The Diffusion of Instability in Authoritarian Regimes

Open Access
Brawner, Thomas W
Graduate Program:
Political Science
Doctor of Philosophy
Document Type:
Date of Defense:
November 10, 2016
Committee Members:
  • Joseph Wright, Dissertation Advisor
  • Joseph Wright, Committee Chair
  • Donna Bahry, Committee Member
  • Bumba Mukherjee, Committee Member
  • Johannes Fedderke, Outside Member
  • Diffusion
  • Authoritarian Regimes
  • Democratic Transition
  • Regularization
  • Machine Learning
  • Prediction
  • Coups
Spatial models of political instability typically constrain the trigger events of interest to lagged values of the dependent variable, and such a lag is often intended to capture positive diffusion of the event. This dissertation explores a wider range of spatiotemporal lags in descriptive and predictive models of authoritarian regime failure, democratic transition, and coups d'etat. Expanding the set of spatiotemporal lags allows for more precise directional hypotheses of transnational learning. The results indicate that lags of autocratic transitions, for example, may inhibit the realization democratic transition, while lags of democratic transition may facilitate it. The dissertation is also inspired by a methodological concern for model generalizability, and therefore the importance of model validation and regularization is given extensive treatment. These principles are applied to descriptive models, in which the primary objective is to explain, and to predictive models, in which the primary objective is to predict unseen data. Moreover, these principles are applied to linear models commonly employed in political science and to more flexible machine learning models that are less commonly used in the field. In the end, a template for improving the generalizability of model estimates is provided, and straightforward tools for interpreting those estimates are explored.