Separation and Rare Events

• DOI: 10.1017/psrm.2020.46
• Publication Date: 2020-12-11T21:58:01Z
• Authors: Liam F. Beiser-Mc...
• Keywords: Firth; Rare events; Separation (statistics)

Abstract When separation is a problem in binary dependent variable models, many researchers use Firth's penalized maximum likelihood order to obtain finite estimates (Firth, 1993; Zorn, 2005; Rainey, 2016). In this paper, I show that approach can lead inferences the opposite direction of when number observations are sufficiently large and both independent variables rare events. As datasets with events frequently used political science, such as dyadic data measuring interstate relations, lack awareness may inferential issues. Simulations an empirical illustration “weakly-informative” prior distributions centered at zero, for example, Cauchy suggested by Gelman et al. (2008), avoid issue. More generally, results caution be aware how choice interacts structure their data, estimating models presence separation.