Over the past few months we have worked on regularly updating our irregular leadership change models and forecasts in order to provide monthly 6-month ahead forecasts of the probability of irregular leadership change in a large number of countries–but excluding the US–worldwide. Part of that effort has been the occasional glance back at our previous predictions, and particularly more in-depth examinations for notable cases that we missed or got right, to see whether we can improve our modeling as a result. This note is one of these glances back: a postmortem of our Yemen predictions for the first half of 2015.
To provide some background, the ILC forecasts are generated from an ensemble of seven thematic1 split-population duration models. For more details on how this works or what irregular leadership changes are and how we code them, take a look at our R&P paper or this longer arXiv writeup.
We made a couple of changes this year, notably adding data for the 1990’s, which in turn cascaded into more changes because of the variation in ICEWS event data volume. This delayed things a bit, but eventually we were able to generate new forecasts for the time period from January to June 2015, using data up to December 2014. Here were the top predictions:
The prediction community owes a great deal to Phil Tetlock, who has been involved in some of the largest and longest evaluations of expert forecasts to date. Tetlock is perhaps most widely known for his two-decade long study of political forecasters, which found that “foxes” (who know a little about a lot of different topics) typically outperform “hedgehogs” (who know a lot about one specific domain) in near-term forecasting. Over the last three years, Tetlock, Barbara Mellers, and Don Moore have led the Good Judgment Project, a large-scale forecasting tournament.
The Good Judgment Project began in mid-2011 as a forecasting tournament between five teams, sponsored by the US Government. (Read early coverage of the project from The Economist here.) Each of these teams had its own methods for leveraging the knowledge of its members to generate accurate forecasts about political and economic developments around the world. For example, the Good Judgment Team now assigns its forecasters to smaller teams of about a dozen members. This allows for collaboration in sharing information, discussing questions, and keeping each member motivated. Example questions include “What will the highest price of one ounce of gold be between January 1, 2014 and May 1, 2014?” or “Who will be the King of Saudi Arabia on March 15, 2014?” Predictions are scored both individually and as a team using Brier scores.
Season 3 of the tournament began this summer, and for the first time forecasters now have access to information from ICEWS, provided directly by the ICEWS project. ICEWS covers five events of interest (insurgency, rebellion, ethnic or religious violence, domestic political crisis, and international crisis) around the world on a monthly basis, and makes forecasts six months into the future. Two current Good Judgment questions related to ICEWS are:
- Will Chad experience an onset of insurgency between October 2013 and March 2014?
- Will Mozambique experience an onset of insurgency between October 2013 and March 2014?
GDELT (gdelt.utdallas.edu) is a global database of events which have been coded from vast quantities of publicly available text that is produced by the world’s new media. It has created a great deal of excitement in the social science community, especially within the field of international relations. But it has had wider visibility as well: in August 2013, there were 150,000 views of a map of protest activity around the world, based on the GDELT database. Event data have been around for several decades, but the GDELT project has generated new interest.
ICEWS is an early warning system designed to help US policy analysts predict a variety of international crises to which the US might have to respond. These include international and domestic crises, ethnic and religious violence, as well as rebellion and insurgency. This project was created at the Defense Advanced Research Projects Agency, but has since been funded (through 2013) by the Office of Naval Research. ICEWS also produces a rich corpus of text which is analyzed with powerful techniques of automated event-data production. Since GDELT and ICEWS are based on similar, though not identical methods and sources, it is interesting to compare them.
ICEWS event data, gray line for stories and black line for events, 2001-2013
One area in which they are most conceptually different is that ICEWS follows a more traditional approach to event data in seeking to encode a chronology of events that reflects in some sense the putative ground truth of what occurred. The figure on the right shows the corpus of stories in ICEWS (gray) and the resulting events (black): total events are fairly stable over time event though the number of media stories increases. GDELT is more concerned with getting a comprehensive catalogue of all media stories (and other text) on reported events, and the corpus of those media stories is increasing exponentially, as the figure below shows. As a result, the number of events in GDELT is also increasing over time, much more so than ICEWS.