For likelihood-based regression contexts, including generalized linear models, this paper presents a boosting algorithm for local constant quasi-likelihood estimators. Its advantages are the following: (a) the one-boosted estimator reduces bias in local constant quasi-likelihood estimators without increasing the order of the variance, (b) the boosting algorithm requires only one-dimensional maximization at each boosting step and (c) the resulting estimators can be written explicitly and simply in some practical cases.
|ジャーナル||Annals of the Institute of Statistical Mathematics|
|出版ステータス||Published - 2010 4月|
ASJC Scopus subject areas