Richard A. Berk

Presumes at least minimal familiarity with the math and practicalities of doing a multiple linear regression. The point of this book is to explain very clearly what regression can and cannot do, and especially to drive home the meaning and force of all the assumptions which are required to make the machinery of statistical inference for linear regression work. These assumptions are, or ought to be treated as, scientific hypotheses, which need to be not just taken for granted, or even listed mechanically and then ignored, but supported. (Many of them are hard to even assert with a straight face about important kinds of real data, never mind back up.) As he rightly says, the requirements for using regression for causal inference are even stronger, and the common practice of ignoring these issues, or hoping that they'll go away if you just use instrumental variables, has nothing to recommend it. (However, the discussion of Judea Pearl's work on causal inference in section 10.5 seems to me to be somewhat superficial, and even to misunderstand Pearl's book more than a little.) Of course, as descriptive summary of a data set, regression has much to recommend it; but not necessarily more than newer methods of data-mining, which he considers briefly in the conclusion.

I won't say this belongs on every practitioner's shelf, because it's not the kind of book you'll come back to again and again. I will say that anyone who is becoming a practitioner of regression needs to learn these lessons thoroughly, for their own sake and for the sake of anyone who might have to rely on their findings.

Errata: p. 66, the statistic oscillates between being S and being W, and the function Q is nowhere defined; p. 129, the first two terms in the numerator of eq. 8.9 should be grouped together in a parenthesis.… (mer)