Joshua D. Angrist

Using Data to Understand Economic Relationships

As a labor economist and econometrician, Angrist is interested in using data to understand economic relationships. In his world, randomized controlled trials are ideal but they aren’t always possible.

“It's not something I get to do very often, a real randomized trial,” he says. “We'd like to study things with controlled experiments. Mostly, we use a naturally occurring variation to kind of mimic that or at least that’s the hope.”

One such area where randomized controlled trials are often impossible to conduct is in education, one of the key pillars of Angrist’s research.

Education and the Returns to Schooling

“Returns to schooling is a very important question in my field of labor economics,” he says.

“We're very interested in who earns what and why. There's a lot of randomness in earnings outcomes, but there are some systematic correlates, and one of the most important systematic correlates of earnings is education. People who have more education earn more. So if you spend a year in school, how much more do you earn later?”

According to his research, each year of schooling is associated with roughly 10 to 12 percent in higher wages, and college graduates earn 60 percent higher wages than high school graduates in the United States. This is important not just for the individual, but also for policy makers, says Angrist, because investing into education is much more than the tuition people pay. Conducting an experiment here is tricky however, as schooling is not randomly assigned. Angrist and his late colleague Alan Krueger began investigating what it would take to truly measure the returns to schooling and developed an instrumental variable (IV) strategy. It’s since become very well known, using the timing of the year in which people were born and compulsory attendance laws in the United States.

“In the United States, you're basically allowed to drop out of high school on your 16th birthday,” he says. “So if you are born later in the year, you're younger when you start school and therefore you'll be kept in school longer before you're allowed to drop out.”

Angrist and Krueger looked not at the current generation, but looked back to their parents’ generation, a time when dropping out of high school was much more common than today.

“They dropped out, for example, because they needed to start work,” he continues. “You could go into some sort of vocational training program or you could go into the armed forces. And especially at that time, that was a very important channel. And if they finished high school, it's because they were forced to. There we find that being forced to finish school really has a large payoff.”

Returns to schooling is a very important question in my field of labor economics. We're interested in who earns what and why.

Using Instrumental Variables to Examine Military Service

While he was a PhD student and the years that followed immediately afterwards, Angrist studied the effects of serving in the military on the earnings of soldiers. He was trying to determine specifically whether joining the military could be seen as an overall good career and financial move. By looking at data from the US’ involvement in the Vietnam War and World War II, he found what looked like insights, but ones that needed further examination.

“If you look at people who served in World War II who were born in the 1920s and compare them to people who didn't serve from those same birth years, you'll see that people who served in the World War II armed forces live longer,” says Angrist. “So it looks like it was good for them to serve, actually improved their health. But if you think a little bit about that, that's probably preliminary to conclude that. The reason the people who served in the military live longer is people who had health problems were not eligible for service and that predicts their future mortality, not whether they served.”

Solving the problem of identifying correlation versus causation in this scenario is complicated in the same way that many areas within economics are not conducive to randomized controlled trials. In his research however, Angrist found an instance where the United States did something in what he calls “the direction of a randomized trial.” In the 1970s, during the Vietnam era, men were called up for conscription to be processed and see whether they were eligible on the basis of random lottery numbers that were assigned to birthdays.

When the US got involved in Iraq and Afghanistan there was some discussion about going back to the draft. And I was quick to point out that that was harmful to the draftees, so we should think hard about whether we really want to do that.

“So that’s not quite a randomized trial because it's not a simple in or out treated control scenario, but there's an element of partial random assignment,” he says. “Those lottery numbers have some predictive power for who serves. So that became my thesis project. And essentially I've been working on empirical strategies to analyze that sort of scenario where there's an element of random assignment. How do you use that to extract and learn about causal effects? The basic tool that allows us to do that is an econometric tool, instrumental variables.”

IV Methods and Refining Interpretation of Estimates

Angrist and his co-Laureate Guido Imbens were awarded the prize not for the creation of IV, but for finding new ways to use IV methods, interpret the results, and refine the interpretation of estimates. Once Angrist applied IV to his work, the findings became much clearer.

“Serving in the military holds you back in your civilian career,” he says. “The impact of serving in that period turns out to be a fairly substantial negative effect, something like 15 percent lower earnings in your thirties. I returned to that in early 2000 when the US got involved in Iraq and Afghanistan and there was some discussion about going back to the draft. And I was quick to point out to the media and to other interested observers that the earlier experience with the draft was that that was harmful to the draftees, so we should think hard about whether we really want to do that.”

A Simpler, Funnier, Approach to Economics

Angrist and his co-author Jörn-Steffen Pischke drew inspiration from unlikely sources: a comedy science fiction series and martial arts. The result, Mastering ‘Metrics: The Path from Cause to Effect, was the funniest econometrics books at the time – at least according to Angrist himself, though he does say their next book Mostly Harmless Econometrics later dethroned Mastering as the funniest. But inspiration and laugh factors aside, Angrist said that he and his co-author had very serious intentions with the books.

The most valuable tools, in our view, were simple tools… There is a model, but it's very simple and easy to describe.

The Impact of Joshua Angrist's Work

“We had this view that there's a lot of potentially dangerous ideas in econometrics, and by that we meant things that are relatively complicated and produce estimates that aren't very transparent,” he explains. “It's hard to describe where they came from and the most valuable tools, in our view, were simple tools, regression, instrumental variables, differences in differences, regression discontinuity designs that are all kind of meant to capture causal effects in one way or another in a very transparent way where you can say, ‘Well, what I'm really doing here is comparing the draft of eligible and ineligible men, and then from that I’m inferring a particular causal effect.’ And there isn’t really an elaborate model between you and the data. There is a model, but it’s very simple and easy to describe.”

Angrist knows that stories help convey bigger messages and they really wanted to bridge a disconnect they saw between econometrics instruction and econometrics practice. It was a bonus that he enjoys writing as a practice and ultimately he found the whole process fun.

“One reason you write is to be understood,” he says. “But you also want to be remembered; you want to make an impact.”