List price: $20.00 (that's 30% off!)
Used price: $13.80
Buy one from zShops for: $13.10
Used price: $2.89
Buy one from zShops for: $2.90
Used price: $45.99
The authors are well informed and recognized experts in the field. Also there is a need for a book such as this in the interventional field. They should clean it up in their next edition.
It is true that most clinical trials are conducted on restricted sets of subjects who must meet carefully planned inclusion/exclusion criteria. Also, clinical centers are chosen by the sponsor (often a pharmaceutical company) and are picked because they have performed well in the past or have a well known investigator. Therefore, the common paradigm of statistical inference that the subjects are a random sample from a larger population is not even closely true. This seems to be a good argument for not taking formal statistical hypothesis testing too seriously in this context and too much attention has been placed on p-values and rigid decision making in the regulatory arena so I was very much interested in seeing how Norleans would get around these problems.
But instead of recognizing the inadequacies of the models and finding better models, he resorts to a subjective form of decision making based solely on graphics and a misunderstanding of methods such as maximum likelihood and analysis of variance. These problems require increased sophistication for solution not the abandonment of statistical modeling.
Graphical techniques are clearly useful and the pioneering work of Tukey led to a deeper appreciation for good exploratory data analysis techniques beginning in the 1960s. However, Norleans ignores or is ignorant of this literature as he fails to use any of the techniques of Tukey. Rather he concentrates on unconventional and not always very informative graphical displays. These techniques seem to be designed to look at individuals more than groups and can often be very cluttered. There is no reference to Tukey, Cleveland or Tufte, the pioneers in statistical graphics.
I find his condemnation of Jerzy Neyman particularly insulting and it shows both naivety and lack of understanding. He doesn't even know how to spell Neyman's first name (sic Jersey)! He claims that statistical hypothesis testing is based on assuming the truth can be known when in fact it is just the opposite. The framework assumes that there is a true state of nature but the truth can never be known and that we can only express our degree of belief in probabilities that particular decision rules lead to incorrect conclusions. Sample size requirements and decisions are made when these error probabilities are sufficiently small.
Neyman-Pearson theory has also been criticized for using a sharp null hypothesis but extensions of the theory got around that problem as can be seen from the famous text by Lehmann on hypothesis testing in the late 1950s. It is commonplace now to use both composite null and alternative hypotheses and appropriate methods for equivalence teting have been devised by switching the usual null and alternative hypothesis. But Norleans appears to be unaware of these advances. Also the generalization by Wald and others to statistical decision theory based on loss functions or utility functions is likewise overlooked.
On the one hand he reject parametric statistical inference because he does not believe in the use of parametric distributions to represent test statistics but yet he accepts the method of maximum likelihood which he does not recognize as parametric. But maximum likelihood methods are not always robust and in some cases not even sensible.
The analysis of clinical trials is challenging. Mixed effects models, censored survival models, handling of missing data, multiplicity adjustment are among the many tools and issues associated with these problems. Probability and statistics have subtleties that cannot always be simplified. It takes sophistication and the clever use of probability to conquer these problems but Norleans offers us none of this.
He appears to be ignorant of the asymptotic theory of statistics which is based on convergence concepts from probability. The only asymptotic result he mentions is the central limit theorem and that he seems to think is based on a Taylor series approximation. With the Poisson model he mentions the problem of overdispersion but instead of recognizing that with medical data more complex models such as compound Poisson can adequate address the issue and make sense clinically he rejects the methodology itself.
Medical researchers who want to understand statistics and its useful role in medicine and other research would be better served by reading David Salsburg's "A Lady Tasting Tea" than the garbage in this book.
I really have serious issue about Dr. Chow's selection of this book for this series and I cannot understand how he can characterize it so favorably in his introduction.
Used price: $5.00
Collectible price: $7.93
Buy one from zShops for: $7.99
There is virtually NO INFORMATION ABOUT COCAINE OR CRACK and the information in the book can be gotten freely from an addiction anonymous program.
A much better book is Cocaine Solutions: help for Cocaine Abusers and their Families.
Used price: $6.95
List price: $14.95 (that's 30% off!)
Buy one from zShops for: $8.99
The book makes very ineffective use of special text arrangements like summaries and comments in the right-hand margin. Apparently, someone advised the authors that readers, even professional therapists, all have ADHD now, so that they cannot read straight text for more than 120 seconds, and text must be chopped up in chunks that can be given different looks and scattered around. The problem is that the material in the margins is not consistently a summary or anything else. To read the book, you have to keep shifting around from one little chunk to another, and you end up wishing the author would have organized the material.
In summary, the book seemed poorly written and not clinically useful.