Dept of Statistics, University of Auckland

2006-09-18

University of Waikato, I1.09 (I Block, 1st floor)

Abstract: This talk will present a unified framework for fitting many types of regression models. The framework mainly consists of 5 classes of models, called vector generalized linear models (VGLMs), vector generalized additive models (VGAMs), reduced-rank VGLMs (RR-VGLMs), quadratic RR-VGLMs (QRR-VGLMs), and reduced-rank VGAMs (RR-VGAMs). Some of these have special application in community ecology where ordination and calibration are natural outcomes. The models are fitted with iteratively reweighted least squares (IRLS), usually involving Fisher scoring. Smoothing and nonlinear optimization also feature in some of the models. The breadth of the framework is illustrated using my VGAM package for R.