Mastering 'Metrics: The Path from Cause to Effect

Citation

Angrist, J. D., & Pischke, J.-S. (2014). Mastering 'Metrics: The Path from Cause to Effect. Princeton University Press.

Chapter Summary

The book has an introduction and six chapters, about 245 pages of text followed by a list of abbreviations, empirical notes and an index. Its organising idea is the "Furious Five", the five methods the authors consider most valuable for causal questions: random assignment, regression, instrumental variables, regression discontinuity designs and differences-in-differences. Chapters 1–5 each take one method, and Chapter 6 applies several of them to a single question. The kung fu framing runs throughout (Chapter 1, for example, opens with an epigraph from the Kung Fu television series and an "Our Path" overview). Chapters 1–5 each end their main text with a "Masters of 'Metrics" profile of a historical figure, and all chapters except Chapter 4 have an appendix that restates the ideas in a more conventional, formula-based way. Each method is developed through a few published studies, whose data and Stata files are available on the book's website.

Introduction

Sets out the book's view of econometrics as the study of cause and effect and introduces the five tools that structure the rest of the book.

Chapter 1: Randomized Trials

Section 1.1 ("In Sickness and in Health (Insurance)") starts from survey data on whether people with health insurance are healthier than those without, shows why such comparisons mix the effect of insurance with selection bias, and presents random assignment as the solution, using the RAND Health Insurance Experiment. Section 1.2 ("The Oregon Trail") turns to the Oregon Health Plan lottery. The profile covers figures "From Daniel to R. A. Fisher", and the appendix ("Mastering Inference") reviews means, standard errors and statistical inference in the experimental framework.

Chapter 2: Regression

The chapter introduces regression through the question of whether attending an expensive private college pays off, in three sections: "A Tale of Two Colleges", "Make Me a Match, Run Me a Regression" and "Ceteris Paribus?". As the titles suggest, regression is presented as a way of making matched, other-things-equal comparisons. The profile is of Galton and Yule; the appendix covers regression theory.

Chapter 3: Instrumental Variables

Three studies illustrate instrumental variables: "The Charter Conundrum" on charter schools, "Abuse Busters" on the Minneapolis Domestic Violence Experiment (where randomly assigned police responses were not always delivered as assigned), and "The Population Bomb". The profile is "The Remarkable Wrights"; the appendix sets out IV theory.

Chapter 4: Regression Discontinuity Designs

"Birthdays and Funerals" uses the minimum legal drinking age to show how outcomes jump at an age cut-off, and "The Elite Illusion" uses admission cut-offs to ask whether selective public high schools improve their students' results. The profile is of Donald Campbell.

Chapter 5: Differences-in-Differences

"A Mississippi Experiment" analyses the effect of monetary policy on bank failures in Mississippi during the 1930 banking crisis, asking whether central banks should step in when depositors run. "Drink, Drank, . . ." returns to the minimum legal drinking age with a regression version of differences-in-differences across states. The profile is of John Snow, and the appendix discusses standard errors for regression DD.

Chapter 6: The Wages of Schooling

The final chapter applies the earlier tools to the economic return to education: "Schooling, Experience, and Earnings", estimates from identical twins ("Twins Double the Fun"), instrumental variables based on child labour laws and quarter of birth ("Econometricians Are Known by Their . . . Instruments"), and "sheepskin effects" of a high school diploma in Texas ("Rustling Sheepskin in the Lone Star State"). The appendix treats bias from measurement error.

Key Concepts

Introduction and Chapter 1

• Ceteris paribus comparison: comparing outcomes with other things equal, the goal that all five methods pursue.
• Selection bias: the part of a naive treated-versus-untreated difference that reflects pre-existing differences between the groups rather than the treatment.
• Random assignment: the benchmark design that removes selection bias and against which other methods are judged.

Chapters 2–3

• Regression with controls: comparing treated and untreated individuals who match on observed characteristics.
• Instrumental variable: a variable that shifts the treatment but affects outcomes only through it, used when compliance with assignment is incomplete.

Chapters 4–6

• Regression discontinuity: estimating effects from jumps in outcomes at a cut-off in a running variable such as age or test score.
• Differences-in-differences: comparing changes over time in a treated group with changes in an untreated group, estimated in regression form across many states in Chapter 5.
• Measurement error bias: how error in a regressor, such as reported schooling, biases estimates.

Critical Analysis

Strengths

• Causal questions first. The book is built around questions about health insurance, colleges, policing, banking and schooling rather than around estimators. Walter Krämer, reviewing it in Statistical Papers (2015), described it as a microeconometrics textbook "quite out of the ordinary" that starts strictly from applications and puts the lack of controlled experiments at the front.
• Very little formal machinery in the main text. Formulas appear mostly in the appendices; Krämer notes that the definition of expectation given early in the book will make mathematically inclined readers wince but helps economists without a statistics background. Simeon Paton's student review similarly highlights the everyday language and the appendices that give the textbook-style treatment.
• Real studies and replicable data. Every example is a published study, and the authors' website provides the data and Stata .do files for the tables and figures of Chapters 1 and 3–6, which makes the book easy to pair with hands-on work.
• Memorable and engaging. Endorsements collected on the authors' website, from Andrew Gelman, Gary King and others, single out the readability and humour, and the "Masters of 'Metrics" profiles give each method a historical background.

Limitations

• Not a full econometrics course. Five designs are covered; time-series methods, limited dependent variable models and the matrix treatment of regression are not part of the contents. Paton notes it may not serve an economics undergraduate course in its entirety.
• A narrow, mainly American set of examples. Krämer calls the examples, from Harvard versus a state university to drinking ages across US states, rather parochial for readers outside that setting.
• Steeper than the style suggests. Paton observes that a basic grounding in econometrics helps because the topics quickly increase in complexity.
• No exercises. The contents list no problem sets; practice has to come from the website data or a companion textbook.

Where It Fits in the Course

Within this course list, Mastering 'Metrics is the short, design-based introduction to causal inference. It pairs naturally with a conventional textbook that develops regression systematically: Studenmund's Using Econometrics, Stock and Watson's Introduction to Econometrics or Wooldridge's Introductory Econometrics supply the estimation theory, diagnostics and further topics that this book leaves out, while Angrist and Pischke explain why particular research designs are credible. A statistics background of the kind covered by Newbold and colleagues' Statistics for Business and Economics makes the inference appendix easier.

The authors' own Mostly Harmless Econometrics covers the same tools (regression, instrumental variables, differences-in-differences, regression discontinuity, plus quantile regression) at a graduate level. Chris Blattman, quoted on the book's website, describes Mastering 'Metrics as the more casual, intuitive introduction that the harder book lacked; reading it first makes Mostly Harmless much more approachable.

How to Use This Book

The book suits economics and social science undergraduates in a first econometrics or research-methods course, policy students, and graduate students from other fields who need to read empirical papers. Some exposure to basic statistics helps, particularly for the appendices.

• Core: the Introduction and Chapters 1–5 in order; each method builds on the selection-bias framework of Chapter 1.
• Appendices: read them on a second pass, or alongside the relevant chapters of a regression textbook if the course is technical.
• Chapter 6: a capstone that shows how several methods are applied to one question; useful for revision.
• Practice: download the data and .do files from the authors' "Killer Apps" page and reproduce the tables, for example the Oregon lottery results or the drinking-age regression discontinuity figures.

For a reading group or a policy course, the main sections can be read without the appendices; for an econometrics course, assign the appendices and add problem sets from a conventional textbook.