research

Dr. Motulsky is an MD who is also a Professor of Pharmacology and President of his own software company. The book's title suggests that he can make biostatistics intuitive for non-statisticians (e.g. physicians, clinicians and nurses). After reading through it he has made a believer out of me! He introduces concepts through examples and touches on most of the important statistical methods that are used in the medical literature. While the book could be used as a classroom text, it seems to me to be more suited as a reference source for medical researchers who want to understand the statistics described in research papers. Although not a statistician by training, Dr. Motulsky has a good understanding of statistical methods and principles and exhibits his wisdom and experience throughout the book. He is deliberate at keeping things simple and to the point. He points out that he intentionally uses fake examples and modifies real examples for simplification of exposition. He avoids mathematics as much as possible. the preface and the introduction are very well written and the reader should read both before reading the rest of the text.

My usual concern with such books is that concepts are oversimplified and the presentation is too cook-bookish. Amazingly that is not the case here. Professor Motulsky carefully explains concepts such as confidence intervals, p-values, multiple comparison issues, Bayesian thinking and Bayesian controversy in a way that should be understandable to his intended audience.

Proportions and the binomial distribution are introduced early. Advanced topics such as sequential methods, survival curves and logistic regression are tackled. These subjects are important in medical research but are often avoided in elementary books. To his credit he also does a very good job of introducing the concepts of sensitivity and specificity. Hypothesis testing is introduced at the same time which makes a lot of sense since for a particularly hypothesis test the specificity and the sensitivity are related to the type I and type II errors. It is a good way for those familiar with medical applications where specificity and sensitivity may be intuitive concepts, to become comfortable with the less familiar null and alternative hypotheses and their associated error probabilities.

Professor Motulsky writes eloquently and this appears to be appreciated by the readers, judging from the other reviews that I have seen on Amazon. Having said all this you might wonder why I didn't give it 5 stars. I found a few things that could have been done better.

I am not completely happy with the way probability is introduced through the binomial distribution and here the wording could be improved. He writes "Mathematicians have developed equations, known as the binomial distribution, to calculate the likelihood of observing any particular outcome when you know the proportion in the overall population." Actually the binomial distribution is a probability distribution (which he has not yet defined as he first uses the term distribution). The equation is a statement that the probability of an event (e.g. exact 7 heads in 10 coin flips) is given by equation (2.2) on page 19 with N=10 and R=7 and p=1/2 (assuming a fair coin).

Another area that could be omitted or else improved is the discussion of Bayesian ideas. Bayes theorem is presented in a limited context related to the example of sensitivity and specificity. While I do think that some Bayesian ideas are well brought out the breadth of applications is missing. Some comparison of the frequentist and Bayesian approaches and philosophy are correctly described but the discussion is too brief to provide good insight. The p-value is strictly a frequentist concept. Motulsky relates it to the Bayesian idea of posterior odds for the null hypothesis to be true. While there is such a formal mathematical relationship, they are conceptually quite different. This is just like relating likelihood to posterior probability. Mathematically the likelihood and posterior probability are related through Bayes theorem as posterior = likelihood x prior but although likelihood is an acceptible frequentist concept posterior probability is not. A real understanding requires some knowledge of the sample space for a frequentist and the treatment of parameters as random quantities by Bayesians. I think this may be something that requires a little more mathematical sophistication than is intended for this readership.

There are a few topics that get little or no treatment but deserve more in a biostatistics texts. These include missing data, resampling methods, hierarchical Bayesian models and longitudinal - repeated measures data. Perhaps we will see intuitive descriptions of some of these topics in the second edition.

I'm a practicing physician who has found it necessary to try to educate myself on the use of biostatistics in the medical literature. I have read over 20 books on biostatistics. This is clearly the best. It is written so that even the non-statistician can understand the concepts, and explains the statistical approach and rationale without scaring the reader away with arcane formulas. It is very logical in its progression and addresses the errors and assumptions that doctors make when trying to evaluate a paper. This book should be required reading not only by every medical student, but by anyone who attempts to write or interpret the medical literature.

I used this book when writing a manuscript for publication. I stumbled across the author's web site, and found his approach remarkably useful. The book is clear and concise. It encompasses all general statistical methods needed for biomedical publishing for both basic and clinical research. It serves as an indespensible introduction for those, who like me, either never understood biostats as taught in medical school or who have not used these skills often enough to develop adequate familiarity with the methods and their application. Highly recommended.