Andrew Gelman reviews Mostly Harmless Econometrics on his stats blog. The book, by labor economics greats Angrist and Pischke, reads like an updated, extended version of Angrist and Kruger’s famous chapter in the Handbook of Labor Economics: Empirical Strategies in Labor Economics.
It’s a retrospective on the search for causal identification, and a more readable guide for practice than the usual fare (it’s only partly written in Greek). Lovingly, geekily, it draws its title from the novel by Douglas Adams.
Andrew laments the absence of data modeling: how to go from raw data to regression functions. Angus Deaton would probably criticize the absence of economic theory. Both are correct, but it’s an excellent book I nevertheless foisted on my graduate statistics class and my research assistants right away.
And you can even buy a t-shirt.