A look at the 2021 Nobel Prize in Economics

Three researchers — David Card, Guido Imbens and Joshua Angrist — won this year's Nobel Prize in Economics for their use of natural experiments to find causation.

This article walks through the logic of natural experiments, referring to a paper written by the Royal Swedish Academy of Sciences on why the Nobel prize was awarded.

There are some bizarre correlations out there. In the 2000s, drownings were highly correlated with the number of movies Nicholas Cage starred in. Between 1940 and 2000, every time the Washington Football Team won their final home game in an election year, the incumbent president won the election.

Clearly, these have absolutely nothing to do with each other. But, if you have enough concurrent events happening in the world, you're going to be able to find some that are highly correlated with each other, despite no relationship actually existing.

Have a look at these bizarre correlations:

The old saying is: 'correlation doesn't imply causation'. If you can actually explain to us how higher divorce rates in Maine cause higher consumption of margarine per capita, we'll buy you a Tesla.

When it comes to education levels and income, the same issue applies. The data shows there is a correlation, but does more schooling actually cause higher incomes? Maybe there are other things are at play? Think: family wealth, intelligence and conscientiousness.

What does causation really mean?

Before getting into the Nobel Prize winning research, we need to first understand what causation really means.

Consider two variables, X and Y. We say X causes Y if a change in X leads to a change in Y when holding all possible other factors the same. This is different from correlation, which does not hold all other factors the same.

In the education example, we need to ensure variables like intelligence and family wealth are held the same.

So, if we can isolate the change in X and see its impact on Y, we can determine if X does indeed cause Y. In our example, if we isolate years of education, we can see if it really does cause incomes to be higher.

The ideal scenario

Ideally, to figure out if staying in school increases your income, we would run a randomised controlled trial (RCT). We would construct a control group of 1000 randomly chosen school children who must drop out at the age of 16 — and a treatment group of 1000 randomly chosen school children who can't drop out until they complete high school.

The reason this experiment allows us to see if there is causation is that it randomises children across the two groups, ensuring both groups are almost identical. The randomising isolates the effect of completing school. So, if the treatment group has higher future incomes than the control group, we know completing more years of school causes a higher future income, on average.

But good luck getting any parents to sign their child up for this experiment.

For most relationships outside of natural sciences, running a controlled experiment is impossible; we have to somehow figure out how to use existing data from the real world. We also have to figure out how to isolate the impact of extra education, whilst controlling for other factors.

Independently, Card, Imbens and Angrist were able to figure out a way to faithfully construct the treatment and control groups from existing data.

This is why their work is so genius; it allowed researchers across all disciplines to answer questions where an experiment could not be conducted.

How did they show causation?

Card, Imbens and Angrist use something called an instrument variable. These are variables that affect X (i.e. how long you stay in school) but have no impact on Y (i.e. what your future income will be). The key is this: the instrument variable must be random.

You can then use this random instrument variable to reconstruct the control and treatment groups used in experimental research. Confused? Let's go through the schooling and income example to show you how it works.

Angrist noticed that in the US a child could leave school once they turned 16 or 17 (depending on the state). So, children born earlier in the school year could leave school earlier, and thus complete less total schooling than those born later in the year. And that is exactly what they found in the data.

Hence they could use a child's birthday as this instrument variable. It,

Is random because the birthday of any given person is random
Impacts variable X (how long you stay in school)
Does not impact variable Y (your future expected income)

Angrist then constructed his two groups. The control group are the children born in the first quarter of the year and the treatment group are the children born in the last quarter of the year. These two groups isolate the effect of years of schooling on future incomes because the allocation of children is randomised; we know your birthday has no impact on your future income (beyond impacting the average years of schooling).

The graph above demonstrates how children born in the fourth quarter of the year, on average, had higher average years of schooling and higher future weekly earnings. This isn't just correlation anymore, this is causation.

In fact, they demonstrated that each additional year of schooling directly caused about a 9% increase in future income.

We know your concern — this is pretty old data so the result is somewhat meaningless for us now. You are right, but this year's Nobel Prize was given for the discovery of the method not for the results themselves.

Conclusion

Despite the fact much of this work was done in the 1990's, its impact is still monumental. The set of evidence available to academics and business people has increased by an incredible amount. So, despite correlation not implying causation, we now have a way to navigate the grey area to make real conclusions about the world around us.

The recent emergence of 'alternative data' (which we will go into further in a future article) uses these principles and techniques to provide a better understanding of the world around us.

Though, we will end this article off on a more somber note. Card, Angrist and Imbens all worked very closely on this research with renowned economist and public servant Alan Kruger. His contributions to the field were massive and without his work Card, Angrist and Imbens could not have gotten as far. He sadly passed away in 2019, and if he was still alive he would have surely co-won the Nobel prize this year alongside his distinguished colleagues.

Two other case studies (Additional reading)

Here, we go through two other natural experiments that contributed to the Nobel Prize. The first of those studies shows increasing the minimum wage doesn't cause more unemployment; the second that migration doesn't necessarily cause more unemployment for locals.

Study 1: Minimum wage and unemployment

If you want to start an argument - get a bunch of economists and politicians in a room and get them to talk about whether the government should raise the minimum wage. Traditional economics says increasing the minimum wage decreases demand for labour, leading to higher unemployment. Some data shows the minimum wage and unemployment are positively correlated and other data says they are uncorrelated. Some data even says they are negatively correlated. Card wanted to figure out whether or not a rise in the minimum wage actually causes more unemployment.

In order to do so, he needed an instrument variable. It needed to be random and it needed to have an impact on the minimum wage but not the employment level. In 1992, New Jersey raised its minimum wage from $4.25 to $5.05. Meanwhile, Pennsylvania held it’s minimum wage constant at $4.25. So, his instrument variable was the state in question.

Card then looked at restaurant employment across both states immediately after the change (he uses restaurant workers because they tend to be paid at or near minimum wage). Because the communities in either state were so identical, the state border randomised the communities into two groups to isolate the impact of the minimum wage rise. Thus, Pennsylvania was the perfect control group and New Jersey was the perfect treatment group.

Having a look at the graphs above, you can see employment in New Jersey actually increased in the wake of the minimum wage rise while employment decreased in Pennsylvania.

Now, this does not mean raising the minimum wage won't lower employment in other scenarios. Almost all economists would agree: if you tripled the minimum wage tomorrow there would be some bad employment effects. But this paper does provide evidence that raising the minimum wage doesn't necessarily cause worse employment outcomes.

Study 2: Immigrants and jobs

If you want to start an even bigger argument - get a bunch of economists and politicians in a room and get them to talk about immigration. Phrases along the lines of "immigrants are stealing my jobs" or "immigrants are buying my house" could easily get thrown around. With a topic like this, there is always a visceral 'us' versus 'them' mentality.

Card decided to check if the job prospects of the local population were actually impacted by large amounts of immigration.

For a bit of context, in 1980 Fidel Castro unexpectedly allowed all Cubans who wished to migrate to the US to do so via boat. In what is now known as the Mariel boatlift, 125,000 Cuban's sailed to the US, where 50% permanently settled in Miami. Over 3 months, Miami's labour force jumped a staggering 7%. This provided Card with the perfect ability to prove if immigration caused worse job prospects for the local population. Similar to the minimum wage example, his 'instrument' variable became cities. If Card picked cities nearly identical in labour market conditions to Miami prior to the migrant intake, he could randomise away all other factors to isolate the impact of immigration.

Card found no evidence the wage and employment of non-Cubans were affected by the Mariel boatlift in Miami. The control group (other US cities similar to Miami) saw the same effects as the treatment group (Miami). In fact, they found native-born citizens in Miami even saw some better outcomes due to the technology investment that occurred in the wake of the sudden migration.

Further research by Card shows continual migration actually does have an effect - but only on very specific industries - and generally only on previous migrants.