Lessons from near real-time forecasting of irregular leadership changes
| Author | Michael D Ward,Andreas Beger |
| DOI | 10.1177/0022343316680858 |
| Date | 01 March 2017 |
| Published date | 01 March 2017 |
| Subject Matter | Research Articles |
Lessons from near real-time forecasting
of irregular leadership changes
Michael D Ward & Andreas Beger
Department of Political Science, Duke University
Abstract
Since 2014, we have been producing regular six-month forecasts of the probability of irregular leadership changes –
coups, rebellions, protests that result in state leader changes – for most countries in the world for the Political
Instability Task Force (PITF). During 2015, we issued new forecasts each month, with a delay as short as five days
and no longer than two weeks into each six-month forecasting window. This article describes the approach we use to
generate our forecasts and presents several examples of how we present forecasts. The forecasts are derived from a
statistical ensemble of seven thematic models, each based on a split-population duration model that aims to capture a
specific argument or related set of covariates. This approach is modular in that thematic models can be swapped out
or new models integrated, and it lessens the need for generalist ‘kitchen sink’ models. Together, the models achieve
high out-of-sample accuracy. Based on our experience, we draw conclusions about the practical, policy, and scientific
aspects of this and similar undertakings. These include thoughts on how to evaluate and present forecasts, the
potential role of ensembles in model comparison, the role of ensembles and prediction in causal research, and
considerations for future efforts in forecasting and predictive modeling.
Keywords
coups, EBMA, ensemble, forecasting, ILC, PITF, prediction, protest, rebellion, split-population duration regression
Introduction
What if the President of Russia is overthrown this year?
How will Turkey’s policies towards Syria and mass
migration to Europe change if the military were to reas-
sert power? In 2014, there were five instances of similar
unexpected and irregular transitions between sitting
leaders of states, including one in Thailand and two in
Burkina Faso, which has had another in 2015. Ukraine is
still embroiled in a conflict that started when the then
president was overthrown in a revolution, as is Yemen
after an irregular change in 2015. Although the means by
which the leaders of these countries were overthrown
vary, ranging from military coups and revolutions to
armed rebellion, we treat them as a common outcome,
which we call irregular leadership changes,orILCfor
short (Beger, Dorff & Ward, 2016).
Over the past several years, we have been providing
regular global six-month forecasts of ILCs with a delay
of as few as five days from the start of the forecast period
to delivery of a forecast. The underlying focus on ILCs
as an outcome of interest, rather than the mechanisms
through which they can occur, is an important empiri-
cal addition to the literature. Processes of change are
often theoretically not separable (Bueno de Mesquita &
Smith, 2015). The coup in Mali in 2012, for example,
was driven by dissatisfaction with how the civilian gov-
ernment handled the Tuareg rebellion in the north,
while the most recent Thai coup involved mass pro-
tests. Trying to predict coups without considering the
impact of mass protests or armed rebellion and vice
versamaythusnotbeasfruitfulasconsideringthejoint
outcome.
From a more general perspective, our effort flows
from research into the duration of political authorities,
regimes, and polities (Eckstein & Gurr, 1975). The basic
Corresponding author:
michael.don.ward@gmail.com
Journal of Peace Research
2017, Vol. 54(2) 141–156
ªThe Author(s) 2017
Reprints and permission:
sagepub.co.uk/journalsPermissions.nav
DOI: 10.1177/0022343316680858
journals.sagepub.com/home/jpr
thrust of Eckstein and Gurr was to characterize situations
by authority relations, defining underlying dimensions as
a way of getting at the duration of different political
constellations. ‘Persistence and change in political sys-
tems, 1800–1971’ (Gurr, 1974) first developed the idea
of calculating the half-lives of different sorts of political
polities, regimes, and authorities. We apply this basic
notion to produce forecasts of their ‘half-life’.
1
Forecasting ILCs is also important in practice. To
the extent that leaders of foreign states matter for
foreign policy, we should care whether one is facing
a risk of irregular and thus unexpected removal. Not
only are ILCs often a result of violent and damaging
processes, but the instability they cause can spark sig-
nificant levels of violence and civil war (Powell &
Thyne, 2011: 256) and also has negative economic
effects (Alesina et al., 1996).
Forecasting rare political events
ILCs are rare events. Over the past 15 years, on averagewe
see between two and four per year (Figure 1). One might
question the utility of trying to develop models that will
help to understandor to forecast rare events; but it is often
said that reality is a low probability event. ILCs are low
probability,high impact eventsthat often change the world
in consequential ways. Imagine the impact of a coup
tomorrowin Russia or Turkey. Suchevents would certainly
change theconfiguration of regionalpolitics in Europe and
the Middle East,let alone what might be the consequences
for citizens living in these states. Despite – or because of –
the rarityof such events, it is useful to developbetter under-
standings of such occurrences. Toward that end, we
develop a model of the duration of incumbent leaders.
Our forecasts are based on an ensemble of several
split-population duration regression models using
monthly data from 1991 to the present (actually to July
2015). The basic idea is to develop a model which is a
mixture that can apply to situations where the risk of an
irregular turnover is rare (for example, New Zealand)
and have a separate but integrated model that applies
to cases in which the risk of irregular turnover is high
(e.g. Fiji or Ukraine). Such split-population models are
widespread in the medical literature, but have only
recently been adapted and employed in the social
sciences. We present only a relatively brief summary of
the methodology, but the Online appendix contains a
more detailed description and estimation results.
2
Figure 1. Spatial and temporal distribution of ILCs, 1991–2015
1
Long ago in a distant galaxy, Ward was a student of Gurr, but it
took him a long time to return to studying the persistence of political
regimes.
2
See also Beger et al. (2016) and the spduration R package.
142 journal of PEACE RESEARCH 54(2)
To continue reading
Request your trial