Can We Predict Politics? Toward What End?
Michael D. Ward
Duke University
Keywords: prediction, security
Shipwrecks and Security Studies
In 1901, off the Greek island of Antikythera, a ship
pulled into a bay to wait out a storm. After the storm
was over, its divers discovered an ancient shipwreck
containing many valuable antiquities, including jewelry,
coins, statues, and pottery. One item was a lump of corroded
bronze and wood. Everything was carted off to
the National Museum of Archeology in Athens. In 1902,
an archeologist noticed that the corroded lump had
what appeared to be gears in it. He assumed that it was
some sort of astrological clock, but it appeared to be too
far advanced given the dating of the other items it was
found with, which were initially dated to about 150
BCE, and it was ignored for five decades. Several years
later, it was X and c rayed, resulting in images of eightytwo
different fragments of the device.
By the end of 2014, it had been established that the
mechanism dated back further, to about 250 BCE. Many
consider it to be the first computing machine (Freeth et al.
2006). This device was able to calculate and display celestial
cycles, including phases of the moon, as well as a solar
calendar. Perhaps more importantly, it was able to predict
eclipses—seen at the time as omens. In fact, many think
of the Antikythera Mechanism as an omen prediction
device, with all the other predictions it makes simply byproducts
of its true purpose.
The main purpose of this mechanism was to generate
accurate predictions, and whoever used it could do so without
a detailed theoretical knowledge of Hipparchosian astronomy
as applied to irregular phases of ellipsoid orbits.
Thus, it was pure prediction and did not “explain” anything.
At the same time, it embodied detailed engineering
that was based on a theoretical mechanism that provided
exquisite details of planetary orbits in general as well as
specifically. For some, that might suffice as explanation. It
seems clear that this accurate prediction device was possible
only because it was based on a deep understanding of
celestial orbits. What is particularly surprising is that this
level of astronomical prediction was apparently lost in the
shipwreck and did not reappear in Europe for over a millennium.
In retrospect, it seems unlikely that this device
would simply disappear, given how powerful it must have
been in 250 BCE. Yet, as far as we know, this is exactly
what happened.
Jumping ahead a thousand years, we see that prediction
is deeply embedded in the philosophy of science.
Although in hiding in the aftermath of the French
Revolution, Marie Jean Antoine Nicolas de Caritat,
marquis de Condorcet, wrote in Historical View of the
Progress of the Human Mind (1795, translated):
If man can, with almost complete assurance, predict phenomena
when he knows their laws, and if, even when he does not,
he can still, with great expectation of success, forecast the future
on the basis of his experience of the past, why, then, should it
be regarded as a fantastic undertaking to sketch, with some pretense
to the truth, the future destiny of man on the basis of his
history. The sole foundation for belief in the natural sciences is
this idea, and the general laws directing the phenomena of the
universe, known or unknown, are necessary and constant.
This theme continues to the current day through
Auguste Comte (1846), John Stuart Mill (1843), A. J.
Ayer (1936), Carl Hempel (1935), Karl Popper (1935),
Thomas Kuhn (1962), and even Jon Elster (1989, 2007),
who notes:
To predict that less of a good will be bought when its price
goes up, there is no need to form a hypothesis about human
behavior. Whatever the springs of individual action—
Ward, Michael D. (2016) Can We Predict Politics? Toward What End? Journal of Global Security Studies, doi: 10.1093/jogss/ogv002
VC The Author 2016. Published by Oxford University Press on behalf of the International Studies Association. All rights reserved. For permissions, please e-mail:
journals.permissions@oup.com
Journal of Global Security Studies, 1(1), 2016, 80–91
doi: 10.1093/jogss/ogv002
Review Essay
Downloadedfromhttps://academic.oup.com/jogss/article/1/1/80/1834930bygueston19March2021
ration, traditional, or simply random—we can predict that
people will buy less of a good simply because they can afford
less of it. Here there are several mechanisms that are
constrained to lead to the same outcome, so that for predictive
purposes there is no need to decide among them. Yet for
explanatory purposes the mechanism is what matters. It
provides understanding whereas prediction at most offers
control. (1989, 9)
The argument is that if you can develop models
that provide an understanding—without a teleology of
why things happen—you should be able to generate
predictions that will not only be accurate, but may also
be useful in a larger societal context. The basic
argument is essentially that it should be possible to
develop a predictive science of human behavior. It is
debatable whether the so-called regularities are constant
or changing.1
During the height of the war in Vietnam, in the mid-
1960s, a graduate student at Stanford began to systematically
study the details of the war, down to the level of
troop numbers on all sides, number of bombing sorties,
casualties, troop attrition, kill ratios (then a big indicator
in McNamara’s Defense Department), and political support
in the United States and South Vietnam for the incumbent
administration and the war itself, along with the
number of Viet Cong and North Vietnamese defectors.
Statistical models of the relationships among these variables
were constructed and validated toward the goal of
generating predictions (from computer simulations) about
the escalation of the war. Jeffrey S. Milstein published his
efforts in Dynamics of the Vietnam War: A Quantitative
Analysis and Predictive Computer Simulation in 1974, and
his working papers on this topic were widely read in the
academic and policy communities. This was one of the first
uses of prediction in the study of international relations
(IR) and security studies. By 1974, articles were starting to
promote forecasting in the realm of world politics
(Choucri 1974), and a few years later, there was a volume
devoted to the topic (Choucri and Robinson 1978).
There have been a number of scholarly efforts at prediction
in the realm of international and domestic conflict.
(Ironically, the Vietnam War was considered
international by many comparativists, but a civil conflict
by many IR scholars). One of the first calls for greater
attention was by Herman Weil (1974), then working in
the defense-consulting sector. After a few efforts in the
later 1970s, an article by Gurr and Lichbach (1986)
took Gurr’s model of Why Men Rebel (1970) and used
it to make out-of-sample forecasts of where there would
be future conflicts. It did so explicitly to test the theoretical
model of conflict that had been initially developed
by Ted Robert Gurr. It forecasts the number of days
of protests as well as the number of deaths in protests in
ten countries. But this study generated forecasts that
used data measured in five-year periods, and the period
from 1971 to 1975 was forecast based on data for
about 1970. Nonetheless, it was one of the first empirically
oriented studies to focus explicitly on forecasts.
Until very recently, most of this thread of work in
security studies had been lost, or if not lost, at least
abandoned.
Cioffi-Revilla (1996) provided a published prediction
of what was likely to happen in the first Iraq War. It
was the only published prediction that I can find, or remember,
though it took a while to get published. By
now, there is a bevy of recent efforts that include a forecasting
component (Gleditsch and Ward 2000, 2010,
2013; King and Zeng 2001; Ward and Gleditsch 2002;
Hegre 2008; Weidmann and Ward 2010; de Mesquita
2011; Metternich et al. 2013; Ward et al. 2013; Koubi
and Bo¨hmelt 2014; Pilster and Bo¨hmelt 2014; Schutte
2014). Why has this thread been so sparse in security
studies? Two words: Kenneth Waltz.
Many social scientists see a sharp distinction between
explanation on the one hand and prediction on the
other. Indeed, this distinction is often sharp enough that
it is argued that doing one of these things cuts you out of
doing the other. Kenneth Waltz (1997) is a good example
of this belief. A long-standing, but incorrect, example
has been the weather, about which it has been
famously argued that while you can understand the various
components of the weather system—the evaporation
of water, its collection in clouds, the changing temperatures
that result in lightning, and the like—having this
explanation does not enable you to make actual predictions
about the weather. It is argued that because the
contexts are sufficiently varied and numerous, they
would defeat our ability to predict the weather. Mill argued
against this idea, the then-prevailing opinion that
tidology would never be a precise endeavor. He proposed
that weather prediction could, in principle, be
successful and become an exact science. Guess what?
Against all the naysayers, he was right.
Nonetheless, the idea that prediction and understanding
are different has persisted and in fact is
widespread in the realm of security studies. In a simple
way this is true, because you can develop a predictive
system without necessarily understanding all the details
that are in play. A classic example often offered is that
1 Others—Hume and Heidegger, for example, as well as those working
on recent hermeneutics—had a different view that I do not elaborate
herein.
MICHAEL D. WARD 81
Downloadedfromhttps://academic.oup.com/jogss/article/1/1/80/1834930bygueston19March2021
an individual can be an excellent pool player without being
able to explain the basic equations of motion. But
this example seems disingenuous, and solutions to the
equations may take different forms. Perhaps, a kind of
“intuitive” understanding may supplant a more “analytic”
understanding. We know that analytic knowledge
is harder to communicate to others than is intuitive information,
but it may be that both types of knowledge
are useful and even accurate. Moreover, who knows exactly
what is going on in the nervous system of an excellent
pool player at a level below cognition?
It is clear that the distinction between these two ideas
about knowledge is prominent, but it is very hard to determine
where it originates. Note the following training guide
on telling the two apart: http://bit.ly/1DErHSD. This comes
from the US National Institutes of Health Office of
Behavioral and Social Science Research and is part of their
online training curriculum that helps practitioners determine
what is useful for prediction and explanation. This
visual guide distinguishes between predictive and explanatory
concepts related to the risk of teen pregnancy, allowing
practitioners to “learn” which category best describes
different variables such as race. The software not only classifies
factors into the mutually exclusive categories of explanation
and prediction, but it also grades you on how
well you can distinguish both.
It turns out that most of the examples of what we
cannot predict have been overturned by scientific inquiry,
accompanied by clever and concerted research
programs. In part, that was Popper’s (1935) point.
Lewis Fry Richardson dreamed in 1922 of thousands of
individuals making thousands of calculations in order to
do weather prediction. Richardson had just spent over
two years making the necessary calculations for a single
day, May 20, 1910. He described his dream:
After so much hard reasoning, may one play with fantasy?
Imagine a large hall like a theatre, except that the circles and
galleries go right round through the space usually occupied by
the stage. The walls of this chamber are painted to form a map
of the globe. The ceiling represents the north polar regions,
England is in the gallery, the tropics in the upper circle,
Australia on the dress circle and the Antarctic in the pit. A myriad
of computers are at work upon the weather of the part of
the map where each sits, but each computer attends only to
one equation or part of an equation.
A picture of that dream was remarkably prescient,
Richardson’s Dream, which had 60,000 “computers”
(individuals with slide rules) working simultaneously to
produce a weather forecast. Detailed data on a wide variety
of variables, coupled with rapid calculations of
some fairly “simple” equations, have produced weather
predictions that are sufficiently accurate that there is
now a derivatives market in weather.2
On the other hand, many scholars (but few others)
will tell you that we need more theory. Doubtless they
are right. Few of them really mean “theory” in the sense
that I reserve for the term. Few of them mean “theory”
in the sense of analytical narratives. Many of them mean
“detailed, plausible stories” about how stuff occurs.
Long ago in a galaxy far away from here, James
Caporaso (echoing Arthur Stinchcombe) taught me that
any good social scientist should be able to come up with
a variety of plausible stories about how something
comes about, about what “causes” things, not just one.
Unfortunately, much of security studies appears to stop
at this junction. At the time, Caporaso intended it as a
challenge to guide research. I have come to appreciate
that it is also a curse. In short, most scholars seem to
want more theories that answer the question of “why”
something occurs. A wide swath of published articles in
security studies broadly includes a section on “theory,”
and even more empirical studies are pilloried for weak
or absent theory.3
What is typically meant by “more theory” falls into a
way to create more nuanced understandings of the
world. Often, what is desired is an attempt to find ways
to encapsulate parts of the theory in a broader picture
aimed at reconciling empirical findings with what is prescribed
by the theory. Vasquez (1997) details how this
typically works in security studies and elsewhere in his
widely cited examination of Waltz’s balancing proposition.
He details a longish list of recommendations,
which he claims is degenerative in that it avoids attempts
at refutation whether they are logical, definitional, or
historical. Waltz’s response (1997) misses the point
but does offer up another defense: theories are meant for
explanations, and he claims that “tests are always
problematic.”
In his most recent article in Security Studies, Daniel
Altman (2015) proposes “a new theory of false optimism
as a cause of war.” Along with formal models,
Altman points to Pearl Harbor as illustrative evidence
that Japan’s decision-making in 1941 was “an instance
of false optimism that lead to war, in this case war
against a country with nearly ten times its military potential.”
Altman argues that when alternative strategies
are proposed, those that are most optimistic are most
2 Among many derivative market products for weather, see http://
www.climetrix.com.
3 By wide swath, I mean virtually all.
82 Can We Predict Politics? Toward What End?
Downloadedfromhttps://academic.oup.com/jogss/article/1/1/80/1834930bygueston19March2021
likely to be chosen and those that are most pessimistic
are likely to be avoided. But this “theory” is not really
examined empirically, nor are there counter examples.
Theory is also used as a goal of process tracing.
Consider Mahoney’s claim that process tracing can be
used for both theory testing and theory development,
developed in many places but available in Mahoney
(2015), in which a scholar is interested in the putative,
generic question “What X caused Y in case Z?” or for
example “What are the possible causes or alternative
explanations of World War I? . . . [T]his [is] the theory
construction task and process tracing is one tool for
pursuing it.” The more prosaic task is to determine
whether this generated theory was validated in a specific
case, although herein the case is already pretty specific.
There are many guidelines, including reasoning logically
(surely a good thing), having knowledge of specific
cases (also beneficial, but potentially misleading), and
having “good knowledge of relevant preexisting theories
and generalizations.” The latter is doubtless useful,
perhaps to know which ideas to throw away and
which to build on. For Mahoney, and many others, a
theory appears to be a list of possible causes or
explanations.
For many others, “theory” is what is missing in
extant research on security studies or simply what they
want to focus on. Beckley (2015, 9) notes that there is a
paucity of studies that focus on theory building in the
realm of alliance studies. He constructs two competing
“perspectives”: entanglement theory that alliances drag
states into war and “freedom of action theory,” suggesting
that loopholes avoid entanglements. Although
anointed as theories, Beckley often refers to each of
these simple, long-standing ideas in world affairs as
“perspectives.” Another approach to “theory” is to refine
extant explanations. A terrific example of this can
be found in Liff and Ikenberry (2014), who focus on the
theory of “security dilemmas” as a guide to understanding
military competition in the Asian Pacific. They take
the idea of military competition back to Herz (1950)
and Jervis (1978), though I would date it back to
Richardson (1960), which was written much earlier
(first published in 1948, but written in the 1930s) and
published posthumously. Liff and Ikenberry suggest that
“on both sides” states prefer to avoid consumption of
security-producing goods (militaries, wars, treaties) if
they can secure credible commitments without them.
Mistrust and uncertainty generate an action–reaction
dynamic that is contextualized in a number of ways,
(notably those introduced by Jervis [1978]), and point
to the importance of distinguishing between putatively
offensive versus defensive postures, actions, and goods.
This article expands the set of things to examine in order
to determine this dynamic balance, adding considerable
nuance along the way (Liff and Ikenberry 2014, 61):
To begin, borrowing from recent scholarship critiquing the
balancing literature, we expand the scope of the metrics typically
employed in scholarly debates on the security dilemma
to include a broader selection of internal and external policy
measures aimed at enhancing a state’s military capabilities.
Indeed, various military force development and force employment
measures aimed at enhancing military capabilities
often overlooked in the existing literature can significantly
influence a state’s perceptions of its would-be adversary’s
intentions. Such measures include qualitatively improving
military capabilities through modernization, innovation, or
rationalization; transforming force structure or posture to
confront changing threats; tightening military ties with
other states short of new, formal mutual defense pacts
through joint exercises and training, hosting or rotating foreign
forces, co-locating military facilities, expanding interoperability
and joint contingency planning, and sharing
intelligence and military technology. In addition, we argue
that independent of shifts in material power, leadership rhetoric
and political statements can generate insecurity in
others. For example, statements seen by one side as supporting
the status quo may be interpreted by the other party as
offensive and threatening.
Each of these may be useful exercises, but whether they
merit the term “theory” or validate the repudiation of
other strategies is open to interpretation. They do add
nuance.
In a soon-to-be-classic piece, Healy (2015) notes that
this sort of theoretical nuance has (also) gained a foothold
in sociology. He finds this dysfunctional, not because theory
or nuance is bad, but like Vasquez, because it hides
ideas from critical examination. This nuance is fractal in
that it appears in calls for higher as well as lower levels of
abstraction. On the one hand, we see calls for the nuance
of ever more detailed empirical analyses that illustrate
which theory needs to be contextualized and amended to
be useful. An opposite approach is to demand a theory of
such a level of generality that it does not apply to anything
precisely observable in the world. Healy’s third nuance
is the epitome of contextualization, “the insinuation
that your sensitivity to nuance is a manifestation of one’s
distinctive. . . ability to grasp and express the richness,
texture, and flow of social reality itself. This is the nuance
of the connoisseur” (3). Healy concludes that sociology is
“gutted” with nuance, and nuance needs to be avoided,
at least for now. I have a similar view of the nuance of
theory in security studies.
MICHAEL D. WARD 83
Downloadedfromhttps://academic.oup.com/jogss/article/1/1/80/1834930bygueston19March2021
I am here to suggest that less is more.4
Thus, let me
be the first to call for less theory in security studies. We
should winnow the many, many such “theories” that occupy
the world of security studies.
Instead, we need more predictions. We need these
predictions for four reasons. First, we need these predictions
to help us make relevant statements about the
world around us. We also need these predictions to help
us throw out the bad “theories” that continue to flourish.
These predictions will help drive our research into
new areas, away from moribund approaches that have
been followed for many decades. Finally, and perhaps
most important, predictions will force us to keep on
track. It is hard to imagine that in ten years hence, the
theories I have highlighted above will have been evaluated
in terms of their accuracy. It is easy to go back in
the literature ten years and find similar pronouncements
that were left hanging out there in the realm of JSTOR
alone. At the end of the last century, John Vasquez and
Kenneth Waltz debated the value of predictions in terms
of their implications for theory (Vasquez 1997; Waltz
1997). As it turns out, this debate, while interesting, is
not relevant for our discussion because neither Vasquez
nor Waltz mean predictions in the sense of a forecast,
but rather in the sense of an empirical regularity or fact
that can be induced or deduced by the theoretical framework.
Indeed, most of the “predictions” in this debate relate
to events in the early twentieth century.
Predictions in the Larger World
In the 1980s, Philip Tetlock was concerned—like
many—about the possibility of a nuclear conflict between
the United States and Soviet Union. Indeed, this
was the main focus of most security studies during the
Cold War. Popular movies and scholarly journals highlighted
what time it was on the so-called “doomsday”
clock. Tetlock wanted to understand why the pundits
made statements about the future that were all over the
place. Frustrated by the fact that pundits justified their
inaccuracies with revisions of history, timing, and other
excuses, he collected forecasts—an amazing number of
them (25,000)—and started to keep track and grade
their accuracy. Tetlock’s book, Expert Political
Judgment (2005, 2010), showed that pundits as well as
social scientists making predictions were equally and totally
bad at it. An oft-cited refrain from the book is that
“dart throwing chimps” were equally accurate to
experts. Indeed, the most well qualified of experts often
turned out to be the worst forecasters. Unlike Bill
Ascher (1979), who made similar points earlier, Philip
Tetlock (2005) undertook to improve forecasting
methodology.
With his team of statisticians and social scientists—
including Barbara Mellers and Don Moore—Tetlock
was able to propose an adventurous and ambitious program
of research for the then newly formed IARPA
(Intelligence Research Advanced Projects Agency), a research
arm of the US intelligence community focused on
addressing hard problems with open-source efforts between
scientists and those in the intelligence community.
The Good Judgment Project (goodjudgmentproject.org)
was the result. Essentially, the effort is to provide specific
forecasting questions that (a) have a discernible
outcome (e.g., Will Kim Jung Un preside at the May 1
parade in Pyongyang in 2014?) and (b) have a precise
time of resolution. These two characteristics of the forecast
can be judged independently from the forecast itself
(a problem when forecasts are self-evaluated). About
4,000 people have signed up for this tournament. They
are also trained to use the computing platform. Each individual
is made part of a group, and these groups as
well as the individuals compete in terms of their accuracy
(measured largely by Brier scores).
Two major lessons are gleaned from the experiments
that are embedded in these tournaments. One is that
understanding the base rates at which some phenomena
occur is very important. Without an understanding of
what the base rate is, it becomes difficult to tell whether
an event is “normal” or exceptional. Is a protest of 1,000
people in Tahrir Square in Cairo something that is unusual,
or does it happen on a monthly basis, for example?
The second lesson follows from the first: How do you
identify an exception? Just these two skills, which can be
easily taught, cause a change in thinking about what is
likely to happen. And it turns out, they also help individuals
make more accurate forecasts. Two other things are
important. One of them is the simple act of keeping track
of your successes and failures, something that is unlikely
to receive widespread assent by media stars and pundits.
The second is more prosaic: Aggregations help reduce
bias and uncertainty. The classic example of this is the
1906 livestock fair in which 800 individuals guessed the
weight of an ox on display. No single individual got the
right answer, but according to Galton, both the mean and
the median of these 800 estimates were within one percent
of the correct weight (Galton 1907).
4 I agree that we need more real theories. More generally, a theory is
an explanation of some aspect of the natural world that is logically
consistent and generates implications that are empirically falsifiable
through observation and/or experimentation. A theory is not a simple
conjecture or hypothesis. Many social science conjectures and
hypotheses are called theories by their authors.
84 Can We Predict Politics? Toward What End?
Downloadedfromhttps://academic.oup.com/jogss/article/1/1/80/1834930bygueston19March2021
This leads to four points:
1. Base rates are important contexts for predictions.
2. Exceptions to base rates are important to identify;
however, what an exception is remains complicated.
3. Keeping track of successes and failures is important.
4. Having a lot of answers may be the only way to get
the right answer, but doing so may require some aggregation
from a variety of perspectives.
How are Tetlock’s forecasters doing in this competition
with prediction markets, subject matter experts,
and other unspecified “work products” of the intelligence
community? So far, there are five independent research
teams, of which Tetlock’s group is one. This
experiment involves over one million judgments from
over 10,000 participants. This is based on about one
hundred real events each year, for example “Will Iran
sign an IAEA Structured Approach document before
June 1, 2014?” A systematic comparison of the various
approaches is underway. In short:
• Aggregations of trained forecasters beat subject matter
experts by about fifty percent, and they also beat
prediction markets.
• Prediction markets significantly out-perform subject
matter experts. Domain expertise is less important
than problem solving and belief updating.
• The top forecasters—the so-called super forecasters—
tend to be the same individuals each year.
Another IARPA project is also interesting to examine.
It is known as the OSI, Open Source Indicators,
project, in which the basic idea is that classified information
is possibly inferior to open-source materials for
making important predictions. Thus, an attempt has
been made to undertake detailed, precise forecasts of the
type often engaged in within the intelligence community,
while only using materials found in the unclassified
world. The goal is to develop and test methods for the
automated analysis of publicly available data to anticipate
or detect significant social events. Such events include
disease outbreaks, political instability, and
elections. The goal is kind of a version of Google
FluTrends on steroids: beat the news by fusing early indicators
of events from diverse data (Ramakrishnan et al.
2014). They are less interested in theory, but do base their
predictions on exactly the kind of data that are widely used
in the social sciences.
Some recent predictive successes include:
1. Riots after impeachment of Paraguay’s president
(2012)
2. The so-called “Brazilian Spring” (June 2013)
3. Hantavirus outbreaks in Argentina and Chile (2013)
4. Venezuelan student uprising (Feb 2014)
What is remarkable here is that these predictions are
graded by independent subject matter experts reading the
local, non-English text. This suggests that not just trained
individuals, but also trained statistical models might be
able to generate accurate and useful predictions.
Recent Work
Over the past eight years, there has been an effort to create
a forecasting platform for decision-making within the defense
and intelligence communities, again based on open
sources. This is known as W-ICEWS, the Worldwide
Integrated Crisis Early Warning System, often abbreviated
ICEWS. It has a lot of components, but the most pertinent
is the suite of models, developed by social scientists, that
forecast major instability events around the world with
high accuracy. The basic idea is to use data that are detailed
but aggregated to the month. These data are both
structural and behavioral. The behavior data are event
data produced using a tested ontology of categories and
an automated procedure for constructing word graphs of
stories order to glean context. The actor dictionary and
verb dictionary are updated monthly, and the data are
made available on a monthly basis. As of March 2015,
these data are publicly available (Boschee et al. 2015;
Lautenschlager et al. 2015; Lustick et al. 2015).
Each of three teams creates several models of political
instability events. These events are rebellions, insurgencies,
ethnic violence, and domestic and international
crises. Several theme models are developed using contemporary
social science literature. The themes each address
different substantive arguments. For example, in
developing models for domestic international crises, we
generated themes that captured (a) the performance of
the macro economy, (b) the demographics of each country,
including youth bulges and other demographic factors,
(c) the structure of the political system, (d) the
institutional infrastructure of each country both politically
and economically, (e) the scope and extent of rebellious
activity, (f) the interactions between the standing
government and various factions of opposition, and (g)
the extent of economic vitality (or stagnation) in neighboring
countries. Each of these themes produced predictions
based on a limited but useful set of forces thought
variously to be important. Without going into a lot of
detail, we use these themes to produce predictive distributions,
which we then combine using a special version
MICHAEL D. WARD 85
Downloadedfromhttps://academic.oup.com/jogss/article/1/1/80/1834930bygueston19March2021
of Ensemble Bayesian Modeling Averaging, which
weights each theme in proportion to its ability to make
accurate predictions in a test set of data set aside for
this calibration test (Raftery et al. 2005; Vrugt et al.
2006; Vrugt, Diks, and Clark 2008; Fraley et al. 2010;
Fraley, Raftery, and Gneiting 2010; Sloughter, Gneiting,
and Raftery 2010; Montgomery, Hollenbach, and Ward
2013, 2014, 2015). Our predictions are then combined
in the same way, by an independent group with the predictions
made in different ways by different modelers
(all, as it turns out, political scientists).
The so-called ICEWS component models developed at
Duke (Ward et al. 2013; Ward 2015) produce accurate
forecasts, and as expected, the ensemble forecasts are generally
more accurate, in terms of having both fewer false
negatives and positives than any individual model. They
are combined with other models within the ICEWS project,
again using Bayesian ensemble techniques. The result
is a model that preserves transparency, yet maximizes prediction.
At the same time, it is based on explanations for
political stability that are found in the literature. The result
of this approach has been successful and more accurate
than other efforts that tried to find the best single model.
Table 1 illustrates these results, which pertain to sixmonth
predictions for 167 countries each month.
What we have learned in the project so far is that aggregation
of different diverse models produces better
predictions than models that are singular in their explanation
of what is occurring in the world. Further, this
principle seems to scale up as well as down, and different
models can benefit from aggregation as well as
diverse approaches. It may be that there is no one best
model from a forecasting perspective. Moreover, we
find that like the weather, other forms of complicated
phenomena that exhibit violent variation—conflict and
crisis—can also be explained and predicted.
There are many more political forecasting projects that
I will not discuss in detail at present, including the attempt
to forecast conflict decades into the future. These include
Choucri (1974), Hildebrand, Laing, and Rosenthal
(1976), Choucri and Robinson (1978), Freeman and Job
(1979), and Singer and Wallace (1979) in the 1970s, and
Vincent (1980), Gurr and Lichbach (1986), Cioffi-Revilla
(1996), Davies and Gurr (1998), and Pevehouse and
Goldstein (1999) in the 1980s and 1990s.
More recently, in this century, there have been
many ongoing efforts at forecasting.5
In April 2015,
new predictions of weather and conflict in Africa at the
1 deg grid (Witmer et al. 2015) were presented at a conference
at PRIO (Peace Research Institute Oslo) along
with a variety of other new studies (Beger and Ward
2015; Brandt 2015; Colaresi 2015; Gleditsch 2015;
Hegre, Høyland, and Nyga˚rd 2015; Hunziker 2015;
Schneider 2015; Schutte 2015) all focusing explicitly
on prediction of conflict (See also the Joshi et al. scenarios
about improved governance [Joshi, Hughes, and
Sisk 2015]).
There will be many successes (and failures) in the
years ahead as the policy and scholarly world turns to a
more rigorous examination of their understandings
through predictions, not just locally in elections, but
globally as well in a wider range of human endeavors.
The bottom line with theory is that for the most part,
it looks back at the world and describes how things occurred.
Often, it attempts to develop a teleology of why
things occurred. Rarely is there a track record kept of
whether the theory was borne out in cases not initially
considered and where there is some ex ante evaluation,
the theoretical response is often to add greater nuance to
encompass the situations or cases that were not well covered.
Theory building does not generally lead to new
ideas. This is a well-known facet of inquiry that has been
presented and debated for decades, if not longer.
Predictions, failed as well as successful, have greater potential
for moving the needle on what we know. The cure
for bad theory is more theory; the cure for bad predictions
is better and often different understandings of what is implied
by the underlying (and changing) social dynamics.
Table 1. Current metrics for out-of-sample results for W-ICEWS ensemble models
Insurgency
(%)
Rebellion
(%)
Domestic political
crisis (%)
Ethnic religious
violence (%)
Dyadic international
crisis (%)
Precision 89 84 68 98 96
Recall 80 88 49 71 95
5 See Gleditsch and Ward (2000), de Mesquita (2002, 2009, 2011),
Feder (2002), O’Brien (2002), Ward and Gleditsch (2002), de Marchi,
Gelpi, and Grynaviski (2004), Brandt, Colaresi, and Freeman (2008),
Hegre (2008), Schneider, Gleditsch, and Carey (2010), Weidmann and
Ward (2010), Ulfelder (2012), Gleditsch and Ward (2013), Hegre et al.
(2013), Blair, Blattman, and Hartman (2014), and Brandt, Freeman,
and Schrodt (2014).
86 Can We Predict Politics? Toward What End?
Downloadedfromhttps://academic.oup.com/jogss/article/1/1/80/1834930bygueston19March2021
Data Constraints
One of the difficulties of doing predictions is that data
are often hard to obtain for the future. More practically,
all quantitative and quality investigations are limited to
information about the past. The typical response to this
kind of approach is to employ all the available information,
because more information is always better. In the
quantitative world, this results in an undesirable dependence
between the data collected and the model posited.
This dependence typically results in overfitting in which
the model has been sculpted to the available data but
might not characterize additional data (such as the future
or different cases). The same can happen qualitatively
wherein new cases are sought to fit different
aspects of the theory being examined or constructed.
Even if we do not have time to wait for data on the
future to appear as a way of examining the external validity
of our explanations, we still have the past. Ideally,
we should divide our cases and samples into different
groups, one of which we use to estimate the model and
evaluate its characteristics as we refine or estimate it statistically.
These cases are known as the training cases. A
second set of observations, known as the test data, can
be then examined to see if they also conform to the theory/model.
This procedure, by now a gold standard in
most of science, is known as cross-validation and serves
as a way to anneal one’s investigations against being too
tied to the data at hand.
What Are the Pros and Cons of a Predictive
Approach to Social Science?
The main pro is that the predictive enterprise helps us
evaluate how well we are doing so that we can improve
our understanding of the world. It is the gold standard
of a scientific approach. We do not yet have an experimental
framework for many important subjects. As a result,
it is important to make sure we can get the same
kind of results with new information that we got with
the data we began investigating. That means we have to
either save some data back (a great idea), use the future
to see how well our modeled understandings perform, or
preferably both. There are only nascent traditions of this
in the social sciences at present. Keeping track of your
success is not collecting significant coefficients. Keeping
track matters. One consequence is that we cannot just
keep using the same data over and over. And over. One
reason that many hate predictions is that talking heads
make many predictions in the media, but few of them
ever keep track of how well they are doing. Their goal is
somewhat akin to a venture capitalist’s make enough
bets that eventually one of them is correct enough that
you get to make a lot more bets. Ascher (1979) long ago
showed that the talking heads were most often wrong.
This is still true. You would think that they would get
better over time, but there is little evidence that this is
the case.
We also will be driven into making more precise investigations
once we start to predict. We will not be satisfied
with annual data for most things. Nor will we
necessarily be satisfied with national-level information
because it becomes even more apparent in the predictive
domain that the world is neither flat nor homogeneous.
As a result, we should get more precise understandings
of how things play out in our social world. At the same
time, we have to recognize that our predictions are probabilistic
and contain a large amount of uncertainty,
more so than in other endeavors.
As a result of these two aspects, better and more precise
understandings of our social world, it is possible to
be more relevant to decision makers at all levels. This
does not mean just inside the beltway. It also means decision
makers at CDC (Centers for Disease Control) as
well as those in non-governmental organizations around
the world.
What are the cons? Several arguments are usually
brought to the fore.
1. The world is inherently unpredictable. But we are
studying it anyway. Go figure. The refrain to this litany
is often “but I know what is going to happen in
this instance.” Maybe, but let us keep track and find
out if you are right. This is the talking heads premise,
and it is demonstrably false. Making cause and effect
statements about politics does imply that politics is
in part, at least, predictable.
2. This will empower the establishment and impoverish
those without power. Actually, it might. But at the
same time, it provides ways in which those outside
the capitals can also affect the future. Why will prediction
be more valuable to the establishment than it
is to the rest of society? Is the same thing true of explanation
and substantive knowledge? Western society
is based in part on the idea that knowledge is a
valuable thing for all. It is true that some take more
advantage of it than others. But having open and
available knowledge can be important for many diverse
groups. As an example, we might think that
clandestine organizations can benefit from open
knowledge, even precise actionable knowledge, but
as we think these thoughts, most of us might not be
thinking about transnational activist networks but
rather large governmental organizations. However,
MICHAEL D. WARD 87
Downloadedfromhttps://academic.oup.com/jogss/article/1/1/80/1834930bygueston19March2021
it is clear that non-governmental actors are also consumers
of knowledge.
3. It is possible to predict things without true understanding
or knowledge. Sure, but rarely is this true
for any but the simplest of systems. I am reminded of
the wonderful essay by Calvin Trillin describing the
chicken on Mott Street in New York’s Chinatown
that played and always won Tic-Tac-Toe. The
chicken did not understand the game. But the
chicken always won.6
It is ridiculous to suggest that
we have models that predict as accurately as the
(now gone) Mott Street chicken but have the same
understanding of the “chicken” game. Even if it
were the case (and I repeat it is not), could the opposite
really be true? If you have deep understanding of
the world, should you not be able to generate accurate
predictions of how it will work in situations you
have not seen before? Will proximate effects lead us
toward distant causes much like proximate causes
can lead us to distant effects?
4. We will disrupt the space-time continuum. If we can
predict conflict, for example, we will be able to prevent
it. Or start it where we want. And then we will
no longer be able to predict conflict. One of my models
attempts to predict where there will be coups de
e´tat and other types of irregular regime changes on a
monthly level. Maybe if Muhammadu Buhari, who
assumed the presidency of Nigeria at the end of May
2015, sees our manuscript, he might be able to prevent
any irregular leadership change from occurring
in the next six months. And maybe that would be
bad. Or good. But in any case, I am pretty sure that
the Nigerian president is already aware of the fragility
of the Nigerian political landscape. This is a frequent
type of criticism of forecasting. I think we can
wait for this to become a real problem before we
stop trying to develop better understandings of the
world.
5. Real social effects occur glacially. Predictions will be
focused on epiphenomenal changes that won’t matter
in the long run. How do you know?
In summary, we need less theory because most theory is
an attempt to rescue or adapt extant theory. We need more
predictions in order to keep track of how well we understand
the world around us. They will tell us how good our
theories are and where we need better explanations.
Predictions are like cell phones. First, they seem arcane and
bizarre. Then, in a few short years, there is no one around
who remembers life without cell phones and your kids use
them in ways you don’t understand. The 2012US presidential
election was the first that was famously and accurately
predicted. But it will be the first of many. All future voters
will vote in an era in which accurately predicting the election
will be the norm, not the exception, though as in the
UK in 2015, there will be exceptions. This will have consequences
for democracy. In the same way that having a
product recommended to us on the web is now normal,
this will become the new normal. Data science (and more
data) will guide us to a better understanding of our future
than we have now. Whether you are involved with commercial
organizations, local government, non-governmental
organizations (NGOs), the federal government, or
international organizations, prediction will be part of the
daily ebb and flow of information, and we shall become
used to seeing accurate predictions about a wide variety of
political phenomena.
But, as we get more accurate, will we be able to begin
manipulating outcomes? Engineer results? It may not
seem like it to everyone, but political beliefs are malleable.
In fact, survey researchers are feigning shock at discovering
that political surveys tend to politicize
respondents. Republicans can turn into Democrats, and
vice versa. Will we get equally adept at predicting what
kinds of information, interactions, and initiatives will
turn the tide in a particular election? Facebook and
Google—and many other less famous firms—think
so, and are gearing up for the 2016 election with tools
that go way beyond surveys that can be used for that
purpose.
What do we expect to see in the global security system?
First, we know that a wider variety of actors will
be consuming and generating data on their activities.
Indeed, we know that a wider variety of national and
non-state actors are using predictive models of behavior
that might broadly be considered in the realm of political
violence, ranging from strikes and protests to attacks
and casualties. China, Russia, the European Union, and
the UK, as well as the United States and the United
Nations and the North Atlantic Treaty Organization, all
have substantial forecasting capabilities in the global
realm. Some of these forecasts are made on a weekly basis,
and others are more long term, looking out a couple
of decades. But the fact that predictive heuristics are
now part and parcel of normal statecraft is important
and recent. Moreover, non-state actors are not to be left
out and are beginning to evolve toward greater foresight.
Consider that an International Red Cross that is
able to predict domestic conflicts will be better able to
pre-position supplies and expertise to deal with the human
toll of such conflicts. A monitoring of violent
6 And if it did not, the winner was ridiculed for being marginally smarter
than a chicken.
88 Can We Predict Politics? Toward What End?
Downloadedfromhttps://academic.oup.com/jogss/article/1/1/80/1834930bygueston19March2021
government actions that can predict the safest time to remove
NGO personnel in conflict zones is another example
of a predictive tool that can affect the global security
system in new and beneficial ways.
The global security system is complicated, multilayered,
unknown, and changing. In some ways, it is exactly
like the solar system. However, it may change
more quickly but maybe less dramatically.7
By developing
explanations and subjecting them to critical evaluations,
we learned more. We can learn more again. We
can build Antikythera Mechanisms. Though they may
or may not tell us about the next sociopolitical eclipse in
the international security system, they are likely to help
us get rid of bad ideas.
Acknowledgments
Deborah Avant, Alex Montgomery, and Cassy L. Dorff helped
guide me through this essay. I appreciate their help and absolve
them of my interpretation of their sage advice.
References
Altman, Daniel. 2015. “The Strategist Curse: A Theory of False
Optimism as a Cause of War.” Security Studies 24(2):
284–315.
Ascher, William. 1979. Forecasting, an Appraisal for PolicyMakers
and Planners. Baltimore, MD: Johns Hopkins University
Press.
Ayer, Alfred Jules. 1936. Language, Truth, and Logic. London:
Gollancz.
Beckley, Michael. 2015. “The Myth of Entangling Alliances:
Reassessing the Security Risks of U.S. Defense Pacts.” International
Security 39(4): 7–48.
Beger, Andreas, and Michael D. Ward. 2015. “Irregular Leadership
Changes in 2015.” In Conference on Forecasting and
Early Warning of Conflict and Political Instability. Oslo:
Peace Research Institute.
Blair, Robert, Christopher Blattman, and Alexandra Hartman.
2014. Predicting Local-Level Violence. SSRN. http://ssrn.
com/abstract¼2497153.
Boschee, Elizabeth, Jennifer Lautenschlager, Sean O’Brien, Steve
Shellman, James Starz, and Michael D. Ward. 2015. ICEWS
Coded Event Data (Harvard Dataverse Network [Distributor]
V1 [Version]). http://dx.doi.org/10.7910/DVN/28075.
Brandt, Patrick. 2015. “Forecasting Conflicts: Long and Short
Term Predictions Based on Different Training Set Considerations.”
In Conference on Forecasting and Early Warning of
Conflict and Political Instability. Oslo: Peace Research
Institute.
Brandt, Patrick T., Michael Colaresi, and John R. Freeman. 2008.
“The Dynamics of Reciprocity, Accountability, and Credibility.”
Journal of Conflict Resolution 52(3): 343–74.
Brandt, Patrick T., John R. Freeman, and Philip A. Schrodt.
2014. “Evaluating Forecasts of Political Conflict Dynamics.”
International Journal of Forecasting 30(4): 944–62.
Choucri, Nazli. 1974. “Forecasting in International Relations:
Problems and Prospects.” International Interactions 1(2):
63–68.
Choucri, Nazli, and Thomas W. Robinson. 1978. Forecasting in
International Relations: Theory, Methods, Problems, Prospects.
San Francisco: W. H. Freeman.
Cioffi-Revilla, Claudio. 1996. “On the Likely Magnitude,
Extent, and Duration of an Iraq-UN War.” Journal of Conflict
Resolution 35(3): 387–411.
Colaresi, Michael. 2015. “Using Explicit Model Criticism to
Build and Improve an Open Source, Policy-Relevant Forecast
of Civil Conflict.” In Conference on Forecasting and Early
Warning of Conflict and Political Instability. Oslo: Peace
Research Institute.
Comte, Auguste. 1846. General View of Positivism. London:
Tru¨bner.
Davies, John L., and Ted Robert Gurr. 1998. Preventive Measures:
Building Risk Assessment and Crisis Early
Warning Systems. Lanham, MD: Rowman & Littlefield
Publishers.
de Caritat (Marquis de Condorcet), Marie Jean Antoine Nicolas.
1795. Cre´ation de Esquisse d’un tableau historique des progre`s
de l’esprit humain. Translated and edited by de Caritat
(Marquis de Condorcet), Marie Jean Antoine Nicolas. Philadelphia:
Lang and Uttick.
de Marchi, Scott, Christopher Gelpi, and Jeffrey D. Grynaviski.
2004. “Untangling Neural Nets.” American Political Science
Review 98(2): 371–78.
de Mesquita, Bruce Bueno. 2002. Predicting Politics. Columbus:
Ohio State University Press.
———. 2009. The Predictioneer’s Game. New York: Random
House.
———. 2011. “A New Model for Predicting Policy Choices: Preliminary
Tests.” Conflict Management and Peace Science
28(1): 65–87.
Elster, Jon. 1989. Nuts and Bolts for the Social Sciences. Cambridge:
Cambridge University Press.
———. 2007. Explaining Social Behavior: More Nuts and Bolts for
the Social Sciences. Cambridge: Cambridge University Press.
Feder, Stanley A. 2002. “Forecasting For Policy Making in the
Post–Cold War Period.” Annual Review of Political Science 5:
111–25.
Fraley, Chris, Adrian E. Raftery, and Tilmann Gneiting. 2010.
“Calibrating Multimodel Forecast Ensembles with Exchangeable
and Missing Members Using Bayesian Model
Averaging.” Monthly Weather Review 138(1): 190–202.
Fraley, Chris, Adrian E. Raftery, J. McLean Sloughter, and Tilmann
Gneiting. 2010. ensembleBMA: Probabilistic Forecasting
using Ensembles and Bayesian Model Averaging
(R package version 4.5). http://CRAN.R-project.org/
package¼ensembleBMA.
7 Though there is a genre of science fiction that suggests otherwise.
Stephenson (2015).
MICHAEL D. WARD 89
Downloadedfromhttps://academic.oup.com/jogss/article/1/1/80/1834930bygueston19March2021
Freeman, John R., and Brian L. Job. 1979. “Scientific Forecasts
in International Relations: Problems of Definition and Epistemology.”
International Studies Quarterly 23(1): 113–43.
Freeth, Tony, Yanis Bitsakis, Xenophon Moussas, John H. Seiradakis,
A. Tselikas, H. Mangou, M. Zafeiropouou, R. Hadland,
D. Bate, A. Ramsey, M. Allen, A. Crawley, P. Hockley,
T. Malzbender, D. Gelb, W. Ambrisco, and M. G. Edmunds.
2006. “Decoding the Ancient Greek Astronomical Calculator
Known as the Antikythera Mechanism.” Nature 444(Supplement
7119): 587–91.
Galton, Francis. 1907. “Vox populi.” Nature 75(1949): 450–51.
Gleditsch, Kristian Skrede. 2015. “Predictoing UCDP/PRIO
Civil Wars: Expanding the Inequality and Grievance Model
with Event Data.” In Conference on Forecasting and Early
Warning of Conflict and Political Instability. Oslo: Peace
Research Institute.
Gleditsch, Kristian Skrede, and Michael D. Ward. 2000. “War
and Peace in Space and Time: The Role of Democratization.”
International Studies Quarterly 44(1): 1–29.
———. 2010. “Contentious Issues and Forecasting Interstate Disputes.”
Presented at the 2010 Annual Meeting of the International
Studies Association. New Orleans.
———. 2013. “Forecasting Is Difficult, Especially the Future:
Using Contentious Issues to Forecast Interstate Disputes.”
Journal of Peace Research 50(1): 17–31.
Gurr, Ted Robert. 1970. Why Men Rebel. Princeton, NJ: Princeton
University Press.
Gurr, Ted Robert, and Mark Irving Lichbach. 1986. “Forecasting
Internal Conflict: A Competitive Evaluation of Empirical Theories.”
Comparative Political Studies 19(3): 3–38.
Healy, Kieran. 2015. “Fuck Nuance.” Working Paper. Accessed
December 2, 2015. http://kieranhealy.org/files/papers/fuck-nu
ance.pdf.
Hegre, Ha˚vard. 2008. “Gravitating toward War: Preponderance
May Pacify, but Power Kills.” Journal of Conflict Resolution
52(4): 566–89.
Hegre, Ha˚vard, Bjørn Høyland, and Ha˚vard Nyga˚rd. 2015.
“Dynamic Prediction of Armed Conflict, 2015–2065.”
In Conference on Forecasting and Early Warning of Conflict
and Political Instability. Oslo: Peace Research Institute.
Hegre, Ha˚vard, Joakim Karlsen, Ha˚vard Mokleiv Nyga˚rd,
Ha˚vard Strand, and Henrik Urdal. 2013. “Predicting Armed
Conflict, 2011–2050.” International Studies Quarterly 57(2):
250–70.
Hempel, Carl G. 1935. “U¨ ber den Gehalt von Wahrscheinli
chkeitsaussagen.” Erkenntnis 5: 228–60.
Herz, John H. 1950. “Idealist Internationalism and the Security
Dilemma.” World Politics 2(January): 157–80.
Hildebrand, David K., James D. Laing, and Howard Rosenthal.
1976. “Prediction Analysis in Political Research.” American
Political Science Review 70(2): 509–35.
Hunziker, Philip. 2015. “Online Change Detection for Conflict
Event Data: Towards an Early Alarm System.” In Conference
on Forecasting and Early Warning of Conflict and Political
Instability. Oslo: Peace Research Institute.
Jervis, Robert. 1978. “Cooperation under the Security
Dilemma.” World Politics 30(2): 167–214.
Joshi, Devin, Barry B. Hughes, and Timothy Sisk. 2015.
“Improving Governance for the Post-2015 Sustainable Development
Goals: Scenario Forecasting the Next 50 Years.”
World Development 70: 286–302.
King, Gary, and Langche Zeng. 2001. “Improving Forecasts of
State Failure.” World Politics 53(4): 623–58.
Koubi, Vally, and Tobias Bo¨hmelt. 2014. “Grievances, Economic
Wealth, and Civil Conflict.” Journal of Peace Research
51(1): 19–33.
Kuhn, Thomas S. 1962. The Structure of Scientific Revolutions.
Chicago: University of Chicago Press.
Lautenschlager, Jennifer, Steve Shellman, and Michael D. Ward.
2015. ICEWS Coded Event Aggregations [Harvard Dataverse
Network (Distributor) V1 (Version)]. http://dx.doi.org/10.
7910/DVN/28117.
Liff, Adam P., and G. John Ikenberry. 2014. “Racing toward
Tragedy? China’s Rise, Military Competition in the Asia Pacific,
and the Security Dilemma.” International Security
2(Fall): 52–91.
Lustick, Ian, Sean O’Brien, Steve Shellman, Timothy Siedlecki,
and Michael D. Ward. 2015. ICEWS Events of Interest Ground
Truth Data Set [Harvard Dataverse Network (Distributor) V1
(Version)]. http://dx.doi.org/10.7910/DVN/28119.
Mahoney, James. 2015. “Process Tracing and Historical
Explanation.” Security Studies 24(2): 200–218.
Metternich, Nils W., Cassy Dorff, Max Gallop, Simon Weschle,
and Michael D. Ward. 2013. “Anti-Government
Networks in Civil Conflicts: How Network Structures Affect
Conflictual Behavior.” American Journal of Political Science
57(3): 892–911.
Mill, John Stuart. 1843. A System of Logic. London: John W.
Parker, West Strand.
Milstein, Jeffery S. 1974. Dynamics of the Vietnam War: A
Quantitative Analysis and Predictive Computer Simulation.
Columbus: Ohio State University Press.
Montgomery, Jacob M., Florian M. Hollenbach, and
Michael D. Ward 2012. “Improving Predictions Using
Ensemble Bayesian Model Averaging.” Political Analysis
20(3): 271–91.
———. 2013. Ensemble BMA Forecasting. http://cran.r-project.
org/web/packages/EBMAforecast/EBMAforecast.pdf.
———. 2014. “Calibrating Ensemble Forecasting Models with
Sparse Data in the Social Sciences.” International Journal of
Forecasting, in press.
———. 2015. “Calibrating Ensemble Forecasting Models with
Sparse Data in the Social Sciences.” International Journal of
Forecasting 31(3): 930–42.
O’Brien, Sean P. 2002. “Anticipating the Good, the Bad, and the
Ugly: An Early Warning Approach to Conflict and Instability
Analysis.” Journal of Conflict Resolution 46(6): 791–811.
Pevehouse, Jon C., and Joshua S. Goldstein. 1999. “Serbian
Compliance or Defiance in Kosovo? Statistical Analysis and
Real-Time Predictions.” Journal of Conflict Resolution 43(4):
538–46.
Pilster, Ulrich, and Tobias Bo¨hmelt. 2014. “Predicting the Duration
of the Syrian Insurgency.” Research & Politics 1(2).
doi:10.1177/2053168014544586.
90 Can We Predict Politics? Toward What End?
Downloadedfromhttps://academic.oup.com/jogss/article/1/1/80/1834930bygueston19March2021
Popper, Karl. 1935. Logik der Forschung: Zur Erkenntnistheorie
der Modernen Naturwissenschaft. Number Band 9. Berlin:
Springer Verlag.
Raftery, Adrian E., Tilmann Gneiting, Fadoua Balabdaoui, and
Michael Polakowski. 2005. “Using Bayesian Model Averaging
to Calibrate Forecast Ensembles.” Monthly Weather
Review 133(5): 1155–74.
Ramakrishnan, Naren, Patrick Butler, Sathappan Muthiah,
Nathan Self, Rupinder Paul Khandpur, Parang Saraf, Wei
Wang, Jose Cadena, Anil Vullikanti, Gizem Korkmaz, Chris J.
Kuhlman, Achla Marathe, Liang Zhao, Ting Hua, Feng Chen,
Chang-Tien Lu, Bert Huang, Aravind Srinivasan, Khoa Trinh,
Lise Getoor, Graham Katz, Andy Doyle, Chris Ackermann,
Ilya Zavorin, Jim Ford, Kristen Maria Summers, Youssef
Fayed, Jaime Arredondo, Dipak Gupta, and David Mares.
2014. “‘Beating the News’ with EMBERS: Forecasting Civil
Unrest using Open Source Indicators.” CoRR abs/1402.7035.
http://arxiv.org/abs/1402.7035.
Richardson, Lewis F. 1960. Arms and Insecurity. Chicago and
Pittsburgh: Quadrangle/Boxwood.
Schneider, Gerald. 2015. “The Oracle or the Crowd? Experts
versus Markets in Forecasting Conflict and Cooperation.” In
Conference on Forecasting and Early Warning of Conflict and
Political Instability. Oslo: Peace Research Institute.
Schneider, Gerald, Nils Petter Gleditsch, and Sabine C. Carey. 2010.
“Exploring the Past, Anticipating the Future: A Symposium.”
International Studies Review 12(1): 1–7. http://search.ebscohost.
com/login.aspx?direct¼true&db¼poh&AN¼48465758&site
ehost-live&scope¼site.
Schutte, Sebastian. 2014. “Geography, Outcome, and Casualties:
A Unified Model of Insurgency.” Journal of Conflict Resolution
59(6): 1101–28.
———. 2015. “Regions at Risk: Predicting Conflict Zones in African
Insurgencies.” In Conference on Forecasting and Early
Warning of Conflict and Political Instability. Oslo: Peace
Research Institute.
Singer, J. David, and Michael D. Wallace. 1979. To Augur Well:
Early Warning Indicators in World Politics. Beverly Hills, CA:
Sage Publications.
Sloughter, J. McLean, Tilmann Gneiting, and Adrian E. Raftery.
2010. “Probabilistic Wind Speed Forecasting Using Ensembles
and Bayesian Model Averaging.” Journal of the American
Statistical Association 105(489): 25–35.
Stephenson, Neal. 2015. Seveneves. New York: Harper Collins.
Tetlock, Philip. 2005. Expert Political Judgment: How Good
Is It? How Can We Know? Princeton, NJ: Princeton University
Press.
———. 2010. “Second Thoughts about Expert Political Judgment:
Reply to the Symposium.” Critical Review 22(4): 467–88.
Ulfelder, Jay. 2012. “Forecasting Onset of Mass Killings.” Paper
presented at the Annual Northeast Political Methodology
Meeting at New York University.
Vasquez, John A. 1997. “The Realist Paradigm and Degenerative
versus Progressive Research Programs: An Appraisal of
Neotraditional Research on Waltz’s Balancing Proposition.”
American Political Science Review 91(4): 899–912.
Vincent, Jack E. 1980. “Scientific Prediction versus Crystal Ball
Gazing: Can the Unknown Be Known?” International Studies
Quarterly 24(3): 450–54.
Vrugt, Jasper A., Martyn P. Clark, Cees G. H. Diks, Qinyun
Duan, and Bruce A. Robinson. 2006. “Multi-Objective Calibration
of Forecast Ensembles Using Bayesian Model
Averaging.” Geophysical Research Letters 33: L19817.
Vrugt, Jasper A., Cees G. H. Diks, and Martyn P. Clark. 2008.
“Ensemble Bayesian Model Averaging Using Markov Chain
Monte Carlo Sampling.” Environmental Fluid Mechanics 8(5):
579–95.
Waltz, Kenneth N. 1997. “Evaluating Theories.” American
Political Science Review 91(4): 913–17.
Ward, Michael D. 2015. Monthly Predictions of Conflict in 167
Countries, March 2015. Monthly Report 4/2015 mdwardlab.
com. http://mdwardlab.com.
Ward, Michael D., and Kristian Skrede Gleditsch. 2002.
“Location, Location, Location: An MCMC Approach to
Modeling the Spatial Context of War and Peace.” Political
Analysis 10(3): 244–60.
Ward, Michael D., Nils W. Metternich, Cassy L. Dorff, Max
Gallop, Florian M. Hollenbach, Anna Schultz, and Simon
Weschle. 2013. “Learning from the Past and Stepping
into the Future: Toward a New Generation of Conflict Prediction.”
International Studies Review 16(4): 473–644.
Weidmann, Nils B., and Michael D. Ward. 2010. “Predicting
Conflict in Space and Time.” Journal of Conflict Resolution
54(6): 883–901.
Weil, Herman M. 1974. “Domestic and International Violence:
A forecasting approach.” Futures 6(6): 477–85.
Witmer, Frank D. W., John V. O’Loughin, Andrew M. Linke,
Arlene Laing, and Andrew Gettelman. 2015. “Climate-Sensitive
Violent Conflict Forecasts for Sub-Saharan Africa, 2015–
2050.” In Conference on Forecasting and Early Warning of
Conflict and Political Instability. Oslo: Peace Research Institute.
MICHAEL D. WARD 91
Downloadedfromhttps://academic.oup.com/jogss/article/1/1/80/1834930bygueston19March2021