Linear Statistical Models

Rating:
Summary: The way linear models should be taught
Review: Approach of linear models from a geometric point of view is often addressed in a single lecture or chapter and does not provide much assistance in the understanding of the material. This book, however, presents all material usually covered in a linear models book from this approach. For those who have learned models both ways, proofs and applications using these geometrical concepts are much less cumbersome than the standard matrix algebra manipulation. Additionally, having a solid understanding of this material is a greater help in the understanding of more advanced topics. As stated, the book is dense and previous exposure is useful if the reader is not assisted by a knowledgeable instructor.

Rating:
Summary: Thanks
Review: I thank Mr. Peng for saying the book "would serve as a decent linear models reference." J. Stapleton

Rating:
Summary: Elegant and practical treatment
Review: I've used this text twice for a one-semester graduate course in linear models, emphasizing Chapters 1,2,3,5,6, and would use it again. It is outstanding.

The best feature of the book is its consistent theme: a least squares estimator is an orthogonal projection onto a subspace, which can be evaluated by orthogonal decomposition of the subspace. This gives the subject the elegance of pure mathematics, while at the same time making complex topics such as two-way and three-way analysis of variance readily accessible.

The second-best feature of the book is the extensive collection of problems. Most are just at the right level, not simply cookbook plug-in type exercises, but problems that require understanding, yet not too difficult for the average student, who is typically not a math major. A few of the problems require statistical software, but most do not.

The only negative aspect of the book is the large number of errata, although this does have the advantage of teaching the students to adopt a healthy degree of skepticism.