Last Monday it was announced that this year’s Nobel Prize in Economics was awarded to David Card, Joshua Angrist and Guido Imbens. The Committee highlighted in all three cases their empirical contributions and their advances in the study of causality and, specifically, in the use of natural experiments in the social sciences. This is an award that goes far beyond economic science. The concern for causality is today common to all social sciences and all of us who take it seriously from other fields are also very satisfied with this award.
But what is a natural experiment and why are they so useful in the social sciences? Lets start by the beginning. When we study any social phenomenon, one of the objectives we normally have is to know what its causes are. At the theoretical level, the task consists of proposing an argument and, with it, a causal explanation that links one factor with another. The fundamental problem of causality in the social sciences appears when we want to empirically test these arguments. This problem resides in the fact that the economic, political or social phenomena that we want to understand are complex and, therefore, the result of many factors, not just one. They are multi-causal phenomena.
Let’s imagine that we want to know how different levels of proportionality in an electoral system can increase or decrease participation in elections. This will be a very relevant question if, for example, the advisability or not of reforming the electoral system is being discussed. The problem is that if we want to investigate what the effect of something is, we need to control the effect that other questions have on our object of study. Returning to our example, if we wanted to know, and quantify, the effect of the electoral system on participation, we should isolate this relationship from the effects that could explain why in some countries there is more voting, such as the diversity of preferences and groups. social that already exist in society (and which, moreover, are often the ones that initially explain why a more or less proportional electoral system is adopted). The key to understanding the problem of causality is therefore to understand, following our example, that it is very difficult, close to the impossible – to use an expression by Adam Przeworski – to separate the effect of the electoral system from the effect of the social environment. Both can be related to participation and both can go hand in hand.
How can we unlink the effect of these two factors to know the true causal effect of the electoral system? This is where the experimental revolution in the social sciences comes in and from where the Nobel prizes to Card, Angrist and Imbens are derived, but also to the laureates in 2019, Esther Duflo, Abhijit Banerjee and Michael Kremer. In order to empirically identify the causal relationship of one factor over another, we need to compare two groups with identical characteristics except for the factor whose effect we want to investigate. Using academic jargon, we need a treatment group and a control group. In our example, the first would be a society with certain characteristics (including a certain configuration of society) to which an electoral reform is applied that makes the system more proportional. The second group would be a company with the same characteristics but to which the reform is not applied. Given that the only difference between the two groups is the treatment (the proportional reform of the electoral system), if we observe differences in the levels of participation, these can only be due to the effect of the treatment. Thus, we would have isolated and identified the causal effect of the electoral system on participation.
Is it possible to carry out this type of experimental exercises in social sciences? Unlike what happens in a laboratory in the natural sciences, the ability to manipulate, to intervene in the reality of economists, political scientists or sociologists to test certain hypotheses is quite limited. Yes, there are many relevant questions to which answers have been found applying experimental field work where researchers explicitly design treatment and control groups (see the classic study by Gerber and Green, 2000). But many other important questions cannot be approached from this perspective since it is not possible to manipulate reality. In social sciences we work with the world we observe -that is, with the real fact that this or which country has a certain electoral system- and when studying the causes of a phenomenon it is not easy to find its counterfactual in that same world to be able to compare the presence of a treatment versus the absence of a treatment. Nor can we call Pedro Sánchez by phone to ask him to apply an electoral system in certain Autonomous Communities and not in others so that we can study their differences.
What we can do is take advantage of all those circumstances where, naturally, without the intervention of the researcher, history has produced quasi-experimental scenarios. That is, where a historical accident or an unexpected event has changed certain factors in some areas while these have remained unchanged in others and both areas are sufficiently comparable. In this way, a treatment group and a control group would have been generated in a natural way, since the treatment assignment is like was random. It is for all this that natural experiments are so useful and why their use is not exclusive to economics. Both in political science and in sociology we have tried to apply them in order to establish causal relationships on phenomena that interest us.
A well-known example is the job de Kern and Hainmueller (2009) in which they ask whether exposure to foreign media can contribute to destabilizing autocratic regimes. In addition to addressing an interesting question from a theoretical point of view, what stands out from this research is a very intelligent empirical identification strategy. Thanks to the fact that the West German television signal reached many, but not all parts of East Germany due to its topographic variation, they take advantage of this quasi-random exposure to check if access to a greater plurality of information is it relates to different forms of support for the communist regime. Contrary to expectations according to theories about the role of the media, the authors find that those populations exposed to Western television (simply because geographically they had more access to it) had increased their support for the autocratic regime. The authors offer an interpretation to these results, that is, that Western television could function as a kind of entertainment and escape from the daily pressures under the communist regime, making life somewhat more bearable and, therefore, decompressing tensions with the own regimen. Regardless of the fact that this interpretation is debatable, what is interesting about the study is that it informs the theoretical debate about the role of the media and the stability of autocratic regimes, continuing and improving the academic conversation in this regard.
Another example is the study Posted by Adam Glynn and Maya Sen, in which they try to isolate the effect of empathy in the decisions that judges make. Studying something as ethereal (and that correlates with so many personality traits) as empathy is a methodological challenge. To solve it, Glynn and Sen look for a natural experiment in which an event totally external to the judges supposes a “shock” of empathy in some judicial decisions. His solution is to study the impact that having a daughter, instead of a son, has on the votes of the US Supreme Court justices on gender issues. We can expect that having children is a more or less voluntary decision that can be related to many issues. On the other hand, that this offspring later is in the form of a daughter or son is a totally random element that, according to the authors, will condition the way in which the judges will look at the world and empathize with some judicial issues. Their results show that those judges who have daughters adopt more progressive positions aligned with the feminist agenda on issues such as abortion, employment discrimination based on gender, or equal opportunities in education, among others.
Natural experiments also allow us to evaluate some relevant events in our recent political history. Montalvo, for example, studies here what impact the 11M attacks and the management that the Aznar government made of those crises had on the 2004 general elections. The polls had been giving a victory for Mariano Rajoy as a more likely result, but, a priori, it would seem almost impossible to determine how much the final effect of the attack was on the electoral result. In fact, the post-election polls gave us uncertain answers because, after the elections, citizens rationalize their vote and give reasons to explain why they finally mobilized or voted for a party that do not necessarily correspond to what would have happened if the attack it would not have happened. The solution that Montalvo adopts is to use the postal vote of Spaniards residing abroad as a natural experiment. Since voting abroad has time restrictions, these voters had to vote before the attacks. The deadline for CERA to vote in person at consulates or by certified mail was March 7. This allows us to assume that voters abroad did not know about the 11M attack when voting while voters who resided in Spain did. The deadline for voting from abroad is a natural experiment that allows us to compare both groups. Montalvo’s conclusion, observing that the PP did relatively better in the CERA vote compared to the residents’ vote, is that Rajoy could have obtained between 5 and 6.7 points more in the elections of March 14, 2004 if 11M and its subsequent management would never have happened.
Another example that falls close to us is the job by Carlos Sanz on the effects of the type of electoral list (open or closed) on participation. On this occasion, the author exploits a little-known peculiarity of the Spanish electoral system for the holding of municipal elections. The electoral law establishes a discontinuity in the way of electing representatives at the local level based on the number of inhabitants of the municipalities: towns with less than 250 inhabitants use an open list system (voters can selectively choose candidates of the same party as well as others); while localities with more than 250 inhabitants use the closed list system. Studying the electoral behavior of municipalities around this threshold arbitrary (established by state law) it is possible to separate the effect of the type of lists from that of other factors that could correlate with the endogenous establishment of the same, and therefore with the levels of electoral participation. In other words, given that the comparison is not between cities like Madrid and Barcelona with towns with less than 250 inhabitants, but between small municipalities but with very similar characteristics, except for the electoral rule, it is easy to think that this accident Historically, a treatment group and a control group have been formed almost randomly. Sanz finds that open lists encourage participation in the order of between 1% and 2%.
These are just a few examples, but we could have brought many more. The causal revolution in the social sciences, beyond economics, is a fact. Their contribution is not only to make us a little wiser about some social, political and economic phenomena, and to establish more precise explanations (which allow us, for example, to better assess the impact than a public policy), but also allow us to discuss, put questioned and question some of the theories that we believed to be more established.