Abstract
A common problem in models for dichotomous dependent variables is “separation,” which occurs when one or more of a model's covariates perfectly predict some binary outcome. Separation raises a particularly difficult set of issues, often forcing researchers to choose between omitting clearly important covariates and undertaking post—hoc data or estimation corrections. In this article I present a method for solving the separation problem, based on a penalized likelihood correction to the standard binomial GLM score function. I then apply this method to data from an important study on the postwar fate of leaders.
Publisher
Cambridge University Press (CUP)
Subject
Political Science and International Relations,Sociology and Political Science
Reference59 articles.
1. Arguably the best textbook discussion of the problem is that in Venables and Ripley (2002, pp. 198–199), though even that treatment stops short of offering a clear remedy for the problem.
2. Limiting Buyer Discretion: Effects on Performance and Price in Long-Term Contracts
3. State-Sponsored Mass Murder
4. Federalism and the Politics of Macroeconomic Policy and Performance
5. A middle–ground approach, taken by SAS and some others, is to calculate the (theoretically infinite) estimates but warn the researcher that separation is present; S–Plus and R also issue warnings in the case of complete separation, but not when separation is quasicomplete.
Cited by
201 articles.
订阅此论文施引文献
订阅此论文施引文献，注册后可以免费订阅5篇论文的施引文献，订阅后可以查看论文全部施引文献