Monday, February 13, 2017

Thoughts on Angrist and Pishke's "Undergraduate Econometrics Instruction"

Joshua Angrist and Jörn-Steffen Pischke, coauthors of "Mastering 'Metrics," have just released a new NBER working paper called "Undergraduate Econometrics Instruction: Through Our Classes, Darkly." They argue that pedagogy has not kept pace with trends in economic research in the past few decades:
In the 1960s and 1970s, an empirical economist’s typical mission was to “explain” economic variables like wages or GDP growth. Applied econometrics has since evolved to prioritize the estimation of specific causal effects and empirical policy analysis over general models of outcome determination. Yet econometric instruction remains mostly abstract, focusing on the search for “true models” and technical concerns associated with classical regression assumptions. Questions of research design and causality still take a back seat in the classroom, in spite of having risen to the top of the modern empirical agenda. This essay traces the divergent development of econometric teaching and empirical practice, arguing for a pedagogical paradigm shift.
The "pedagogical paradigm shift" they call for would include three main components:
One is a focus on causal questions and empirical examples, rather than models and math. Another is a revision of the anachronistic classical regression framework, away from explaining economic processes and towards controlled statistical comparisons. The third is an emphasis on modern quasiexperimental tools.  
Since I am relatively new to both teaching and economics-- I didn't major in economics as an undergraduate, and did my Ph.D. from 2010 to 2015-- the first economics course that I designed and taught at Haverford quite naturally adhered to many of Angrist and Pischke's recommendations. The course, which I taught in Fall 2015 and Fall 2016, is called Advanced Macroeconomics, but is essentially an applied econometrics course on empirical macroeconomic policy analysis. The students in the course are typically juniors and seniors who have already taken econometrics.

On the first day of class, we read excerpts from the 1968 paper "Monetary and Fiscal Actions: A Test of Their Relative Importance in Economic Stabilization" by Andersen and Jordan. The authors want to test whether "the response of economic activity to fiscal actions relative to that of monetary actions is (1) greater, (2) more predictable, and (3) faster." They use very simple regression analysis, essentially regressing changes in GNP on changes in measures of monetary and fiscal actions. This type of regression is now called a "St. Louis Equation," since Andersen and Jordan were at the St. Louis Fed. I ask my students to interpret the regression results and evaluate the validity of the authors' conclusions about policy effectiveness. With some prodding, the students come up with some ideas about potential omitted variable bias and data concerns. But they don't think about reverse causality or the idea of a "controlled statistical comparison." I introduce the reverse causality issue, and much of the rest of the course focuses on quasiexperimental tools.

The course has no textbook, but we use "Natural Experiments in Macroeconomics" by Nicola Fuchs-Schundeln and Tarek Hassan as the main reference. The course has four units: consumption, monetary policy, fiscal policy, and growth and distribution. In each unit, I assign natural experiment or quasiexperimental papers as well as other papers that attempt to achieve identification via other means, to varying degrees of success. The reading list was influenced by Christina Romer and David Romer's graduate course on Macroeconomic History at Berkeley, which introduced me to the notion of identification and ignited my interest in macroeconomics.

Angrist and Pischke also argue that "Regression should be taught the way it’s now most often used: as a tool to control for confounding factors" in contrast to "the traditional regression framework in which all regressors are treated equally." In other words, the coefficient of interest is on one of the regressors, while the other regressors serve as "control variables needed to insure that the regression-estimated effect of the variable of interest has a causal interpretation."

This advice on teaching regression resonates with my experience co-teaching the economics senior thesis seminar at Haverford for the past two years. Over the summer, my research assistant Alex Rodrigue read through several years' worth of senior theses in the archives and documented the research question in each thesis. We noticed that many students use research questions of the form "What are the factors that affect Y?" and run a regression of Y on all the variables they can think of, treating all regressors equally and not attempting to investigate any particular causal relationship from one variable X to Y. The more successful theses posit a causal relationship from X to Y driven by specific economic mechanisms, then use regression analysis and other methods to estimate and interpret the effect. The latter type of thesis has more pedagogical benefits, whether or not the student can ultimately achieve convincing identification, because it leads the student to think more seriously about economic mechanisms.