Diagnostics in a simple correspondence analysis model: An approach based on Cook's distance for log-linear models Articles uri icon

authors

  • MARTIN APAOLAZA, NIRIAN

publication date

  • April 2015

start page

  • 175

end page

  • 189

volume

  • 136

international standard serial number (ISSN)

  • 0047-259X

electronic international standard serial number (EISSN)

  • 1095-7243

abstract

  • Diagnostics have not received much attention in the literature of simple correspondence analysis models. Since Cook's distance was defined to identify influential observations of the linear regression model, it has been extended to different models, in particular to log-linear models. In this paper we provide the asymptotic distribution of Cook's distance of any kind of log-linear models and also a method for diagnostics, based on it. By using Goodman's RC (K) model as a log-linear model to approximate the ordinary simple correspondence analysis procedure, we follow a Cook's distance approach to identify influential cells and three examples illustrate the performance of this method. (C) 2015 Elsevier Inc. All rights reserved.