Mostly Harmless Econometrics: An Empiricist's Companion

Authors: Joshua D. Angrist and Jorn-Steffen Pischke

Edition: First

Publisher: Princeton University Press

Pages: 392

Price: £54.00 and £24.95

ISBN 9780691120348 and 0355

In case you wondered, Mostly Harmless is the title of the fifth part of Douglas Adams' Hitchhiker's Guide to the Galaxy. These authors are clearly big fans of Adams. Their book provides a quirky and thought-provoking read for any budding econometrician, but be warned that they are applied labour economists who practise micro-econometrics. Their raw material is exclusively drawn from surveys and censuses of individuals, households or firms, and sometimes from panels. How to measure the effect of schooling on earnings from survey data is the empirical question used as an anchor for a large part of the technical discussion. The book has nothing to say about time series econometrics, or financial econometrics or macro-econometrics.

The style is informal and the notation and mathematical formalities are kept to a minimum; yet some concepts are advanced and the discussion is often subtle. Two theoretical topics are treated in detail: least squares regression and instrumental variables. The authors offer a careful and often original explanation of the role of covariates and instruments in estimating causal relations. Refreshingly, the instrumental variables method is motivated in terms of dealing with unobservables in causal relations, rather than telling the traditional "simultaneous equations" story. There are supplementary chapters on panel data, quantile regression and standard error estimation, among other topics.

The cover advertises the authors' alarming claim that "fancier econometric techniques are typically unnecessary and even dangerous". This seems to boil down chiefly to an advocacy of ordinary least squares in preference to probit and Tobit estimation for limited dependent variables (LDVs). I see their point that the assumptions underlying the standard LDV models can be a little peculiar, but would also warn that students need a decent understanding of maximum likelihood methods to properly evaluate their arguments.

This book's chief value will be as a guide for people embarking for the first time on applied research. As such, it is insightful and refreshing.

Who is it for? Graduates and micro-econometric practitioners.

Presentation: Accessible paperback with funny illustrations.

Would you recommend it? Highly - but not to beginners.