Abstract:
Examining onsets of political instability in countries worldwide from 1955 to 2003, we develop a model that distinguishes countries that experienced instability from those that remained stable with a two-year lead time and over 80% accuracy. Intriguingly, the model uses few variables and a simple specification. The model is accurate in forecasting the onsets of both violent civil wars and nonviolent democratic reversals, suggesting common factors in both types of change. Whereas regime type is typically measured using linear or binary indicators of democracy/autocracy derived from the 21-point Polity scale, the model uses a nonlinear five-category measure of regime type based on the Polity components. This new measure of regime type emerges as the most powerful predictor of instability onsets, leading us to conclude that political institutions, properly specified, and not economic conditions, demography, or geography, are the most important predictors of the onset of political instability.
A Global Model for Forecasting Political Instability
by Goldstone, Jack A.; Bates, Robert H.; Epstein, David L.; Gurr, Ted Robert; Lustik, Michael B.; Marshall, Monty G.; Ulfelder, Jay; Woodward, Mark
The cost is $41.89 + tax.
http://www.ingentaconnect.com/content/bpl/ajps/2010/00000054/00000001 /art00013;jsessionid=5oxe60hemkwc.alexandra
My Comment:
I haven’t read this, yet, but wanted to alert blog readers because it attempts forecasting of a political factor and considers other plausably related factors.
They say, “Intriguingly, the model uses few variables and a simple specification.” But of course this is common for forecasting models, as opposed to scientific models (which are trying to find a model and coefficients with underlying stability over varying circumstances); for instance, see http://www.forecastingprinciples.com/files/standardshort.pdf (Principles 6.6 and 7.1).
The sentence “The model is accurate in forecasting the onsets of both violent civil wars and nonviolent democratic reversals, suggesting common factors in both types of change.” suggests that they started out trying to predict them separately and ended up modifying their model to predict the combined “instability”. This would mean that they were data dredging, which is fine if the dredging was done without using the holdout/test data. But I would want to look closely at the sequence by which the model was derived and tested and not just the methodology for fitting coefficients.
