IJE vol.33 no.6 © International Epidemiological Association 2004; all rights reserved.
Book Review |
Essential Medical Statistics (2nd edn). Kirkwood BR, Sternc JAC. Malden, MA: Blackwell Publishing, 2003, pp. 288, $52.95 (PB). ISBN 0865428719.
Like the claim that the people who want to be politicians are the least suitable, one may worry that a reader of an elementary statistics textbook is least suitable to perform analyses. However, we believe that a statistically naive researcher who reads this book before collecting data would reach a broadly correct conclusion. Aimed at medical workers and students Essential Medical Statistics is an introduction to the basic methods and ideas of medical statistics and covers the techniques that are regularly used in medical journals. This restriction on scope means that only a low level of mathematical understanding is required; yet the authors ensure that the key ideas are easily comprehended.
The authors exploit the insight that a principal factor in determining what statistical analysis will be performed is the form of the variable of interest: continuous, binary, or longitudinal. Consequently the book is organized according to this classification. Unfortunately, experimental design, the area of statistics that gives the most ‘yield per thought’, is relegated to the back of the book.
The book commences with the basic concepts of statistics and exploratory data analysis in Part A, including population and sample, type of variables, frequencies, histograms, cumulative frequency distribution, quantiles and percentiles, cross-tabulations, and scatter plots.
Chapter 4 introduces the normal distribution, which is admirably justified by its role in the central limit theorem, on which there is a very insightful worked example where the mean is repeatedly calculated by random sampling from a given finite population. The remainder of Part B works through the calculation of means, standard deviations, comparison of two means, analysis of variance, linear regression, diagnostic checking, and transformations. The authors give practical considerations that explain useful statistical techniques rather than being motivated by the principles of statistics, see for example, the summary in Chapter 13 for practitioners of different choices of transformations under particular situations.
Part C is dedicated to binary outcomes and starts with a chapter on basic probability. This just about covers the basics, but not really the concept and importance of independence: could a reader be misled into believing, for example, that the probability of two cot-deaths in a family was equal to the square of the marginal probability of a cot-death? Chapters 15–21 introduce statistical methods for the analysis of binary outcomes: odds ratios, comparing proportions, chi-squared test, and contingency tables. Some advanced concepts such as confounding, stratification, and conditional logistic regression are also explained briefly.
Chapters 22–27 in Part D discuss the analysis of rates and survival times. The ideas of rate and risk, with their links to the Poisson distribution, along with Poisson regression, hazard functions, Kaplan–Meier estimation of survival, and Cox proportional hazards regression are explained briefly using simple mathematics and well illustrated by figures, tables, and examples. The topics cover a wide range within relatively few pages, for example, parametric regression models for survival analysis, such as the Weibull model, are mentioned in the last section of Chapter 27, leaving all the mathematical details and some statistical inference issues aside.
Part E discusses statistical modelling in a more generalized fashion. Chapter 28 introduces the concept of likelihood and its use for confidence intervals and hypothesis testing. Chapter 29 discusses some practical issues in regression modelling. Advanced techniques, such as non-parametric methods, bootstrapping, random-effect models, generalized estimating equations (GEE), and meta-analysis, are also briefly explained in Chapters 30–32 where the focus is appropriately on illustration of the ideas and their applications.
Although comprehensible, worked examples are vital in a book on elementary statistics, we would ask if a chapter that introduces the reader to contemporary statistical software could act as a safeguard against the risk of oversimplified statistical analyses that might otherwise be suggested by the examples. Perhaps some effort should therefore have gone into explaining the unifying and simplifying qualities of the general regression framework with a few more real-life worked examples that presented the statistical art, rather than the mechanics, of model selection.
We were pleased to note that the book does consistently point out the common pitfalls, including confounding, heteroscedasticity, non-linearity, Simpson's paradox, and regression dilution bias, that would not be noticed had the reader unthinkingly used the software currently available.
However, as far as confounding is concerned, only in the very last chapter is the advocacy of Mantel–Haenszel methods that are present throughout most of the book redressed by mentioning its criticisms or alternatives in a balanced comparison with the regression-based alternative.
In summary, we would recommend this book both to medical researchers with little statistical knowledge and to people, including those in China, with a more mathematical background who wish to learn the nuts and bolts of applied medical statistics. The text could easily be understood by a reader whose first language was not English, but Chinese for example!
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||