Annette J. Dobson and Adrian G. Barnett: An introduction to generalized linear models
Annette J. Dobson and Adrian G. Barnett: An introduction to generalized linear models
复制标题
Annette J. Dobson 和 Adrian G. Barnett：广义线性模型简介
作者：
Christian Kleiber
这是关于广义线性模型(GLMS)的著名介绍性文本的第三版，现在由第二作者A.G.Barnett撰写。与前一版一样，第三版有了相当大的扩展：第二版提供了关于名义和顺序回归的新章节，也提供了一些生存分析和纵向数据分析，而当前版本增加了另外三章，都是关于贝叶斯分析和方法论。具体地说，现在有一章涵盖了MCMC基础知识。这些众多扩展的一个副作用是，标题现在有点误导，因为并不是这里涉及的所有模型都是GLM，尽管在某种意义上，那些不是GLM的模型接近GLM并重用GLM方法。例子包括多项式或有序逻辑回归。因此，新版本更合适的标题应该是“回归建模导论”。与早期版本相比，还有更多的代码展示了如何在各种软件包中使用GLM方法，特别是R和Stata；贝叶斯章节使用了R和WinBUGS的组合。有时，R代码比需要的更难看。从一个网页上可以获得大约40个数据集。数据和例子大多来自生物统计学和相关领域，因此这本书对社会科学教师可能不那么有吸引力。不可避免的是，对一些话题的讨论相当简短。例如，只简单地提到了计数数据中的过度离散性这一重要问题，以及负二项分布作为补救措施之一。因此，打算更详细地介绍计数数据的教师可能不得不利用其他来源。
This is the 3rd edition of the well-known introductory text on generalized linear models (GLMs), now with a second author, A.G. Barnett. Like the preceding edition the 3rd edition has been considerably expanded: the 2nd edition provided new chapters on nominal and ordinal regression, and also some survival analysis and analysis of longitudinal data, while the current edition adds three further chapters, all on Bayesian analysis and methodology. Specifically, there is now a chapter covering MCMC basics. A side effect of these numerous extensions is that the title is now somewhat misleading, in that not all models covered here are GLMs, although those that are not are, in a sense, close to GLMs and reuse GLM methodology. Examples include multinomial or ordinal logistic regression. A more appropriate title for the new edition would thus be “Introduction to Regression Modelling.” Compared to earlier editions there is also more code showing how to use GLM methodology in various software packages, notably R and Stata; the Bayesian chapters employ a combination of R and WinBUGS. On occasion, the R code is somewhat uglier than need be. Some 40 data sets are available from a Web page. Data and examples are mostly from biostatistics and related fields, thus the book will perhaps be less appealing to instructors in the social sciences. Inevitably, discussion of some topics is rather brief. For example, the important issue of overdispersion in count data is only briefly mentioned, along with the negative binomial distribution as one of its remedies. Thus instructors intending to cover count data in some detail might have to draw on additional sources.