Food, Familiarity, and Forecasting: Modeling Coups with Computational Methods

The final dissertation is openly available through UCF. In an effort to condense the research to a more approachable format, I have put together a few pages that reflect some of the key arguments and findings.

Abstract

Military coups are the most consequential breakdown of civil-military relations. This dissertation contributes to the explanation and prediction of coups through three independent quantitative analyses. First, I argue that food insecurity is an important determinant of coups. The presence of hunger can generate discontent in society and subsequently alter coup plotter opportunities. Furthermore, I show that the presence of chronic hunger can condition the effect of increasing development. While increasing levels of development have been shown to limit coup proclivity, a state experiencing chronic hunger will recognize the fundamental failure of basic needs provision. As development increases, the presence of chronic hunger in a state will therefore increase the likelihood of a coup when compared to its absence. Findings indicate that food insecurity, and specifically the conditioning influence of chronic hunger, are important explanatory predictors of coups. In the second analysis, I argue that existing tests of the Coup-Contagion hypothesis have not been sensitive to the specific pathways through which coups may diffuse. After a robust analysis of spatial autocorrelation, I derive a novel feature of contagion that is sensitive to both shocks and historical legacy of neighborhood coups. Regression models including coup contagion as a predictor, provide substantive support for my hypotheses. In the final assessment, I synthesize explanatory models and provide a machine learning framework to forecast coups. This framework builds on a growing effort in social science to predict episodes of political instability. I leverage a rolling origin technique for cross-validation, sequential feature selection, and an ensemble voting classifier to provide forecasts for coups at the yearly level. I find that predictive sensitivity to coups is increasing over time using these methods and can result in practical forecasts for policy makers.