Customer Reviews

Multivariate Statistical Modelling Based on Generalized Linear Models (Undergraduate Texts in Mathematics)

5 star:		(2)
4 star:		(1)
3 star:		(0)
2 star:		(0)
1 star:		(0)

 

 

Back in 2000 Stephen Fienberg gave a talk at the University of California at Irvine on the 2000 census and his book "Who Counts". After the talk I went to dinner with him, my colleague Bob Newcomb and Anita Iannucci. Driving to dinner Bob ask Steve for a recommendation on a multivariate textbook. A number of choice were mentioned. Bob's favorite was Cooley and Lohnes but that was a bit dated. He was definitely looking for an applied text and not a theoretical one. I learned my multivariate analysis out of the first edition of Ted Anderson's book. But that is traditional multivariate Gaussian theory and is not at all an applied text. I always liked Gnanadesikan's book and I mentioned that. Srivastava and carter is an applied text that I like and there are many other choices.
I don't recall many of Fienberg's suggestions but I do distinctly recall that he did say that now you can teach it as a special case of the generalized linear models. The idea seemed to make sense to me but I couldn't picture the details. This book is apparently the book Fienberg had in mind. He might have been thinking about the first edition because this second edition was not out then.

The book is very applied and modern and covers many important topics for biostatisticians. Coverage includes multicategorical responses, semi and nonparametric modelling, time series and longitudinal data, random effects models, state space models including Kalman Filters and nonlinear models, and survival analysis. This is not traditional multivariate data but covers many type of multivariate data and models that do not fit the standard multivariate Gaussian theory.

Chapter 4 on selecting and checking models seems to deal with the classical linear models taking a non-standard approach through the methods of generalized linear models.

Excellent text for an applied course and for a reference book. It also covers hidden Markov models and Bayesian methods (including the MCMC implementation and the WinBugs software).

[1] Studying bioinformatics? You must be familiar with multivariate analysis. This book is absolutely an important reference.
[2] A researcher of statistical pattern recognition? Without doubt, you need this up-to-date book to stuff your toolbox.

Great book! Presents clear information about statistical computational details, as well as a number of nonstandard models (including those of Tutz's original work). The book has a transparent build-up, from more easy modeling exercises to advanced applications. I like best the part on generalized linear time series modeling, using the extended Kalman filter in the context of the EM algorithm. The only critique I have concerns the handling of (the variance of) the measurement error term in the associated generalized state space model (this measurement error should be modeled as a constrained martingale difference).