An Empirical Evaluation of Theories in Government Formation and Duration

Abstract

Theories of government formation and duration have been subjected to rigorous empirical testing. Numerous articles each argue for their own approach, pointing to an empirical model with significant explanation power. Being in a position to gather data not used in government formation and duration classics, this thesis has the benefit of testing their predictive power out-of-sample. The main question is: Can the models predict what they say they can explain when facing unseen observations? Another question at the forefront of explaining the life cycle of governments is the trade-offs regarding a cornerstone in political science methodology - decreasing or increasing complexity of the models which are tested. Recent literature on the life cycle of governments have pulled in the respective directions. This thesis sets out to empirically evaluate the trade-off between parsimony and complexity. The empirical evaluation can be summed up in two main points: Only one out of the four original models predict new observations better than how they predict the original sample. Complex modeling of government formation yields more predictive power compared to the more parsimonious approach. Parsimonious modeling of government duration has more predictive power compared to complex modeling.