<< 1 >>
Rating: 
Summary: Interesting... but needs a better edit
Review: The book presents a very interesting method: reducing power analysis to the F distribution. The authors provide very compelling and convincing arguments for the use of power analysis. At times you feel as if their arguments are not well referenced or backed-up, however, especially if you have read a number of technically-oriented statistical texts. Nevertheless, the arguments are provided a good intuitive feel.
The one problem with the book is its editing, or lack thereof. For example, on page 49 the following appears: "If you set a more stringent alpha (e.g., a = .01) is set,..". The sentence was clearly edited, but the edited-out part was left in. This happens in multiple places. Also, on page 41, the (non)-word "irged" is used instead of "urged." All of this should have been caught and fixed prior to publication and prior to asking for $22.50 for the book. I can understand a few errors making it into the final printed edition, but this bordered on ridiculous. I would say that the editorial errors actually became a distraction and took away from the central theme of the book.
Rating: 
Summary: Clear, concise, useful
Review: This book is clearly and concisely written and provides both an introduction to power analysis and a description of a single set of procedures for power analysis when using any of the procedures covered by the general linear model. If you're familiar with power analysis it's fairly easy to skip the sections you don't need to read without impairing your ability to follow the development of the power analysis model.Murphy and Myors also take some positions which are debatable (especially by those of us who often don't have the luxury of restricting our sample sizes) but always well argued.
Rating: 
Summary: Almost good.
Review: This text was interesting and informative, but belabored the value of minimum effect hypothesis testing and pretty much ignored confidence intervals as an alternative. Worse, this book contains some mistakes: the noncentral F distribution formula (A3 in Appendix A) is written with parameters that are not explained, and also the authors state that the classical hypothesis testing is false "by definition". This is simply not true; it may be false more often than not, but it is not false by definition. But the worst shortcomings of this book are that it propagates the use of statistical tables instead of clearly explaining the underlying formulae. With ubiquitous computers, it is ridiculous to think that people still need to consult tables, which are restrictive in the alpha values. After reading this text, it is clear that power depends on effect size, alpha, the standard deviations of the treated and untreated populations, and the sample size, but nowhere do the authors clearly show what this functional relationship is. I guess they think that gamma functions and the like are just too difficult mathematics and force people to blindly work with tables in a haze of confusion, wondering the functional relationship of these variables. Finally, they do point out the desired relative seriousness of type I vs. type II errors (a major plus) but fail to emphasize this point as much as it deserves. For example, if there is no a priori reason to favor type I over type II errors or vice versa, then these should be set equal to each other and the sample size calculated from the formulae. Using power = 0.8 with alpha = 0.1 may be acceptable in their field of psychology but is incongruous with the point that they belabor - that type II errors are typically more serious. In conclusion, I would say to read this book from a library and hold off on buying until they (hopefully) correct these flaws in a second edition. Unfortunately, I have yet to see a better text that does clearly explain the functional relationship between the variables involved in power calculations.
Rating:
Summary: Almost good.
Review: This text was interesting and informative, but belabored the value of minimum effect hypothesis testing and pretty much ignored confidence intervals as an alternative. Worse, this book contains some mistakes: the noncentral F distribution formula (A3 in Appendix A) is written with parameters that are not explained, and also the authors state that the classical hypothesis testing is false "by definition". This is simply not true; it may be false more often than not, but it is not false by definition. But the worst shortcomings of this book are that it propagates the use of statistical tables instead of clearly explaining the underlying formulae. With ubiquitous computers, it is ridiculous to think that people still need to consult tables, which are restrictive in the alpha values. After reading this text, it is clear that power depends on effect size, alpha, the standard deviations of the treated and untreated populations, and the sample size, but nowhere do the authors clearly show what this functional relationship is. I guess they think that gamma functions and the like are just too difficult mathematics and force people to blindly work with tables in a haze of confusion, wondering the functional relationship of these variables. Finally, they do point out the desired relative seriousness of type I vs. type II errors (a major plus) but fail to emphasize this point as much as it deserves. For example, if there is no a priori reason to favor type I over type II errors or vice versa, then these should be set equal to each other and the sample size calculated from the formulae. Using power = 0.8 with alpha = 0.1 may be acceptable in their field of psychology but is incongruous with the point that they belabor - that type II errors are typically more serious. In conclusion, I would say to read this book from a library and hold off on buying until they (hopefully) correct these flaws in a second edition. Unfortunately, I have yet to see a better text that does clearly explain the functional relationship between the variables involved in power calculations.
<< 1 >>
|
