Premium
Theory & Methods: Semi‐parametric estimation of ROC curves based on binomial regression modelling
Author(s) -
Lloyd Chris J.
Publication year - 2002
Publication title -
australian and new zealand journal of statistics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.434
H-Index - 41
eISSN - 1467-842X
pISSN - 1369-1473
DOI - 10.1111/1467-842x.00209
Subject(s) - mathematics , receiver operating characteristic , statistics , parametric statistics , regression analysis , binomial (polynomial) , model selection , negative binomial distribution , regression , poisson distribution
This paper presents a method of estimating a receiver operating characteristic (ROC) curve when the underlying diagnostic variable X is continuous and fully observed. The new method is based on modelling the probability of response given X , rather than the distribution of X given response. The method offers advantages in modelling flexibility and computational simplicity. The resulting ROC curve estimates are semi‐parametric and can, in principle, take an infinite variety of shapes. Moreover, model selection can be based on standard methods within the binomial regression framework. Statistical accuracy of the curve estimate is provided by a simply implemented bootstrap approach.