Monday, January 19, 2015

Mastering Metrics

I went through a copy of Mastering Metrics by Joshua Angrist and Steve Pischke. Their previous book Mostly Harmless Econometrics is widely used as a companion to advanced undergraduate and introductory graduate courses in Microeconometrics. Both books are very useful. Mastering Metrics has six chapters on: 1. Randomised Trials; 2. Regression; 3. Instrumental Variables; 4. Regression Discontinuity Designs; 5. Differences in Differences and 6. The Wages of Schooling.

The RCT chapter is a clean and concise overview of the main economic parameters that can be estimated with an RCT and provides a good and well worked-through example. The chapters on regression and IV don't cover too much more ground than in their previous work but have some very interesting extra examples that students will find useful. Similarly, RDD and Difference-in-Difference designs are covered in Mostly Harmless Econometrics but are extended with more examples in this book. The final chapter goes through in detail the case of estimating returns to education and goes through the problem of omitted variable bias in depth. Each of the chapters will certainly be very useful companion chapters to lectures on the topics.

The style of both MHE and MM (whether you appreciate the jokes or not) is rightly popular among students, combining informal text with worked examples and appendices providing the mathematical foundations. Many people's first introduction to topics like IV or RDD is in the context of highly formal and stylised mathematical models and the approach here has probably led to many students figuring out for the first time that the key concepts are interesting and applicable. Armed with these insights it is easier to step back into the fight with the notation.

It is clear they have the potential to continue this into a series of short and accessible books.  I think a clear demand from my students would be to extend the RCT chapter into a book that addressed not just the basic parameters but the wider set of parameters that can obtained from RCTs as well as how to think about common problems with RCT designs in a more formal way.

No comments: