Mostly Harmless Econometrics

Mostly Harmless Econometrics: An Empiricist's Companion is an econometrics book written by two labour economists Angrist and Pischke. Jan Kmenta, also a labour economist, notes that the book is not a textbook as such but rather a book describing a series of econometric issues encountered by the authors in their empirical research and implicitly as an advocacy of the approach they have taken.

The book has eight substantial chapters organised in 3 sections: preliminaries, the core and extensions: The first section on Preliminaries outlines the basic approach taken highlighting the importance of identifying what the causal relationships of interest are. They stress the importance of research design and random assignment. The second section, The Core stresses the importance of trying to make regression make sense. The first of the major approaches they advocate is the use of instrumental variables . They then recognising that good instruments can be difficult to come by they suggest approaches to deal with unobserved confounders. Here under the headings: fixed effects, differences-in-differences, and panel data, they suggest the use data with a time or cohort dimension to control for unobserved but fixed omitted variables. The final section has chapters on Regression Discontinuity Designs and quantile Regressions and ways of coping with Nonstandard Standard Error Issues.

The book has been widely reviewed. One such review is by Kmenta. He has a few quibbles with some issues but is overall positive arguing that the book will help extend the vocabulary of econometricians by introducing terminology not encountered in standard econometric textbooks and that the book will prove particularly interesting for micro-econometricians in general and labour economists in particular. Another review, by Gelman, a statistician rather than a labour economist, is perhaps a little more critical arguing that despite the breadth promised by its title the coverage of econometrics is rather limited with the data being addressed being largely cross-sectional with little acknowledgement of time series data and the fact that the only problems addressed are causal whereas econometricians are typically also interested forecasting, description and testing of theories. The lack of acknowledgement of nonparametric methods, Bayesian inference, or models other than the standard linear regression are also raised as issues. The lack of model building is at the heart of Gelman's worries.

The same two authors have written a more basic book covering similar ground and taking a similar approach: Mastering'metrics: The path from cause to effect

The book

• Angrist, Joshua D., and Jörn-Steffen Pischke. (2008) Mostly harmless econometrics: An empiricist's companion. Princeton university press (ebook published 2008 - copyrighted 2009)

The authors' home page for the book

• https://www.mostlyharmlesseconometrics.com/