International Journal of Epidemiology 2002;31:701-702
© International Epidemiological Association 2002
Book Review |
Models for Repeated Measurements (Second Edition). JK Lindsey. Oxford: Oxford University Press, 1999, pp. xx + 515, £45.00, $75.00 (HB). ISBN: 0-19-850599-0.
This, the enlarged second edition, expands the first edition's four parts in places and adds a fifth part on design issues, leading to a 25% increase in length.
Briefly, Part I describes the basic concepts of modelling repeated measures data, showing how familiar and unfamiliar distributions can be generalized to describe complex data structures. Part II details models for continuous responses, focusing on longitudinal data and incorporating discussions of heterogeneity and non-normality. This leads naturally to Part III, where models for over-dispersed and then longitudinal discrete data are considered. Part IV is devoted to duration data and comprises a discussion of frailty models and models for event histories. Finally, Part V discusses design issues and missing data.
Two other significant differences from the first edition are the omission of the section on generalized estimating equations (GEE) and the shorter, less comprehensive bibliography.
The omission of the section on GEE is regrettable in my opinion, and indicative of the book's focus on parametric models, interpreted through the ‘likelihood’ school of inference1 and selected using Akaike2 information criterion (AIC). While Lindsey claims the conclusions arrived at using AIC ‘always appear to be more plausible’ it is not clear why in general this should be so, especially when the X2 calibrated likelihood ratio test is applicable.
The strength of this book is the detailed analysis of a wealth of examples from a wide variety of sources. Additionally, numerous data sets are provided, so readers can experiment for themselves. While few computational details are given, R code for both the examples and thought provoking exercises can be found at www.luc.ac.be/cemstat.
My reservations centre around Lindsey's slightly idiosyncratic approach, which is captured by an unusual phrase in the preface, ‘... much of the most interesting new material comes from articles rejected by well-known statistical journals’. This, and the implicit criticism of established texts, suggests that uncritical adoption of the methods presented here may not find universal favour.
The technical difficulties of the book mean that epidemiologists without a strong statistical background will struggle. Nevertheless, when faced with a novel problem, this book is a useful place to seek appropriate parametric approaches. Thus, I would recommend the library have a copy, but not choose it for my bookshelf.
References
1 Lindsey JK. Some statistical heresies. Statistician 1999;48:1–40.
2 Akaike H. Information theory and an extension of the maximum likelihood principle. In: Petrov BN, and Csàka F. (eds). Second International Symposium on Inference Theory. Budapest: Akadémiai Kiadó, pp. 267–81.
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||