Separation phenomena logistic regression

Authors

  • Ikaro Daniel de Carvalho Barreto Universidade Federal de Sergipe
  • Suzana Leitão Russo
  • Gutemberg Hespanha Brasil
  • Vitor Hugo Simon

DOI:

https://doi.org/10.47059/geintecmagazine.v4i1.378

Abstract

This paper proposes an application of concepts about the maximum likelihood estimation of the binomial logistic regression model to the separation phenomena. It generates bias in the estimation and provides different interpretations of the estimates on the different statistical tests (Wald, Likelihood Ratio and Score) and provides different estimates on the different iterative methods (Newton-Raphson and Fisher Score). It also presents an example that demonstrates the direct implications for the validation of the model and validation of variables, the implications for estimates of odds ratios and confidence intervals, generated from the Wald statistics. Furthermore, we present, briefly, the Firth correction to circumvent the phenomena of separation.

Author Biography

Ikaro Daniel de Carvalho Barreto, Universidade Federal de Sergipe

Departamento de Estatística e Ciências Atuariais, Modelos Lineares Gereneralizados, Séries Temporais.

Published

2014-03-17

Issue

Section

Articles