New research addresses a persistent challenge in political science: how to handle unobserved unit-level heterogeneity when analyzing rare events data. The study tackles the debate surrounding fixed effects models by clarifying their main limitation—not inaccurate coefficient estimates but biased marginal effects.
When examining only event-experiencing units, these models produce inflated baseline risk estimates that skew predictor effect calculations for rare outcomes like civil war onset or genocide occurrence. This insight leads to a novel solution: the Penalized Maximum Likelihood Fixed Effects (PML-FE) estimator preserves all data points while still estimating finite fixed effects.
Testing through Monte Carlo simulations reveals how PML-FE performs compared to other methods in practice, not just theoretically. The approach's effectiveness is demonstrated using real-world civil war onset data as an illustrative example.
| Fixed Effects in Rare Events Data: A Penalized Maximum Likelihood Solution was authored by Scott Cook, Jude Hays and Rob Franzese. It was published by Cambridge in PSR&M in 2020. |
