Convergence Problems in Logistic Regression
2003; Wiley; Linguagem: Inglês
10.1002/0471475769.ch10
ISSN1940-6347
Autores Tópico(s)Advanced Scientific Research Methods
ResumoChapter 10 Convergence Problems in Logistic Regression Paul Allison, Paul AllisonSearch for more papers by this author Paul Allison, Paul AllisonSearch for more papers by this author Micah Altman, Micah AltmanSearch for more papers by this authorJeff Gill, Jeff GillSearch for more papers by this authorMichael P. McDonald, Michael P. McDonaldSearch for more papers by this author Book Author(s):Micah Altman, Micah AltmanSearch for more papers by this authorJeff Gill, Jeff GillSearch for more papers by this authorMichael P. McDonald, Michael P. McDonaldSearch for more papers by this author First published: 12 December 2003 https://doi.org/10.1002/0471475769.ch10Citations: 33Book Series:Wiley Series in Probability and Statistics Series Editor(s): Walter A. Shewhart, Walter A. ShewhartSearch for more papers by this authorSamuel S. Wilks, Samuel S. WilksSearch for more papers by this author AboutPDFPDF ToolsRequest permissionExport citationAdd to favoritesTrack citation ShareShareShare a linkShare onEmailFacebookTwitterLinkedInRedditWechat Summary Chapter ten shows how logistic regression models can produce inaccurate estimates or fail to converge altogether because of numerical problems. The chapter then provides methods to detect false convergence, and to make accurate estimation of logistic regressions. Topics include: maximum likelihood estimation of logistic regression computational problems arising from complete separation and quasi-complete separation a comparison of the ability of popular statistical packages to detect and report false convergence methods for avoiding quasi-separation: omitting variables, combing categories methods for robust and accurate logistic regression: exact inference, Bayesian estimation, penalized maximum likelihood estimation Citing Literature Numerical Issues in Statistical Computing for the Social Scientist RelatedInformation
Referência(s)