10 of 10 people found the following review helpful:
4.0 out of 5 stars
An amazing book, but with two serious shortcomings, December 12, 2007
This review is from: Regression Analysis: A Constructive Critique (Advanced Quantitative Techniques in the Social Sciences) (v. 11) (Hardcover)
This book is a must-read for any statistician or researcher in the social science who desires to use regression in their work, or to interpret or critique others' work that makes use of regression. The focus of the book is on identifying the potential for abuse of regression and finding ways to use regression effectively without abusing it. This book is extremely disciplined in its philosophical approach and aims to cultivate that same critical outlook in others. This book has two notable weaknesses, however: the author offers neither creativity nor encouragement when it comes to exploring new techniques. Also, the author makes almost exclusive use of fabricated data, instead of real data, to illustrate his points.
The author claims that this book would be most useful for people with prior background in regression, but I find this book to be quite accessible. If anything, the author may slightly exaggerate the prerequisites (in the interest of caution and rigor), and in addition to this, he makes up for the subtle and difficult nature of the material through his excellent writing style. In the end, though, this book is more difficult on a philosophical than a mathematical level; this book requires deep thought and self-motivation to understand, but not much in the way of mathematical background.
The core message of this book is that regression has the greatest potential as a descriptive and data-analytic technique, and that its use in inference, particularly in causal inference, is almost always misguided. It is hard to disagree with the author's viewpoint without being in rather severe denial of reality; the author's criticisms are very sound, regardless of whether they are what most statisticians or social scientists "want to hear".
One thing I find annoying about this book, however, is that the author seems to unconditionally support the status-quo when it comes to which mathematical techniques are used. He mentions things like Bayesian methods and quantile regression in passing, comments that they are "not widely used" or "have not caught on", and ignores them from then on. I think there is something deeply hypocritical about calling others to think critically about regression, and yet not being at all critical when it comes to reflecting on what techniques to use, or thinking about developing new techniques. Even if he does not want to explore these techniques himself, he ought to talk about them more positively and with more encouragement, instead of dismissing them the way he does.
I have a second criticism of this book: when illustrating his points, the author uses fabricated or "toy" datasets almost exclusively. I find this to be objectionable, and I think it weakens the author's argument. Although I agree with the conclusions of this book (outside of the dismissal of "non-mainstream" techniques), I think that it is crucial that the author's message gets out there to both students and experienced statisticians alike, and I think his reliance on fabricated datasets may be offputting to many people. It may take time an effort to find real data that gives elegant examples of the phenomena he is discussing, but the examples are out there and I think it is pretty weak that he did not put in the work to do this.
Bottom line? This book is especially useful for undoing the damage done by many of the common textbooks and courses on regression. I wish this book were more widely used as either a textbook or a supplement for graduate courses in statistics. The ideas in this book really need to "get out there" so to speak. People who want a book with a similarly thoughtful philosophy but more exposition of mathematics might want to check out the book by Freedman "Statistical Models". People who want a more introductory book on regression that is still more honest than most about the limitations of regression may want to check out "Introduction to Regression Models" by Abraham and Ledolter. But I wish Berk would revise this book in a new edition, making use of real datasets, and I wish he would give a bit more encouragement to methods and techniques that are outside the mainstream. Also, this book needs to be supplemented by a creative mind and by other books that explore some of the other techniques out there--I especially think quantile regression is worth exploring.
Help other customers find the most helpful reviews
Was this review helpful to you? Yes
No
1 of 4 people found the following review helpful:
5.0 out of 5 stars
A different take..., March 2, 2006
This review is from: Regression Analysis: A Constructive Critique (Advanced Quantitative Techniques in the Social Sciences) (v. 11) (Hardcover)
Emphasizes concepts and conditions underlying good regression. While typical books on regression hurl equations at you, this book details the meaningfulness of various model assumptions.
Help other customers find the most helpful reviews
Was this review helpful to you? Yes
No
