The 2019 FIFA Women’s World Cup just ended on July 7. The host nation, France, provided nine venues across the country to hold the matches. The FRED Blog team and the St. Louis Fed in general loves using real-world situations to showcase applications of economics, and sports is a fairly popular real-world situation. (Also check out these essays on stadium subsidies, the Olympics, and France’s World Cup win in 1998.) One might expect that a large sporting event such as the World Cup would give an economic boon to the host, but let’s test that expectation by looking at data in FRED. Specifically, let’s see if the literal build-up to this World Cup affected construction in France.
One could easily assume that building stadiums to host the World Cup would cause a spike in construction. And the first graph does show that construction in France has gradually risen since the prior Women’s World Cup in 2015, which was held in Canada. At first glance, then, one might conclude that construction increased because France needed to expand its sports and tourism infrastructure ahead of hosting the World Cup.
But, before making sweeping claims about the causal effect of the World Cup on construction, we should be thoughtful about the theory. First, did France actually construct new stadiums? The short answer is no. Eight of the nine stadiums that hosted matches this year were built before France even learned it would host the World Cup in 2019. The lone exception is Parc Olympique Lyonnais, which opened in 2016. Although this venue hosted the World Cup Final, in which the U.S. defeated the Netherlands 2-0, it was not built specifically for this event. The stadium is better known as the home of Olympique Lyonnais, one of France’s most successful football (soccer) clubs.
Another key sanity check is to see whether the same result holds in similar periods. If it’s true that construction increases leading up to a major sporting event, then we’d expect to see the same trend before the 2016 Men’s European Championship, the 1998 Men’s World Cup, and the 1992 Olympics, all of which were hosted in France. However, when we expand the graph to a larger time window (shown below, with these events marked in red), the hypothesis doesn’t seem to hold. Rather, we see a decrease in construction leading up to both the 2016 Euro and the 1998 World Cup and an upward trend prior to the 1992 Olympics that seems to continue over too long of a time to truly be the result of hosting. Looking again at the period from 2015 to 2019, it seems more likely that there are other factors at play—beyond this year’s World Cup—driving changes in construction.
So what’s the takeaway here? It’s not likely that hosting the World Cup caused an uptick in construction in France. But there’s an important lesson about confirmation bias to consider: It was easy for us to theorize that a few years of increasing construction would precede the World Cup. It was easy to find a short time window of data in which we could see this expected effect. So it was easy to assume we proved our hypothesis by theorizing about the effects and finding evidence to support the claim. But as we saw, the longer-term time series showed something different. And that should serve as a warning against simply finding a small sample of data to support a theory and asserting it as truth. Rather, in making claims like this one, we should consider as many confounding factors as possible before being convinced. In other words, we should be our own biggest skeptics.
How these graphs were created: Search for “France construction,” select the quarterly series for “Production of Total Construction in France,” and click “Add to Graph.” For the first chart, use either the date boxes above the graph or the timeline below it to set the start date to June 2015. For the second graph, either set the start date to January 1960 using the same method or click the “Max” button above the graph. To add the vertical lines marking the major events, go to the “Edit Graph” panel and use the “Add Line” tab to create user-defined lines. Try different date ranges to see how the choice of time sample affects the appearance of the graph.
Suggested by Darren Chang, Matthew Kaye, Andrew Spewak, and Christian Zimmermann.
