@article{JSSv010i04,
title={Analysis of Robust Quasi-deviances for Generalized Linear Models},
volume={10},
url={https://www.jstatsoft.org/index.php/jss/article/view/v010i04},
doi={10.18637/jss.v010.i04},
abstract={Generalized linear models (McCullagh and Nelder 1989) are a popular technique for modeling a large variety of continuous and discrete data. They assume that the response variables Y<sub>i</sub> , for i = 1, . . . , n, come from a distribution belonging to the exponential family, such that E[Y<sub>i</sub> ] = µ<sub>i</sub> and V[Y<sub>i</sub> ] = V (µ<sub>i</sub> ), and that η<sub>i</sub> = g(µ<sub>i</sub> ) = x<sub>i</sub><sup>T</sup>β, where β ∈ IR <sup>p</sup> is the vector of parameters, x<sub>i</sub> ∈ IR <sup>p</sup>, and g(.) is the link function.</p><p>The non-robustness of the maximum likelihood and the maximum quasi-likelihood estimators has been studied extensively in the literature. For model selection, the classical analysis-of-deviance approach shares the same bad robustness properties. To cope with this, Cantoni and Ronchetti (2001) propose a robust approach based on robust quasi-deviance functions for estimation and variable selection. We refer to that paper for a deeper discussion and the review of the literature.},
number={4},
journal={Journal of Statistical Software},
author={Cantoni, Eva},
year={2004},
pages={1–9}
}