This article was written by David A. Jaeger, expanding on his The Conversation article, “Nobel economics prize winners showed economists how to turn the real world into their laboratory”.
The awarding of the Sveriges Riksbank Prize in Economic Sciences to David Card, Josh Angrist, and Guido Imbens for “Answering Causal Questions Using Observational Data” marks the culmination of a revolution in the way empirical economists approach the world that began more than 30 years ago. Along the way, the entire field of economics has shifted towards being more data-oriented and less driven by pure economic theory.
Around the late 1980s, labour economists started thinking deeply about how to estimate the effects of phenomena like immigration or the minimum wage in ways that could plausibly be interpreted as representing a causal relationship, even in the absence of randomised experiments. Along with this focus on causality, came a new focus on data and measurement of individuals’ behaviour.
Both Card and Angrist had Orley Ashenfelter as their doctoral supervisor, and Ashenfelter deserves substantial credit along with Card for pushing labour economics and empirical economics, more generally, towards the search for “natural experiments” that mimic the randomisation of treatment that occurs, for example, when pharmaceutical companies try to establish the efficacy of a new drug. While experiments have become increasingly important and popular in economics (the Nobel Prize was awarded in 2019 to three economists, Abhijit Banerjee, Esther Duflo, and Michael Kremer, for their work in using randomised experiments to establish policies in developing countries to reduce poverty), they were relatively uncommon in the 1980s, and most economists who worked on applied topics used “observational” data from surveys (like the Census) or administrative sources (like Social Security).
Ashenfelter and Card were among the first in economics to recognise the importance of the “counterfactual” in establishing that a relationship between two economic variables, like education and wages, was causal rather than just a correlation. Simply put, just because we observe that, on average, those with higher levels of education also earn higher wages does not mean that those higher wages are caused by more education. Other factors, such as a privileged family background or higher innate ability, could be associated with both higher levels of education and higher wages. If those factors aren’t observed by the researcher, it is difficult to establish causality without randomising the “treatment,” in this case education. Natural experiments, in which some mechanism like a policy change or twin birth, assigns the treatment as if it were random, allow us to establish what the counterfactual outcome would have been, on average, for those who were actually treated.
David Card and the real world as laboratory
Two of David Card’s most influential papers employed natural experiments to great effect. In “The Impact of the Mariel Boatlift on the Miami Labor Market” (Industrial and Labor Relations Review 43(2):245-257, 1990), he examined how more than 120,000 migrants who left the port of Mariel in Cuba between April and 1980 affected labour market outcomes in Miami, where most of the migrants landed. One might be tempted to do a “before” and “after” comparison of wages and unemployment in Miami, but this would ignore the fact that the US economy was very different in 1979 (expansion) than is was in 1981 (recession) for reasons that had nothing to do with the influx of migrants. To address this, Card used the average change in wages and unemployment in Atlanta, Houston, Los Angeles, and Tampa-St. Petersburg to measure what would have happened in Miami in the absence of the “Marielito” refugees. By subtracting this change from the change in labour market outcomes in Miami, he was able to (arguably) calculate the effect of the influx of immigrants on wages and unemployment in Miami. Card found, remarkably, that there was virtually no effect on the wages of lower-skilled non-Cubans in Miami nor an increase in unemployment among blacks or non-Cuban workers. This result was controversial when it was published 31 years ago and is still controversial today. A small cottage industry re-examining Card’s paper has emerged, with some researchers confirming his results and others disagreeing vehemently that immigration has little impact. What has remained most influential, however, is Card’s attempt to measure the counterfactual outcome and estimate a causal effect with using the method of “differences [over time] in differences [between the treated and control places]”.
The second of Card’s papers that has been enormously influential is work with the late Alan Krueger, Card’s and Ashenfelter’s colleague at Princeton, who tragically died at the age of 58 in 2019. In “Minimum Wages and Employment: A Case Study of the Fast-Food Industry in New Jersey and Pennsylvania” (American Economic Review84(4):772-793, 1993), Card and Kreuger examined the impact of the minimum wage on employment. Standard economic theory dictates that imposing a minimum wage in the labour market should have no effect on employment (if it is below the market equilibrium wage) or decrease employment – but is highly unlikely to increase employment. Living in New Jersey, Card and Krueger were aware that New Jersey would raise its minimum wage from $4.25 to $5.05 an hour (above the Federal minimum wage) on 1 April 1992. Utilising the vast resources of Princeton’s Industrial Relations Section, where both were faculty members, they gathered data from fast food restaurants in New Jersey and, as a counterfactual, Pennsylvania, before and after the change in New Jersey’s minimum wage. Contrary to the conventional theoretical predictions, Card and Krueger found that employment actually went up in New Jersey’s fast-food restaurants relative to those in Pennsylvania. As with Card’s work on immigration, this result was immediately (and still is) controversial, spawning much further research, including a book of their own on the minimum wage, Myth and Measurement (Princeton University Press, 1995). Krueger strongly advocated for increasing the Federal minimum wage in the U.S. during his time as the Chief Economist at the Department of Labour under Bill Clinton, resulting in a 21 percent hike in August of 1996.
Josh Angrist and the causal impact of schooling
Like Card, Josh Angrist is a product of the fertile environment at the Industrial Relations Section at Princeton of the 1980s. His two textbooks, Mastering Metrics: The Path from Cause to Effect and Mostly Harmless Econometrics: An Empiricist’s Companion (Princeton University Press, 2015 and 2009, respectively, both written with Jörn-Steffen Pischke, who also completed his PhD at Princeton) have revolutionised the teaching of econometrics at the undergraduate and graduate level and codified the techniques of causal inference. The Nobel award cites Angrist’s work in econometrics, but his work on the economics of education is equally important, particularly that on the impact of charter schools on educational outcomes. What distinguishes Angrist’s work is the absolute laser-like focus on exploiting quasi-random variation to identify causal effects. His contributions have given us new methods and better ways of interpreting the results of old methods.
Apart from his textbooks with Steve Pischke and his work with Guidio Imbens discussed below, one of Angrist’s most influential contributions crystalises the approach for which this Nobel Prize was given. In “Does Compulsory School Attendance Affect Schooling and Earnings?” Quarterly Journal of Economics 106(4):979-1014, 1991), written jointly with Alan Krueger, who surely would have shared this award, were he alive, Angrist tries to tease out the causal effect of schooling on earnings. This was a notoriously difficult empirical problem because unobserved factors like family background and innate ability may be correlated both with schooling and with earnings and muddy empirical estimates of the wage increase due to education alone. Angrist and Krueger noted that school age starting laws in the US, which usually state that kids need to start first grade in the calendar year in which they turn 6, interact with compulsory schooling laws, which usually state that one can drop out of school after turning 16, induce a relationship between quarter of birth and education, with individuals born in the last quarter of the year (who start school at age five and can drop out at age 16) getting more education than those born in the first quarter of the year (who start school at age 6 and can drop out at age 16). In the extreme, someone born on 31 December would get an entire additional year of schooling than someone born on 1 January. Because it seemed implausible at the time that when one is born in the year should be related to family background or innate ability, Angrist and Krueger asserted that we could use the correlation between quarter of birth (as an “instrument”) and schooling to purge the relationship between schooling and earnings of any of the unobserved factors. What they found was as surprising as Card’s or Card’s and Krueger’s results – they estimated that effect of schooling on earnings was actually greater than previous estimates using conventional methods! While there is still some controversy over whether these results are fully reliable, what is indisputable is that Angrist and Krueger’s paper set the standard for how empirical economists employ “instruments” as a means of estimating causal effects with observational data.
Guido Imbens and causal inference methods
Unlike Card and Angrist, Guido Imbens did his PhD at Brown University (1991) and was advised by Tony Lancaster, an econometrican who moved to Brown in 1986 from Hull University in England and brought Imbens with him as a student. Imbens’s contributions are more purely methodological in nature than either Card’s or Angrist’s but are equally concerned with estimating causal effects in non-experimental data. Imbens’s body of work provides tools for those who work empirically to estimate causal effects – or to know when there are limits in how we can interpret our results. It has been enormously influential framing how we should think about policy evaluation.
Imbens’ most influential paper, “Identification of Causal Effects using Instrumental Variables” (Journal of the American Statistical Association 91(434):444-455, 1996), was co-authored with Angrist and Donald Rubin, a statistician at Harvard who also could easily have shared the Prize. This paper lays out a general framework that helps us to understand how to evaluate policies when compliance with being assigned to treatment is imperfect and when the effects of policies differ across individuals. In another influential paper, “Identification and Estimation of Local Average Treatment Effects” (Econometrica 62(2):467-475, 1994), Angrist and Imbens define exactly for whom the estimates from a natural experiment give the causal effect, what is known as the “local average treatment effect”. To return to the quarter-of-birth example above, it turns out that the estimates that Angrist and Krueger produced are most relevant for those individuals who would have dropped out before age 16 but who were forced to stay in school longer because of compulsory schooling laws. If the return to education for them somehow differs from the return to education for those who would have stayed in school past age 16 regardless of compulsory schooling laws, Angrist’s and Krueger’s results are relevant only for those who were forced to stay in school.
It is worth noting that all three – Card, Angrist, and Imbens – are not only among the world’s best economists but are also phenomenal educators. Card has been tremendously prolific at Princeton and at Berkeley in training graduate students, many of whom are themselves at the top of the profession. Angrist regularly gives classes around the world on causal inference and his instructional videos on Marginal Revolution (marginalrevolution.com) are at once crystal clear and amusing. Imbens brings his insights on causal inference to his MBA teaching at Stanford and his 2004 “What’s New in Econometrics” lectures at the National Bureau of Economic Research Summer Institute (along with Jefferey Wooldridge) are legendary. What is clear from their examples is that research and teaching are highly complementary.
For Card, Angrist, and Imbens, the “credibility revolution” in economics is about providing defensible estimates of causal effects from non-experimental data – even if those estimates run counter to economic theory. They fundamentally believe that data, from the “real world”, will reveal the truth, and have developed methods for showing us that truth. Their Nobel Prize is richly deserved.
David A. Jaeger is Professor of Economics at the University of St Andrews, a Research Fellow of the Centre for Economic Policy Research, and the Editor of the Scottish Journal of Political Economy. He is a member of the Applied Microeconomics Group (http://applied-microecon.wp.st-andrews.ac.uk) at St Andrews.