When it comes to proving the reliability of a theory or a hypothesis, fields like physics, chemistry or medicine have a substantial leg up over the field of social science due to their ability to control for different variables through randomised controlled trials (RCTs) and establish causality.
For example, if researchers wanted to test the effectiveness of a vaccine, they could easily give the vaccine to one randomly selected group and the placebo to the other to see if there was a significant improvement.
The same cannot be done in the field of social science because of ethical and practical limitations. For instance, to study the impact of poverty on citizens' health, we cannot randomly subject a large group of people to economic hardship to see how their health changes.
Similarly, other variables could impact both poverty and health that we cannot know or control. So, how can social scientists establish causality between real wages and living standards, lockdowns and infection rates or the impact of different government policies?
The answer is natural experiments - a policy or an event that randomly segments individuals into different treatment and control groups that helps researchers establish causality with minimum assumptions.
In fact, this year's Nobel Prize in Economics (or the Sveriges Riksbank Prize in Economic Sciences in Memory of Alfred Nobel 2021) was jointly awarded to scholars David Card (UC Berkeley), Joshua Angrist (MIT) and Guido Imbens (Stanford University) for their pioneering work in using natural experiments as a reliable research method and "revolutionising empirical research in the economic sciences".
Famously, in the early 1990s, David Card with the late Alan Krueger studied the impact of higher minimum wage on the employment of labour by analysing a natural experiment (conventionally, the assumption was that higher wages created unemployment as it raised the cost of production for firms).
To mimic the conditions of an RCT, they collected data regarding the employment at fast-food restaurants in the adjacent areas on either side of the state border between Pennsylvania and New Jersey, assuming that the businesses in either jurisdiction would be quite similar.
By comparing the findings from both areas before and after an 8% increase in minimum wage was issued only in New Jersey, the researchers found out that unemployment did not rise as expected despite the introduction of minimum wages as the theory predicted.
Although we now know that the negative impact of higher minimum wages was quite low as firms just pass on the costs to customers, Card and Krueger's work spurred countless studies and pioneered the use of natural experiments in economic sciences which was later adopted by other fields of social science.
By studying natural experiments, Card was later able to publish some remarkable findings like the impact of immigration on employment (the natives actually became better off) and the impact of investment in schools on the future earnings of pupils (yes, better teachers, books, and facilities did increase the earning potential of the students).
However, it was still difficult to ensure causality in a natural experiment because, unlike in an RCT, the researcher cannot control who gets the treatment and who does not, making the results difficult to interpret.
To illustrate this problem, let us explore a scenario. Imagine two similar companies A and B in the same industry where A gave its employees bicycles as a bonus and the other did not. Clearly, this can be a useful natural experiment to find causality between variables like time spent bicycling and health conditions.
However, the problem arises with the level of participation of the participants themselves. For example, despite being gifted a bicycle, some employees from A may never ride it to work. In addition, there are probably those in A who would have ridden a bike to work regardless of the company's gift. Finally, there could be employees in B who decide to bike without being gifted.
These possibilities made drawing precise conclusions about the causality of the experiments difficult until the work of Joshua Angrist and Guido Imbens in the mid-1990s.
Using probability and a limited set of assumptions, they developed the LATE (localised average treatment effect) framework. This framework could estimate the causal effect among the participants who changed their behaviour (i.e., decided to ride the bicycle only because it was gifted to them) as a result of the natural experiment.
This framework was more credible and transparent as it prompted researchers to state their assumptions and was widely adopted by other researchers in the field of economics and beyond.
It has been over 30 years since the findings of the three scholars mentioned above transformed empiric research, resulted in better policy advice and helped researchers answer questions that we could not investigate by controlling the variables in a lab.
Opportunities to study natural experiments have become relatively abundant in the context of the pandemic and the rise of despotic regimes and political instability.
In that regard, on a call with the Chief Scientific Officer of Nobel Media, Adam Smith (fitting, I know!), while being congratulated on his Nobel Prize, David Card said, "Crazy political regimes have a lot of disadvantages but one of their advantages is that they do create very good conditions for causal analysis".