The notion of quasi-variances, as a device for both simplifying and enhancing the presentation of categorical-predictor effects in statistical models, was developed in Firth and de Menezes (Biometrika, 2004, 65–80). The approach generalizes the earlier ideas of Ridout (GLIM Proceedings, 1989) and of Easton, Peto and Babiker (Statistics in Medicine, 1991 — ‘floating absolute risk’, which has become rather controversial in epidemiology). In this talk I will outline and exemplify the method to show how it can be useful, and discuss its extension to some other contexts such as parameters that may be arbitrarily scaled and/or rotated.

content: "@firthQuasivariancesExtensions2004" -file:@firthQuasivariancesExtensions2004