Unpacking the Unpredictable: Using NLP and LLMs to Examine Cabinet Politics and Responses to Stochastic Events in Presidential Democracies

About the project

**Principal Investigador:** Bastián González-Bustamante Drawing upon strands of the stochastic events concept from the event-based approach of the literature on government survival, the main argument of this project is that these events are not critical in themselves. Social protests, economic crises, media scandals and even natural disasters could be stochastic events. Their potential to become critical and affect cabinet stability depends on a number of institutional factors and actors, but in particular, on the president in presidential systems. Thus, a cabinet duration should be related to the type of unexpected crises governments face. Using a novel dataset that comprises indicators on ministerial turnover and resignation calls in 12 Latin American presidential democracies from the mid-1970s to the early 2020s, this project offers a proof-of-concept focused on four relevant country cases (i.e., Brazil, Venezuela, Costa Rica, and Mexico). It creates and evaluates specific indicators for stochastic events using cutting-edge machine learning and generative artificial intelligence techniques, including large language models (LLMs), to better understand the differentiated impact of these crises on executive politics and cabinet stability.