
20
25
30
35
1955 1960 1965 1970 1975 1980 1985 1990 1995
Percentage
Number
Year
Figure 9: Sub-Saharan Africa State Failures, 1955-96
State Failure Task Force
⋅
State Failure Task Force Report: Phase II Findings
Figure 10: Number of Sub-Saharan African State Failures by Type,
1955-96
0 10 20 30 40 50
Revolutionary wars
Ethnic wars
Adverse regime transitions
Genocide/politicides
Total cases
range of living standards, the level of material living standards—as measured by
infant mortality (or by GDP per capita or a basket of health and welfare measures)—
was found to be a powerful discriminator of risks of state failure. In the Sub-Saharan
African cases, where most countries are clustered at the low end of the scale of material
living standards, recent changes in living standards emerged as a stronger indicator of
failure risks than did absolute levels. In particular, other things being equal, countries
that had experienced a negative annual change in GDP per capita were on average twice as likely to experience a serious political crisis
two years later than countries that had had a positive change in GDP per capita.
12
Colonial Heritage
The Task Force—along with Sub-Saharan Africa regional experts—discussed the possibility that differences in colonial
heritage affect vulnerability to state failure. Although states of all varieties of colonial background did experience problems, the
data showed that, holding other factors equal, former French colonies on average had only one-third the risk of failure of other
African countries. This was a firmly statistically significant result. However, we note that until recently France has also maintained
a higher level of engagement—political, financial, and military—with its former colonies than most other powers. As these levels
of engagement decline, it may well be that French colonial heritage will become less significant as a moderating factor in regard
to state crises.
Variables Tested for the Sub-Saharan
Africa Model
Economic
Trade openness
GDP per capita
Change in GDP per capita
Land burden
Urban population
Population density
Change in reserves
Political/leadership
Separatist activity
Democracy
Change in democracy level
Economic discrimination
Political discrimination
Ethnic discrimination
Party fractionalization
Parliamentary responsibility
Party legitimacy
Demographic/societal
Youth bulge
Colonial heritage
Labor force
Annual change in employment
Secondary school enrollment ratio
Ethno-linguistic fractionalization
Amnesty International political terror scale
US State Department political terror index
Government repudiation of contracts
Risk of expropriation
Agricultural
Cropland area
Irrigated land
Population in agriculture
Population in subsistence agriculture
Energy
Commercial energy use
Commercial energy production
Percentage of African Countries in Failure
in Africa

ENVIRONMENTAL CHANGE & SECURITY PROJECT REPORT, ISSUE 5 (SUMMER 1999)
Special Reports
58
Level of Urbanization
Although the absolute level of GDP per capita was
not a significant predictor of state failure, when combined
with the level of urbanization—as measured by the
proportion of population living in urban areas—the impact
was statistically significant. Having a high proportion of
urban population increased the risk of political crisis only
in countries whose GDP per capita was below the average
for Sub-Saharan Africa. Among such low GDP per capita
nations, the risk of failure was twice as high as for countries
with higher levels of urban population.
Interestingly, the effect of the share of population in urban
areas on failure risks becomes favorable in countries with higher
levels of GDP per capita. Other things being equal, for countries
that had—by Sub-Saharan African standards—above average GDP
per capita, those that also were above average in their proportions
of urban population were only one-fifth as likely to fail as those
that had lower levels of urbanization. In sum, countries with either
high GDP per capita and higher levels of urbanization—relative to
other Sub-Saharan African countries—or low GDP per capita and
low urbanization were more stable; it was only when relative levels
A Pilot Event-Data Analysis
The general models of state failure identify risk factors associated with serious political crises, but they are less useful in
forecasting outcomes for individual countries. To better understand the factors that might precipitate a failure in a high-risk
country during the two-year period before a crisis, the Task Force conducted a pilot analysisª of events in twelve Sub-Saharan
African cases—four ethnic wars, four regime crises, and four control cases—since the mid-1980s.
b
We used the Global Events Data
System at the University of Maryland—which relies on Reuters’ international wire service—to track daily events over a period of
two years before the onset of state failure (or, for the control cases, during a two-year period in which no state failure occurred) and
to identify:
• “Accelerators.” Feedback events that affect the general conditions underlying conflict development, which also have a
cumulative interaction effect that may increase escalation.
• “De-accelerators.” Events such as negotiations and policy reforms that are likely to de-escalate a crisis.
• “Triggers.” Events that are likely to propel a high-risk situation to the next phase of crisis escalation.
Based on previous analyses using this approach, we would expect to observe an increase in the number and severity of
accelerator events shortly before the onset of state failure.
The method analyzes political events over time, with separate models for ethnic warfare and regime crisis. Examples of
accelerators of ethnic warfare are “attacks on or threats to core symbols of ethnic group identity” and “external support for communal
group objectives from international actors.” For example, whereas external support for communal groups is typically thought to be
a factor that facilitates conflict escalation, tracking accelerators allows us to trace the ebb and flow of the types, quality, and
quantity of support over time.
On the basis of the pilot study, the Task Force concluded that the results of the analysis are sufficiently interesting to merit
further study. Although the sample size was too small for rigorous statistical analysis, the time clusterings of events for countries in
crisis were more similar to those of other countries in crisis—either regime crisis or ethnic war—than to countries not in crisis,
C
suggesting that further analysis by methodologists and area experts may be fruitful. A side benefit of the analysis was that it allowed
the start dates of four of the crises to be adjusted, because—based on the sequence of daily events—some of the crises apparently
began either earlier or later than the Task Force had previously specified in the list of historical crises. In general, the pilot study
results demonstrate that monitoring accelerators is a potentially powerful analytic tool that allows analysts to observe the development
of crises in high-risk countries in fine-grained steps, rather than being constrained by the limitations of yearly data.
The graphic illustrates the pattern of accelerators in former Zaire, a case of ethnic war beginning in April 1992.
d
It shows a
gradual buildup of events from April 1990 to a peak in October 1991, but deaccelerators seem to check complete breakdown up to
that point. Accelerators of ethnic warfare (disunity with the elite and elite responses to threats) reach a high level from January to
March 1992.
a
The accelerators approach used here is derived from a study of the accelerators of genocide and politicide reported by Barbara Harff, “Early
Warning of Genocide: The Cases of Rwanda, Burundi, and Abkhazia.” In Ted Robert Gurr and Barbara Harff, Early Warning of Communal
Conflicts and Genocide: Linking Empirical Research to International Responses. Tokyo: United Nations University Press, Monograph
Series on Governance and Conflict Resolution No. 05, 1996.
b
For a description of the cases, see appendix B.
c
See appendix B for details on the method.
d
For details on the Liberia case, see appendix B, figure B-1: Liberia: Regime Crisis Case.

Author ⋅ Title
59ENVIRONMENTAL CHANGE & SECURITY PROJECT REPORT, ISSUE 5 (SUMMER 1999)
State Failure Task Force
⋅
State Failure Task Force Report: Phase II Findings
of urbanization were “out of balance” with relative levels of economic
development that political risks increased.
This confirms the bimodal effect of urbanization on political risks
described by Jack Goldstone in his work on early modern European states;
13
namely, that if the economy is doing well, and urbanization takes place in
the context of good employment opportunities, then migrants to cities are
socialized into an urban context that they view as rewarding hard work
and promising a better future. This is politically stabilizing. In contrast, if
the economy is doing poorly and urban migrants find poor opportunities
for employment, then migrants are socialized into an urban context that is
frustrating and that they view as hostile and unresponsive. This situation
greatly aggravates the risks of political crisis.
Ethnic Discrimination
The presence of communal groups that are subject
to significant economic or political constraints appears
to increase the risks of political failure, all other things
equal, by almost a factor of two. However, this result
was only weakly statistically significant and should be
viewed as suggestive rather than conclusively
demonstrated.
The Sub-Saharan Africa model had roughly the
same accuracy as the global model—about two-thirds—
in discriminating between state failure and stable cases
14
but resulted in substantially reduced “false positives”
for Sub-Saharan African countries.
15
III. TRANSITIONS TO DEMOCRACY AND
AUTOCRACY
Trends
Institutionalized democracies have increased
significantly in number since the late 1980s. At the
end of the Cold War, the number of full democracies
in the world system exceeded the number of autocracies
for the first time since World War II. As of 1991, full
democracies numbered 57, compared with 55
autocracies. By 1996 the number of full democracies
Table 3: Sub-Saharan Africa Model Results
Key Variables Countries at
Greater Risk
Countries at
Lesser Risk
Relative Risk of
Failure
Material Living
Standards Change
Negative annual GDP
per capita change
Positive annual GDP
per capita change
2.0
Trade Openness
(imports+exports)/
GDP
Below median Above median 1.9
Level of Democracy Partial democracies Autocracies 11.0
Full democracies Autocracies 2.6
Level of Urbanization High urbanization and
low GDP per capita
Low urbanization and
low GDP per capita
2.0
Low urbanization and
high GDP per capita
High urbanization and
high GDP per capita
4.9
Colonial Heritage Not French French 2.6
Ethnic Discrimination Higher Lower 1.9
Figure 11: Former Zaire Ethnic Conflict (Accelerators, De-accelerators, and Triggers)
-4
-2
0
2
4
6
8
10
12
Ap
Ma
Ju
Jul
90
Au
Se Oc No De Ja Fe Ma Ap
Ma
Ju
Jul
91
Au
Acc. 8: Opposition by Kindred Groups in Neighboring Countries
Acc. 7: Responses by State Elites to Perceived Threats from Domestic Challengers Short of Open Rebellion
Acc. 6: Disunity within the State Elite, Conflict and Inefficiency in the Conduct of Routine Government
Acc. 5: Increase in Symbolic, Political, or Military Support for Communal Group Objectives from International Actors
v
Acc. 4: Increase in Symbolic or Political Support for Group Objectives from Domestic Actors
Acc. 3: Increases in the Disposition and Capacity of Elements within the Group to use Force and Violence in Pursuit of their Objectives
Acc. 2: Qualitative Changes in Demands Made on Behalf of an Ethnic Group
Acc. 1: Attacks on or Threats to Core Symbols of Ethnic Group Identity
Trigger: Single Event Capable of Negatively Changing the Direction of a Crisis (Usually Conflictive)
De-Accelerator: Single Event Capable of Defusing a Crisis (Usually Cooperative)
0
Ma
r
90
y
97
n
90
g
90
p
90
t
90
v
90
c
90
n
91
b
91
r
91
r
91
Mar
91
Sept
91
Jan
92
Dec
91
Apr
92
Mar
92
Feb
92
Nov
91
Oct
91
g
91
n
91
y
91
➞
Crisis Begins

ENVIRONMENTAL CHANGE & SECURITY PROJECT REPORT, ISSUE 5 (SUMMER 1999)
Special Reports
60
had increased to 71, whereas autocracies had declined to 49.
The post–Cold War transition—which Samuel Huntington calls
“the third wave of democratization”
16
—also has seen the
establishment of a large number of partial democracies. In 1996
there were 27 such polities, double their numbers in the 1980s.
The long-run trend by which democracies have come to
outnumber autocracies has two sources. One is the significant
number of new democracies established in the post-Communist
states. The other, and more important factor, is that many
countries that tried and failed to establish democratic polities
tried again. South Korea, for example, shifted from autocracy
to full democracy in 1960, but a year later lapsed back to
autocracy. In 1963 it shifted again to partial democracy but
returned to autocratic rule in 1980. South Korea’s most recent
transition began in the mid-1980s and was completed in 1988
when it became, and has thus far remained, a full democracy.
In short, South Korea accounts for three transitions toward
democracy and two cases of backsliding to autocracy. Pakistan,
Turkey, Thailand, and Bangladesh—all full or partial
democracies by 1997—also made three or more democratic
transitions between 1955 and 1996.
Transitions are defined in terms of shifts among the three
categories of regime type—full democracy, partial democracy,
and autocracy. For the analysis of trends, the Task Force defined
transitions to democracy as shifts from autocracy to either partial
or full democracy as well as shifts from partial to full democracy.
17
These transitions are said to be stable if the regime does not
regress toward autocracy in the first five years after the initial
transition.
18
A regime is unstable if it regresses toward autocracy
within five years. Thus, a country that changes from autocracy
to partial democracy, then two years later transitions from partial
to full democracy, is counted as having made one stable
transition. A country changing from partial democracy to
autocracy and remaining an autocracy for five years is counted
as a stable downward transition; whereas a country that shifts
from democracy to autocracy, then within five years returns to
partial democracy, would be counted as an unstable downward
transition.
Four major observations can be made about the evidence:
• Many democratic transitions do not endure. Between
1957 and 1991 there were 54 durable transitions— that
persisted for at least five years—toward full or partial
democracy in independent countries, including 16
democracies established during the period 1989-91 in the
Soviet and Yugoslav successor states. Another 20 democratic
transitions were attempted between 1957 and 1991 but
reverted to autocracy during their first five years. An
additional 33 democratic failures—durable democracies
that shifted toward autocracy for at least five years—
occurred.
• Post–Cold War democratic transitions may be more
durable than earlier ones. Before 1986, 24 regimes made
durable transitions toward democracy, more than offset
by 44 failures—reversion to autocracy—of full or partial
democracies.
19
The 38 durable transitions toward democracy
between 1986 and 1991, however, were offset by only nine
failures. A more precise comparison looks only at the
outcome of democratic transitions that were attempted
between 1957 and 1991. Of the 36 transitions that
occurred before 1986, 12 countries (33 percent) reverted
to autocracy within five years; whereas, for the 38
transitions in 1986 or later, only eight (21 percent) failed
to survive. The short-term survival of democratic transitions
thus appears to have increased slightly in the post–Cold
War period, although the difference is not quite statistically
significant.
• World regions differ substantially in the success of
democratic transitions. Before 1986, Africa south of the
Sahara had only one durable democratic transition and
the record in Asia was only slightly better. In Latin America
and the Caribbean, half of the pre-1986 transitions endured
to early 1997. The success rates of recent democratic
transitions are highest in Asia—where Cambodia is the
only recent democratizing regime to backslide (in 1997)—
and in Latin America. Despite a great deal of concern about
the durability of the post-Communist states, 14 of the 19
that became partial or full democracies during 1989-91
have maintained democratic regimes. The exceptions are
Azerbaijan and Armenia—where democratic governance
was undermined by civil war—and Belarus, Kazakhstan,
and Albania where it was subverted by chief executives
who dissolved or emasculated legislatures that constrained
their power.
• Partial democracies are less durable than either autocracies
or full democracies. There are inherent political
contradictions in most partial democracies—a tension
between demands for greater and more effective
participation on the one hand, and the desire of political
elites to maintain or enhance their control. Most partial
democracies transition within a decade or so either to full
democracies or revert to autocracy.
Region Total Transitions,
1957-1991
Percent That Survive
for
Five Years or More
Europe 14 93
Latin
America
24 83
Newly
Independent
States
12
1
67
Asia 14 64
Africa 10 40
TOTAL 74 73
Table 4: Democratic Transition Success Rates, by Region
1
Uzbekistan, Turkmenistan, and Tajikistan did not make initial
transitions to democracy.

Author ⋅ Title
61ENVIRONMENTAL CHANGE & SECURITY PROJECT REPORT, ISSUE 5 (SUMMER 1999)
State Failure Task Force
⋅
State Failure Task Force Report: Phase II Findings
Models
In developing statistical models of transitions, the Task Force used a
narrower definition of transition than it did for the analysis of trends.
20
Because
crossing the autocracy-democracy divide was thought to be the more critical
transition, and because the number of shifts between partial and full democracy
was relatively small, the Task Force decided to limit its statistical analysis to
transitions from autocracy to partial or full democracy and those from partial
or full democracy to autocracy. In this analysis, models were developed that attempted to answer two research questions:
• What social, economic, and political conditions differentiate countries that make durable democratic transitions from others?
• What conditions characterize countries in which democratic regimes fail to succeed?
These questions are different from the issue of the conditions of “state failure” because the democratic transitions are defined
and measured differently from state failures. Moreover, few transitions from autocracy to democracy, and only about half of the
transitions from democracy to autocracy, meet the criteria of adverse regime transitions.
Transitions from Autocracy to Democracy.
21
A total of 39 transitions to democracy were available for analysis and were matched with 68 control cases—autocracies in the
same region that did not shift to democracy during the matching years.
22
Experts examined the state failure database to identify
variables that they thought should contribute to democratic transitions, and statistical tests were used to determine which of them
differentiated significantly between the transitions and the controls.
Then various combinations of these variables were analyzed to determine the most efficient set. From more than 60 models
analyzed, the one with the highest accuracy included two variables: relatively low land burden—an index that is highest for
Table 5: Democratic Transition Model Results
Variables Tested for the Democratic
Transition Models
Demographic
Infant mortality, normalized
Secondary school enrollment ratio
Youth bulge, normalized
Annual change in infant mortality
Political/leadership
Ethnic character of ruling elite
Years national leader was in office
Regime durability
Democracy minus autocracy index
Autocracy index
Regime duration
Political rights
Civil liberties
Economic
Real investment share of GDP, normalized
Trade openness
Land burden
Real GDP per capita, normalized
Autocracy to Partial or Full Democra
c
Key variables Countries More Likely
To Transition
Countri
e
To T
Regime durability Below median Abov
e
Land burden Below median Abov
e
Partial or Full Democracy to Autocra
c
Key variables Countries More Likely
To Transition
Countri
e
To T
Infant mortality, normalized Above median Belo
w
Regime durability Below median Abov
e
1955 1960 1965 1970 1975 1980 1985 1990 1995
10
20
30
40
50
60
70
80
Autocracies
Partial Democracies
Democracies
Percent
Year
Figure 12: Democracy Trends, 1955-96
Full
Number of Countries
Full Democracies

ENVIRONMENTAL CHANGE & SECURITY PROJECT REPORT, ISSUE 5 (SUMMER 1999)
Special Reports
62
countries with largely agricultural populations and scarce
cropland—and low durability of the regime before the
transition. This model correctly classified two-thirds of the cases
in a set of 39 transitions and 68 controls. The best three-variable
model correctly classified two-thirds of the cases and showed
that durable democratic transitions were most likely when infant
mortality was relatively stable, autocracy was already restricted,
and land burden was low.
These models suggest some interesting substantive findings.
The regimes most likely to undergo stable democratic transitions
during the last 40 years:
• Already had shifted away from purely autocratic forms of
government.
• Tended to have had less durable regimes; that is, they had
attempted previous political experiments.
Transitions were also more likely to occur in societies with
greater economic capabilities (measured by low land burden)
and less short-term variability in quality of life (measured by
changes in infant mortality).
Once a country has transitioned to democracy, the Task
Force found that the likelihood that the transition will be stable
depends on several factors:
• Countries whose democratic transitions are most likely to
succeed have greater annual improvement in infant
mortality, a lower level of infant mortality, greater trade
openness, a higher proportion of the population in urban
areas, and more years of experience as a democracy.
Transitions from Democracy to Autocracy.
23
A total of 35 democratic failures—transitions from full or
partial democracy toward autocracy—were available for analysis
and were matched with 98 control cases;
24
that is, democratic
countries in the same region that did not fail during the
matching years. The two-variable model with the highest
accuracy—nearly three-quarters of cases correctly classified—
included infant mortality normalized by world average and
regime durability. High infant mortality and low regime
durability characterized transitions to autocracy.
It is not surprising that newer democracies—those of low
durability—are more likely to fail than long-lived ones, based
on the evidence that many democracies fail during their first
five years. The role of infant mortality—and by extension, other
aspects of quality of life—in raising the prospects for democratic
survival is consistent with the results of the general models of
state failure.
IV. THE ROLE OF THE ENVIRONMENT IN STATE FAILURE
Goals and Hypotheses
We set out to determine whether the proposition that there
is a measurable connection between environmental degradation
Investigating Links Between Conflict and the Environment
The efforts reported here build on a thriving set of research programs at a variety of institutions investigating the environment’s
role in violent conflict. Early hypotheses centered on environmental degradation and resource depletion directly precipitating
violent conflict. Two major sets of case studies in the 1990s suggested that environmental causal pathways to conflict were more
complicated. Environmental variables—which alone were neither necessary nor sufficient to cause conflict—were found to play
multiple roles along a complex causal chain involving intervening social, political and economic variables.
• Dr. Thomas F. Homer-Dixon of the University of Toronto found that when “environmental scarcity” of renewable resources
did play a causal role, it was most likely to be through impacts that were “sub-national, persistent, and diffuse.” These impacts
indirectly contributed to acute conflict by exacerbating more familiar sources of conflict—for example, ethnic divisions or
relative deprivation.
• Drs. Guenther Bachler and Kurt Spillman, codirectors of the Swiss Environmental Conflicts Project (ENCOP), identified
seven types of “environ-mentally-induced conflict” in a typology that distinguished levels of conflict and parties to conflict.
ENCOP case studies also highlighted “environmental conflicts” as traditional conflicts “induced by environmental degradation.”
As the number of case studies accumulated through these projects and other efforts such as those at the International Peace
Research Institute, Oslo, and Columbia University, it became clear that intervening “institutional capacities,” or coping mechanisms,
to address environmental challenges were critical in determining whether conflict would occur.
Until very recently, a gap in the research program has been the use of statistical analysis examining a large number of
countries over time. The need for this kind of study is made clear by the highly qualified conclusions that the case studies produced.
The work of the State Failure Task Force is one of only two such studies undertaken to date, the other being the work of Hauge and
Ellingsen. In addition, ours is the only study to explore systematically the interactions between environmental change, vulnerability,
and capacity in this context, and the only study to use quantitative measures to attempt to uncover these relationships.
a
a
See appendix D for selected bibliography.

Author ⋅ Title
63ENVIRONMENTAL CHANGE & SECURITY PROJECT REPORT, ISSUE 5 (SUMMER 1999)
and state failure was true. Our goals were to:
• Test the argument with data drawn from all countries, over
an appropriate time period. Although a number of scholars
in recent years have claimed that there is a connection
between environmental degradation and political violence,
these claims have been largely based on individual case
studies.
25
These individual studies, albeit largely of high
quality, fail to rigorously test the correlative claim.
• Determine whether it was possible to offer analytical
guidance to decisionmakers as they face new security
challenges. US policymakers—in the State Department,
National Security Council, Defense Department, and other
agencies—have increasingly framed environmental issues
in security terms.
26
No clear consensus exists, however, as
to what kinds of environmental changes are most important,
what factors make a given level of environmental change
more or less dangerous, or what types of policy interventions
are most promising.
• Construct a specific model, and test it with empirical data,
to provide the foundation for monitoring and forecasting
potential trouble spots, where environmental deterioration
could potentially enhance the likelihood of state failure.
Two primary expectations guided our analysis:
• We did not expect to find any direct, measurable
correlation between environmental change and state
failure. Although this expectation is at odds with some of
the literature,
27
we were guided by the following logic: models
of environmentally induced political violence all include
numerous intervening variables that are held to interact in
a complex fashion.
28
The large number of intervening
variables makes it hard to find strong direct relationships
between the environment and state failure. The complex
interaction means that whatever relationships do exist are
likely to be different from case to case. As a result, the linkages
between environmental change and state failure are unlikely
to be discovered by simply adding environmental variables
to a state failure model.
29
• We did expect that environmental change might have a
significant, negative impact on one of the factors
associated with state failure in the general model. In
particular, we sought to explore whether environmental
degradation would have an impact on quality of life
measures such as infant mortality. If so, then this would
demonstrate an important, though indirect, connection
between environmental degradation and state failure.
Analytically, we conceived of the factors interacting in the
following manner: a given change in environmental conditions
generates an impact on a society that varies according to the
underlying environmental conditions—a society’s
vulnerability—and which is mediated by a nation’s capacity to
respond effectively. Where capacity is high, harm will be avoided.
To illustrate, consider crop yields as the impact and drought
as the environmental change. Vulnerability is the degree to which
crop yields might be expected to fall in the absence of effective
intervention. It might be measured through extent of irrigation
or sensitivity of crops to rainfall. Capacity is the degree to which
the government and social actors are able to lower the actual
impact, and might be measured as the size of the government
budget, number of scientifically trained experts, or extent of
communications infrastructure.
To be even more concrete, for the 1991-1992 growing
State Failure Task Force
⋅
State Failure Task Force Report: Phase II Findings
Figure 13: Mediated Environmental Model
The relationship can be expressed as
Impact = a function of (environmental
change, vulnerability, and capacity), where
• Environmental change is a change
in environmental resources.
• Vulnerability is the magnitude of
the potential impact per unit of
change in environmental
conditions.
• Capacity is the ratio of actual
impact to potential impact.
This formula provides an analytic
framework for understanding the key
relationships among environmental
change, vulnerability, and capacity for response. Given data limitations and the lack of appropriately denominated indicators,
we tested a simple formulation of the model in which environmental impact is a linear function of environmental change,
vulnerability, and capacity.
Material well
being (quality of
life)
- infant mortality
- income loss
- epidemics
Material well
being (quality of
life)
- infant mortality
- income loss
- epidemics
Environmental
Change
- deforestation
- soil erosion
- water loss/
contamination
Environmental
Change
- deforestation
- soil erosion
- water loss/
contamination
Capacity
- skills
- resources
- personnel
Capacity
- skills
- resources
- personnel
Likelihoo
d
State Fail
u
Likelihood
State Fail
u
Trad
Openn
Trad
e
Openn
e
Democratization
Democratization
Vulnerability
- agricultural dependence
- proximity to critical
thresholds (e.g. water)
Vulnerability
- agricultural dependence
- proximity to critical
thresholds (e.g. water)

ENVIRONMENTAL CHANGE & SECURITY PROJECT REPORT, ISSUE 5 (SUMMER 1999)
Special Reports
64
season, El Niño–driven droughts were forecast for northeastern
Brazil and for Zimbabwe, with more or less equivalent lead
times given to decisionmakers and a comparable projected and
actual change in environmental resources—rainfall. The
vulnerability—the potential drop in agricultural production
divided by loss in rainfall—was also about the same. However,
the net social impact, or actual loss in output, was very small in
Brazil but quite high in Zimbabwe, where 80 percent of the
maize crop was lost. Many analysts attribute this difference to
different levels of capacity in the two settings. Officials in Brazil
acted on the knowledge early, implementing effective strategies,
whereas in Zimbabwe the information was never used, and no
responsive strategies were developed.
30
Findings
Environmental change does not appear to be directly
linked to state failure. To determine whether it was possible to
find a statistical correlation between environmental change and
state failure, we tested variables that measured deforestation
and freshwater supply, but both failed to generate significant
results. This was consistent with our hypothesis that the more
direct effects of democratization, trade openness and quality of
life—measured by infant mortality—had such a strong impact
on state failure that they masked any impact of environmental
deterioration.
This result is at odds with recent work by Hauge and
Ellingsen,
31
the only other study we are aware of that employs
statistical tests to evaluate claims about the direct impact of
environmental harm on political violence. Hauge and Ellingsen
found a significant impact from deforestation, soil degradation,
and freshwater access, results that we believe are due to
differences in how the dependent variables are operationalized
and how the independent variables are used. Some of these
differences are potentially large enough to account for the
different results by themselves; taken together they make the
two models essentially incomparable. Because the state failure
model covers a greater time period and includes trade openness
as an explanatory variable, we think its results have more validity.
Nevertheless, the Hauge and Ellingsen model shows that there
is more than one way to approach these questions, and we
welcome the opportunity for scholarly debate.
Environmental change is significantly associated with
changes in infant mortality. To investigate the merits of the
mediated model, we assembled data on environmental change,
vulnerability, and state capacity. Because of data limitations,
we limited our scope to the period 1980-90; extending the time
frame back further would have seriously reduced the number
of countries and variables available for testing.
We chose infant mortality as the dependent variable
because of the availability of data, the high significance of infant
mortality as a factor associated with state failure, and the high
correlation of infant mortality with a number of other measures
of material well-being. We would have preferred to use a basket
of indicators that captured the level of material well-being or
quality of life, but the only well-being indices we located covered
too few countries, spanned too few years, or included factors
that were not relevant to our analysis.
Table 6: Hague and Ellingsen and the State Failure Study: Differences
Study
Hague and Ellingsen
Operationalization of Dependent
Variables
Definition of failure Used incidence of civil
w
in one model; armed
conflict in another
Overall time period 1980-1992
Unit of observation1 Each year of civil war or
armed conflict
Use of Independent Variables
Treatment of deforestation variable Categorized
Range of variables included Some overlap with State
Failure, but nothing
analogous to trade open
n
1
This is a major difference. The State Failure Task Force chose to develop a model th
a
outbreak of state failure. Hauge and Ellingsen's model, in effect, combines out break
a
asked not only to estimate the likelihood of when a civil war will start, but also when
1

Author ⋅ Title
65ENVIRONMENTAL CHANGE & SECURITY PROJECT REPORT, ISSUE 5 (SUMMER 1999)
Once the data were assembled, we screened potential
capacity and vulnerability variables by computing their
correlation with infant mortality. Those that were significantly
correlated—telephones per capita, population in subsistence
agriculture, and land burden—were then tested in combination
with an environmental stress variable in a multiple linear
regression model.
32
As we expected, deforestation proved to be statistically
significant only when tested in a model that included measures
of vulnerability and capacity. For given levels of vulnerability,
capacity, and baseline infant mortality rates, we found that the
greater the loss of forest cover, the higher the increase in infant
mortality rate.
The results for the model using soil degradation as the
environmental stress were more complex, and no linear
relationship could be measured. We obtained significant results,
however, by multiplying the rate of degradation by its severity
and including it as an interactive term. The results suggest that
soil degradation has a negative impact when severe degradation
occurs at a rapid rate; otherwise the impact is positive. One
possible interpretation of this finding is that the same practices
that induce soil degradation—such as agricultural production—
might have a positive net impact, for example, by improving
nutrition or incomes, if the degradation does not proceed too
rapidly.
33
Insights
One major insight that emerges from the analysis is that
available measures of environmental degradation do not
currently serve as a direct signal of impending state failure. In
part, this is a function of the long, complex chain of association
between environmental change and state failure, with a number
of factors intervening along the way. Those factors are strong
enough to push some societies blessed with benign
environmental conditions into failure and to prevent other
societies suffering serious environmental damage from slipping
into political instability. This finding is also a function of the
State Failure Task Force
⋅
State Failure Task Force Report: Phase II Findings
Variables Tested for the Environmental Model
Environmental Change
Deforestation
Soil degradation
Change in agricultural land
Access to fresh water (urban, rural, and total
population)
Fraction of freshwater reserves withdrawn
Sulfur dioxide emissions
Population density
Vulnerability
Percent of population engaged in subsistence
agriculture
Land burden: (farmers per area of cropland) x
(farmers per labor force)
Storm damage
Share of national income by lowest 20 percent of
population
Capacity
Secondary school enrollment ratio
Adult female literacy
Public expenditures on education
Telephone lines per capita
Bureaucratic quality
Corruption
Number of bribery cases
Law and order tradition
GDP per capita
Debt service
Rail mileage per square mile
Rail-ton miles per capita
Road density
Environmental Data Limitations
Our analysis was seriously constrained by the paucity of
available data. Whereas the overall state failure model was
able to test some 75 economic, political, and demographic
variables, the environment model could test only a handful.
This data constraint meant that some important
environmental factors could not be examined. For example,
water quality—consistently mentioned in the literature as
the most serious environmental problem facing developing
countries—could not be included because reliable time series
data are available for only 38 countries.a Air quality suffers
from similar deficiencies.
Useful indicators of vulnerability were also scarce.
Because the best environmental change indicators—
deforestation and soil degradation—that we had were related
to terrestrial ecosystems, we were able to rely on vulnerability
measures that tapped the degree of sensitivity to agricultural
perturbations. However, measures relevant to other
environmental shocks, such as declines in air quality, would
have been harder to construct.
The available measures of capacity were especially
disappointing. The ideal measure, in our view, would take
into account the financial resources, quality and extent of
infrastructure, and knowledge and skills of public and
government officials available for monitoring, assessing, and
responding to major environmental problems. Despite the
great attention paid to issues of capacity building in recent
years,
b
we were unable to identify any useful indicators that
came close to capturing this concept and, instead, had to rely
on proxies that imperfectly measured a few aspects of capacity.
a
Even for these countries, data are taken from single-point
monitoring stations.
b
See, for example, the UN Development Programme’s Capacity
21 program.

ENVIRONMENTAL CHANGE & SECURITY PROJECT REPORT, ISSUE 5 (SUMMER 1999)
Special Reports
66
seriously limited data at our disposal. On balance, we cannot say
how large an impact environmental damage has on the risk of
state failure.
Nevertheless, the results of our analysis provide evidence
for an indirect connection between environmental change and
state failure. Deforestation and soil degradation appear to
diminish the quality of life, as measured by infant mortality
rates, for low-capacity states that are socially vulnerable to
disruptions in soil ecosystems; and infant mortality has been
shown to have a direct impact on the likelihood of state failure.
Caveats on the Findings
While we believe that the results of the mediated
environmental model are useful and significant, the model has
several limitations:
• The process of converting analytic concepts into
measurable variables has necessarily resulted in variables
that are more narrow and arbitrary than the analytic
constructs that they represent. This is most true for our
core capacity variable—telephones per capita, which we
recognize to be a very limited measure of governmental
and societal response capability—but to a degree it is true
for all the variables.
• The findings represent a general tendency that applies to
the set of all countries for which data were available, over
the ten-year period studied. That does not mean that this
tendency will be true for each individual country at every
point in time. Some countries might experience far more
direct connections between environmental change and state
failure than we observe; other countries might experience
less connection between environmental change and infant
mortality than our results suggest.
• Environmental data limitations mean that our conclusions
are far from the last word. We simply did not have measures
for some very important environmental changes—
including water quality, with its impact on public health—
that might prove more significant as precursors of state
failure than those we tested. Data constraints also prevented
us from testing whether state failure is associated with
aggregate processes of environmental deterioration,
encompassing the degradation of soil, air, and water
systems.
IMPLICATIONS OF PHASE II FINDINGS FOR
FORECASTING AND POLICY
The main result from retesting and refining the global
model is a solid confirmation of the work undertaken in the
first phase of the Task Force’s work. Even with an updated and
expanded problem set, different control sets, and more refined
measures of democracy, the basic global model continued to
accurately classify roughly two-thirds of historical cases.
Moreover, the same independent variables emerged as
statistically significant in a variety of retests.
The major implication for forecasting is that as far as
statistical data are concerned—given current limitations in
accuracy and coverage for global data—using a large number
of variables does not add to the effectiveness of forecasting
models. In many cases, we found that the gaps in either the
temporal or geographic range of particular variables were so
great that any possible gains in prediction were offset by
statistical uncertainties or missing data problems associated with
measuring those additional variables. Thus, in all models and
regional sub-models, a handful of variables emerged as providing
significant power in discriminating between state failures and
stable cases over the past 40 years. Although many additional
variables—including those measuring nutrition, education,
droughts, and civil rights—showed significant correlations with
risks of state failure, they did not add statistical power to models
based on our key variables. Those variables, which consistently
emerged in a wide variety of models, are material living
standards, trade openness, and democracy, and in more limited
circumstances, youth bulge, regime duration, ethnic dominance
or discrimination, and the urban proportion of the population.
34
We shall have to wait until the accuracy and coverage of global
data series improves before we can gain further accuracy by
building more complex models. In the meantime, there is a
compelling need to improve global and regional data on these
key dimensions, and on many other social, economic, political,
and particularly, environmental conditions.
A secondary implication is that the accuracy of statistical
models forecasting state failure risks two years in advance
remains at a level that is useful, but insufficient for refined
predictions. In order to bridge the gap between the two-thirds
accuracy of our statistical model, and the better than 90-percent
accuracy required for effective policy responses, the skills of
individual country analysts and policymakers in assessing rapidly
changing local conditions remain absolutely crucial.
The mathematical data analysis cannot prove causality, but
the correlations are consistent with causal interpretations. Our
findings also suggest policy implications that are interesting and
complex, although the best focus and mix of policy responses
will, of course, vary from case to case.
Involvement in international trade, as measured by trade
openness, is associated with a lower risk of state failure in
virtually all states and all contexts. This finding suggests that
policies or measures—including internal factors such as
dependable enforcement of contracts, modest or low corruption,
and improved infrastructure, as well as bilateral or multilateral
efforts to eliminate trade barriers—that help to foster higher
levels of international trade could help prevent political crises.
Interestingly, it appears that it is the involvement in international
trade itself, and not the eventual prosperity that such trade
provides, that is the key to this effect. The work of Etel Solingen
has shown that free trade, if sustained, helps bring together
coalitions of elite actors that support the rule of law and stable
property relationships, as a condition for building wealth.
35
Such

Author ⋅ Title
67ENVIRONMENTAL CHANGE & SECURITY PROJECT REPORT, ISSUE 5 (SUMMER 1999)
coalitions may or may not be democratic, but in either case, they
promote political stability.
Partial democracies—particularly in lower-income
countries where the quality of life remains poor—are associated
with elevated risks of failure. Although full democracies and
autocracies are fairly stable, the in-between forms of government
are at high risk of undergoing abrupt or violent change. This
suggests that while a policy of promoting democracy may
eventually lead to a world of stable liberal states, one cannot
presume that the inevitable intermediate stages will also be
stable. Policymakers need to be particularly attentive to the risks
of failure in such states, and should seek and encourage progress
toward full democracy. Moreover, if helping to increase the odds
of stability in such states is a goal, then policymakers need to
focus on developing policies that help foster international trade
and on supplementing democratization programs with broad
development programs that help improve the overall level of
material living standards.
Material living standards have an undeniable effect on
the risks of state failure. In some models, it is the overall level
of material living standards that emerges as important; in other
models, such as that for Sub-Saharan Africa, it is the direction
of change that appears crucial. In either case, the policy
implication is that efforts to improve material living standards
are a significant way to reduce risks of state failure. In Sub-
Saharan Africa, it turns out that high levels of urbanization
reinforce this effect—for states with high levels of urbanization,
states experiencing growth in GDP per capita have only a
fraction of the risks of state failure of those states experiencing
economic stagnation or decline.
Despite the prevalence of ethnic conflicts—especially in
Sub-Saharan Africa—ethnic discrimination or domination is
not the sole, or even the most important, correlate of state
failure. Because ethnic factors do not emerge as the most
powerful—or most statistically significant—factors associated
with state failure, they bear monitoring, but other policy levers
may be more readily available and more effective.
Environmental stress, vulnerability, and capacity form an
interdependent triad that affects quality of life and, indirectly,
the risk of state failure. Our findings imply that analysts
concerned with the social impact of environmental change need
to monitor not simply the environment, but also changes in a
country’s vulnerability to environmental changes and its capacity
to cope effectively with them. The increased appreciation of
the need to develop indicators of environmental change and of
sustainability should be complemented with equally vigorous
efforts to develop useful indicators of vulnerability and capacity,
where the recent track record has been less encouraging. At the
broadest level, our findings also suggest that when it comes to
minimizing declines in quality of life, increases in capacity and
reductions in vulnerability are equally appropriate targets for
policy intervention as increases in environmental protection.
Newer democracies, especially in countries where quality
of life is relatively low, are more likely to fail than long-lived
ones. The Task Force’s models and data can be used to inform
policymaking about the conditions under which democratic
transitions are likely to succeed or fail. Most contemporary
democracies in Latin America, Asia, and Africa established
democratic institutions one or several times, then reverted to
autocratic rule before making their most recent transitions to
democracy. The problem-ridden history of democratic
transitions in these regions raises questions about the future
durability of newly established democracies there and in the
post-Communist states. Analytic results suggest it is crucial that
international support for democratic institutions be reinforced
by policies that promote improvement in the quality of life.
FUTURE DIRECTIONS
Potentially fruitful future analytic directions that are
suggested by the Phase II results include:
• Forming a better understanding of the conditions of
successful democratic transitions. Initial results suggest
that successful democratic transitions tend to be preceded
by political experimentation–including previous
unsuccessful attempts to establish democratic institutions–
and to occur in countries where agricultural stress is low
and material living standards are higher. On the other hand,
backsliding to democracy tends to occur within a few years
after democratic institutions are introduced, and in
countries with relatively low quality of life and high
agricultural stress. Analyses are needed of the extent to
which successful democratic transitions depend on
improvements in the quality of life, and economic
performance generally, during the early years. Models of
these relationships should also take account of factors such
as elite ethnicity, urban growth, and youth bulge, which
have been shown to correlate with other kinds of state
failure, especially revolutionary and ethnic wars.
• Further developing the concept that the impact of
environmental degradation on state failure is mediated
by vulnerability and capacity, and more thorough testing
of the model. Additional steps would include:
→ Constructing additional indicators of environmental
change—such as water and air quality—vulnerability,
and capacity from currently available sources.
→ Building a set of “watch lists” for specific ongoing
environmental threats that would focus attention on
environmental deterioration in countries with high
vulnerability and low capacity.
→ Developing a core set of environmental indicators—
measured consistently across countries and over
time—that could be used in future analyses. This effort
would include using the next generation of remote-
sensing satellites to gather terrestrial and atmospheric
data and using intensive on-site monitoring to build
an adequate database for other environmental
problems such as water quality, air quality, and
State Failure Task Force
⋅
State Failure Task Force Report: Phase II Findings

ENVIRONMENTAL CHANGE & SECURITY PROJECT REPORT, ISSUE 5 (SUMMER 1999)
Special Reports
68
chemical hazards.
→ Developing models that capture regional variation—or
localized “hot spots”—within a country that are masked
by national level analysis. We know that the
environmental impact on material quality of life will
be stronger if there is a spatial correlation among the
variables. For example, if a given unit of land has a
high rate of deforestation, a high land burden, and
poor institutional capacity, we would expect a larger
local impact on infant mortality, an hypothesis that
could be tested using currently available high-spatial-
resolution data sets.
→ As additional data become available, continuing to
test the hypothesis that environmental damage directly
contributes to the likelihood of state failure.
• Developing a more detailed concept of “state capacity”
to test as a mediating factor in general and regional
models. Building on the results of the mediated
environmental model, further examine and develop in
more depth the concept of state capacity, develop
quantitative measures that tap this dimension, and
incorporate this concept as a mediating factor. We should
also seek or develop data sets that are better able to capture
state capacity.
• Investigating the usefulness of pilot studies of event data
for bridging the gap between model-based risk
assessments and “early warnings.” The general models of
state failure identify risk factors measured two years before
the expected onset of failure. Even the best models identify
a substantial number of false positives and fail to predict
correctly some failures. The goal is to supplement general
models with early warning models that track the immediate
precursors of failure and provide more accurate and timely
warnings than do risk assessments that are based on
background conditions. Specifically, monitoring of events
should concentrate on situations judged to be at high risk
through expert- and model-based analysis, and statistical
techniques should be applied to study the clustering of
events before a crisis.
• Investigating the impact of international support on the
risk of state failure. Many policymakers and analysts
assume that bilateral and multilateral policies can forestall
some state failures and minimize the severity of others.
Previous Task Force analyses have assessed the impact of
some kinds of international economic policies—such as
IMF standby agreements—on the likelihood of state failure,
but these analyses have not shown strong and consistent
results. The impact of other kinds of international
engagement, such as diplomatic and military support,
development programs, and assistance with institution
building remain to be studied. Appropriate data and
indicators need to be gathered and tested in new models.
Because the objectives and hence the likely outcomes of
international policies have changed since the peak of the
Cold War, such models should distinguish between pre-
and post–Cold War patterns of international policy and
their consequences.
Appendix A: Global Model and General Material
DEFINING STATE FAILURES AND CONTROL CASES
State Failure
State failure and state collapse are new labels for a type of
severe political crisis exemplified by events of the early 1990s
in Somalia, Bosnia and Herzegovina, Liberia, and Afghanistan.
In these instances, the institutions of the central state were so
weakened that they could no longer maintain authority or
political order beyond the capital city, and sometimes not even
there. Such state failures usually occur in circumstances of
widespread and violent civil conflict, and are often accompanied
by severe humanitarian crises. In a general sense, they are all
part of a syndrome of serious political crisis which, in the extreme
case, leads to the collapse of governance.
Only 18 complete collapses of state authority have occurred
during the last 40 years, too few for meaningful statistical
generalization. Therefore, the Task Force broadened its focus
and sought to identify systematically all occurrences of partial
as well as complete state failures that began between 1954 and
1996. We began from existing compilations of data on
revolutionary and ethnic conflicts, regime crises, and massive
human rights violations of the types categorized as genocides
and politicides (political mass murders). An initial list—the basis
for the Phase I analysis—was critically evaluated, updated, and
refined for the present study. The four types of internal wars
and failures of governance are:
36
• Revolutionary wars. Episodes of violent conflict between
governments and politically organized challengers that seek
to overthrow the central government, to replace its leaders,
or to seize power in one region. From the 1950s through
the late 1980s, most revolutionary wars were fought by
guerrilla armies organized by clandestine political
movements. A few, like the Iranian revolution of 1978 and
the student revolutionary movement in China in 1989,
were mass movements that organized campaigns of
demonstrations. These mass movements are included only
if one or both parties used substantial violence.
• Ethnic wars. Episodes of violent conflict in which national,
ethnic, religious, or other communal minorities challenge
governments seeking major changes in their status. Most
ethnic wars since 1955 have been guerrilla or civil wars in
which the challengers sought independence or regional
autonomy. A few, like those in South Africa’s black
townships in 1976-77, involved large-scale, violent protests
aimed at sweeping political reforms. Warfare between rival
community groups is not considered ethnic warfare unless
it involves conflict over political power.

69
Author ⋅ Title
ENVIRONMENTAL CHANGE & SECURITY PROJECT REPORT, ISSUE 5 (SUMMER 1999)
it involves conflict over political power.
• Adverse or disruptive regime transitions. Major, abrupt
shifts in patterns of governance, including state collapse,
periods of severe elite or regime instability, and shifts away
from democratic toward authoritarian rule. Some are
preceded by revolutionary or ethnic wars as in Cuba 1959
and Liberia 1990. They also may precipitate internal wars
and be followed by massive human rights violations. They
are analytically distinct from internal wars, however, and
sometimes occur with minimal open violence. Note that
abrupt nonviolent transitions from autocracy to democracy
are not considered “adverse” and thus are not included as
failure cases.
• Genocides and politicides. Sustained policies by states or
their agents–or in civil wars, by either of the contending
authorities—that result in the deaths of a substantial
portion of a communal or political group. In genocides the
victimized groups are defined primarily in terms
of their communal (ethnolinguistic or religious)
characteristics. In politicides victims are defined primarily
in terms of their political opposition to the regime and
dominant groups.
The 233 internal wars and failures of governance are the
basis of the problem set; that is, the study’s dependent variable.
The list is known to omit low-magnitude cases but is thought
to include all serious cases of these types that began between
1955 and the end of 1996 in all states in the international
system with 1996 populations greater than 500,000.
37
One problematic issue is that internal wars, regime crises,
and gross human rights violations often co-occur. Moreover,
multiple events of the same type sometimes occur sequentially
in the same country. Where wars or crises overlapped or came
in quick succession, they were combined. The final problem
set consists of 127 consolidated cases that include 71 discrete
cases plus 56 complex cases, such as linked sequences of events
(of any kind) in which four years or less elapsed between the
beginning and end of successive cases. The analyses reported
here were based on 125 cases, after excluding two low-
magnitude ethnic conflicts.
Appendix D: Environment
M
EDIATED ENVIRONMENTAL MODEL METHODOLOGY
For the environmental model, the infant mortality rate in
1990 is assumed to be a function of its baseline in 1980, plus
the effects of intervening changes—from 1980 to 1990–in
environmental stresses, vulnerability, and capacity factors, while
controlling for baseline levels in 1980. Symbolically, the model
can be expressed as:
IM
t
= a+b
o
IM
t
o
+Σ(b
i
E
i
+b
il
∆E
i
)+Σ(c
j
C
j
+c
jl
∆C
j
)
+Σ(d
k
V
k
+d
kl
∆V
k
)+ε
Where t is the year 1990, t
0
is the year 1980, IM is infant
mortality, E
i
are environmental stresses, C
j
are state capacities,
and V
k
are vulnerabilities.
Because the number of explanatory variables in a multiple
regression model must be limited to avoid “overfitting,”
38
and
because only about 100 countries have nonmissing values for
all variables needed to estimate the environmental coefficients,
we could include a maximum of 10 independent variables in
the model. Each stress, capacity, and vulnerability factor
contributes two variables—a baseline and a change measure—
with an additional variable required to measure baseline infant
mortality rate. Thus, only one variable from each of the stress,
capacity, and vulnerability categories can be accommodated in
the model, plus at most one additional variable.
To select appropriate covariates for the model we initially
screened potential capacity and vulnerability variables by
computing their correlation with infant mortality. Those that
were significantly correlated were then tested together with an
environmental stress variable in a multiple linear regression
model of the general type shown above. Each combination of
one capacity, one vulnerability, and one environmental stress
variable defined a separate regression model. In addition, since
it was hypothesized that tropical countries respond differently
to environmental stresses, a tropics variable was included.
A lack of data further limited our ability to test variables
in the model, and we were only able to test deforestation and
soil degradation variables as environmental stresses and
telephones per capita, population in subsistence agriculture,
Table D-1: Best Environmental Models
Dependent
Variable
Independent Variables
Environmental
Stress
Vulnerability Capacity
Infant mortality Deforestation rate Percent of population in
subsistence agriculture
Telephon
e
per capita
Infant mortality Deforestation rate Land burden Telephon
e
per capita
Infant Mortality Soil Degradation
(severity times rate)
Land burden Telephon
e
per capita
Table D-1: Best Environmental Models
State Failure Task Force
⋅
State Failure Task Force Report: Phase II Findings

70 ENVIRONMENTAL CHANGE & SECURITY PROJECT REPORT, ISSUE 5 (SUMMER 1999)
Special Reports
and land burden as capacity and vulnerability variables:
• Soil degradation data came from a UN Environment
Program data set—Global Assessment of Human Induced
Soil Degradation (GLASOD)
39
—that contains assessments
by regional soil experts about the severity and rate of
human-induced soil degradation. The assessments—
completed in 1990—reflect processes of degradation over
the previous five to 10 years. We converted the data from
GIS format to country values. The severity of soil
degradation is classified on a 0-4 scale, with 4 being the most
severe. The rate is classified from 0-3, with 3 being
the fastest. We created a composite severity score by
multiplying each classification score by the corresponding
percentage of area and taking the sum. We created
alternative scores by weighting the higher classes of
degradation more heavily and obtained similar results.
• The deforestation rate—defined as the annualized rate of
Table D-2: Environmental Model Coefficients
change in forest area from 1980 to 1990—verged on
statistical significance (p=0.06) in models with telephones
per capita as a measure of state capacity and either land
burden or population in subsistence agriculture as a
measure of vulnerability.
• The tropics indicator was not significant, nor were any
interactions among the capacity, vulnerability, and stress
variables.
None of the soil variables were significant when tested
individually or in simple sums (such as the age of land in class
3 plus the age of land in class 4). However, when the interaction
between severity and rate was tested we achieved significant
results, with telephones per capita as the capacity variable and
land burden as the vulnerability variable. The interaction can
be interpreted as suggesting that the impact of soil degradation
on infant mortality is nonlinear; soil degradation increases infant
mortality only when the degradation is severe and takes place
Table D-3: Environmental Model Output
FPO TEXT Shoot Original
FPO TEXT Shoot Original

71
Author ⋅ Title
ENVIRONMENTAL CHANGE & SECURITY PROJECT REPORT, ISSUE 5 (SUMMER 1999)
1
Esty, Daniel C. Jack Goldstone, Ted Robert Gurr, Pamela Surko,
and Alan Unger. Working Papers: State Failure Task Force Report.
McLean, VA: Science Applications International Corporation, 30
November 1995.
2
For a list of countries included in the study, see appendix A, table
A-1 : Country List.
3
For a list of state failure cases, see appendix A, table A-3: Historical
State Conflicts, Crises, and Transitions, 1955-96.
4
See appendix A for details on the procedure for revising the set of
state failures.
5
For a list of control cases, see appendix A, table A-4 : Control
Cases Used for the Global Model.
6
See appendix A for details on the logistic regression and genetic
algorithm techniques; see Esty, Daniel, Jack Goldstone, Ted Robert
Gurr, Pamela Surko, and Alan Unger, Working Papers: State Failure
Task Force Report. McLean, VA: Science Applications International
Corporation, 30 November 1995, for details on neural network
analysis.
7
Jaggers, Keith, and Ted Robert Gurr. “Tracking Democracy’s Third
Wave with the Polity III Data.” Journal of Peace Research vol. 31(4):469-
482, 1995. For details on the scoring and a list of indicators and
weightings for each index, see appendix C, table C-1: Indicators of
Institutional Democracy and Autocracy, in the full text report.
8
For a list of country scores, see appendix C, table C-2: Democracy,
Autocracy, and Democracy Minus Autocracy Scores by Country, 1996.
9
Zakaria, Fareed. “The Rise of Illiberal Democracies.” Foreign
Affairs, vol. 76(6):22-44, 1997.
10
For a list of control cases, see appendix B, table B-1: Control
Cases Used for the Sub-Saharan Africa Model. Sub-Saharan Africa
crises are included in appendix A, table A-3, Historical State Conflicts,
Crises, and Transitions, 1955-96
rapidly.
The environmental model was obtained by regressing 1990
infant mortality rates on annual deforestation rates, adjusting
for differences in states’ baseline (1980) infant mortality and
differences in their capacity and vulnerability. The adjustment
was accomplished by including as covariates the factors listed
in Table D-2 (telephones per capita serves as a surrogate for
capacity, whereas land burden is a proxy for vulnerability). The
R-squared statistic, which ranges from 0 to 1, measures the
fraction of variability accounted for by the model and therefore
is an indicator of how well the model fits the data. The value of
R-squared in this case is deceptively large, because most of the
variability in states’ 1990 infant mortality is in fact explained
by 1980 infant mortality alone. The model suggests, however,
that even after taking this dependence into account, there
remains an association between deforestation rate and infant
mortality, as indicated by the p-value of 0.06 for deforestation,
almost meeting the conventional statistical significance level of
0.05.
[Editor’s Note: These excerpts of the Phase II Findings of the State
Failure Task Force exclude the Executive Summary portions of
Appendices A (Global Model and General Material and D
(Environment) and all of Appendices B (Sub-Saharan Africa) , C
(Democracy), and E (Data Sources).]
11
See appendix B for further details on the model.
12
It should be noted, however, that this finding did not have quite
as much statistical significance (p=.10) as the other findings in this
model.
13
Goldstone, Jack A. Revolution and Rebellion in the Early Modern
World. Berkely: University of California Press, 1991.
14
On the basis of data two years in advance of the crises.
15
The global model had the best accuracy for Western industrialized
countries and poorer accuracy for Sub-Saharan Africa, where it tended
to misidentify too many countries as failures.
16
Huntington, Samuel P. The Third Wave: Democratization in the
Late Twentieth Century. Norman: University of Oklahoma Press, 1991.
17
For a complete list of transitions, see appendix C, tables C-3,
Transitions From Autocracy to Partial or Full Democracy, or from
Partial to Full Democracy, 1957-91, and C-4, Transitions from Full
or Partial Democracy to Autocracy, or from Full to Partial Democracy,
1957-91.
18
This is a minimum criteria. The median age at which democracies
regressed toward autocracy in the period studied is four years. The
analysis could also be done using a more stringent criterion for stability;
for instance, 10 or even 20 years.
19
Note that failures outnumbered durable transitions because some
failures occurred in countries whose democracies were established
before 1957 and thus were not counted as transitions for this analysis.
20
For details on the method, see appendix C.
21
For a list, see appendix C, table C-6: Transitions from Autocracy
to Partial or Full Democracy Used in Model Derivation.
22
Data were missing for other cases.
23
For a list, see appendix C, table C-5: Transitions From Full or
Partial Democracy to Autocracy Used in Model Derivation.
24
Data were missing for other cases
25
For a useful review of these claims, see Geoffrey D. Dabelko and
P. J. Simmons. “Environment and Security: Core Ideas and US
Government Initiatives.” SAIS Review 17(1):127-146, 1997.
26
These developments are covered in the issues of the Environmental
Change and Security Project Report, Woodrow Wilson Center,
Washington, DC.
27
For example, Robert D. Kaplan. “The Coming Anarchy.” Atlantic
Monthly 44-76, February 1994. For a bibliography on environment
and conflict, see appendix D.
28
Homer-Dixon. Thomas F. Environment, Scarcity, and Violence.
Princeton: Princeton University Press, forthcoming 1999.
29
Levy, Marc A. “Is the Environment a National Security Issue?”
International Security 20(2):35-62, Fall 1995.
30
Glantz, Michael, Michele Betsill, and Kristine Crandall. Food
Security in Southern Africa: Assessing the Use and Value of ENSO
Information. Boulder, CO: National Center for Atmospheric Research,
Environmental and Societal Impacts Group, 1997.
31
Hauge, Wenche and Tanja Ellingsen. “The Causal Pathway to
Conflict: Beyond Environmental Scarcity.” Journal of Peace Research
35:3, 1998.
32
See appendix D for details of the method; for a list of models and
coefficients, see table D-1: Best Environmental Models.
33
Of course, our measure of soil degradation is too imprecise, and
our time frame is too limited for us to determine whether there is an
“optimal” rate of soil degradation. It is entirely possible that extending
the time frame from one to two decades, for example, would have a
negative impact on infant mortality at all levels of soil degradation.
34
Youth bulge—a large proportion of the adult population
concentrated in the young adult years—was a significant factor in a
model of ethnic war that was developed during Phase I. For details,
see Daniel Esty, Jack Goldstone, Ted Robert Gurr, Pamela Surko, and
Alan Unger, Working Papers: State Failure Task Force Report. McLean
State Failure Task Force
⋅
State Failure Task Force Report: Phase II Findings

72 ENVIRONMENTAL CHANGE & SECURITY PROJECT REPORT, ISSUE 5 (SUMMER 1999)
Special Reports
VA: Science Applications International Corporation, 30 November
1995.
35
Solingen, Etel. Regional Orders at Century’s Dawn: Global and
Domestic Influences on Grand Strategy. Princeton University Press, 1998.
36
For sources and more detailed descriptions, see Esty, Daniel C.,
Jack Goldstone, Ted Robert Gurr, Pamela Surko, and Alan Unger.
Working Papers: State Failure Task Force Report. McLean, VA: Science
Applications International Corporation, 30 November 1995.
37
Eritrea and Qatar, which have populations over 500,000, were
inadvertently omitted; Luxembourg was inadvertently included, despite
falling below our population size cutoff, according to the US Census
Bureau’s International Data Base. These deviations from the rule do
not contribute significant error because the number of countries in
the study was large.
38
A widely used rule of thumb constrains the number to about 10
percent of the sample size.
39
“Global Assessment Of Human Induced Soil Degradation
(Glasod): A Users Guide To The Global Digital Database,” UNEP/
GRID, July 1, 1991.
Interested in back copies of the Environmental Change and Security Project Report or the China Environment Series?
These ECSP publications or others such as Climate Action in the United States and China, working papers from conferences
on the toxic legacy of the Cold War in the former Soviet Union, European Seas, or environmental confidence building
are available upon request.
For single copies, please contact ecspwwic@wwic.si.edu or call (202) 691-4130.