Regression Analysis: A Constructive Critique

Advanced Quantitative Techniques in the Social Sciences #11

by Richard A. Berk

Regression A Constructive Critique identifies a wide variety of problems with regression analysis as it is commonly used and then provides a number of ways in which practice could be improved. Regression is most useful for data reduction, leading to relatively simple but rich and precise descriptions of patterns in a data set. The emphasis on description provides readers with an insightful rethinking from the ground up of what regression analysis can do, so that readers can better match regression analysis with useful empirical questions and improved policy-related research. "An interesting and lively text, rich in practical wisdom, written for people who do empirical work in the social sciences and their graduate students." --David A. Freedman, Professor of Statistics, University of California, Berkeley

280 pages, Hardcover

First published July 17, 2003

Reviews

[g.lkoa · Rating 3 out of 5]

3.8/5 | To-be-read-again, but not again and again. Arguably well carved more than a little, by a knowledgeable top notch statistician – but only a bit interesting. Berk breezes through a large number of relevant scope/validity domain’s issue: what regression (multiple linear) are good at and what they are totally not good at. I’d have been glad the author could elaborate a bit better and and in a more systematic manner on this issue, but nontheless it conjures up a pretty good case for how cautious we should be regarding the regression machinery and their assumption/hypothesis (or support), and the theoretical impedments to use them as for causal inference. Nice discussion of how these complications get usually ignored and how instrumental variable are damn good at concealing the pachyderm in the room. Not entirely sure the section dealig with Pearl's work on causal inference and DO-calculus do bring justice to this wide topic, certainly no more than a descriptive summary in any - of the few - comprehensive textbook.

[John Myles · Rating 5 out of 5]

I found this to be a wonderful book about the dangers of contemporary statistics approaches. I think the book has an unusually strong description of descriptive statistics and makes a clear delineation between the styles of statistical argument one sees in publication. These delineations are much too rare in our current scientific culture and it's a joy to read such careful exposition of what's wrong with contemporary science.

That said, I found the book a little hard to follow at times because topics that seemed to require more mathematical detail were discussed in English and, vice versa, topics that required English were discussed using mathematical formalisms.