Apples, oranges and spurious correlations

Here are some fun factoids: data shows there is a link (85% correlation) between the age of every Miss America that appeared between 1999 and 2009 and murders in the United States during the period; murder by steam, hot vapors, and hot objects- to be exact. Odd?

Data also shows that mozzarella cheese consumption per capita highly correlates (96%) with civil engineering doctorates awarded. Perhaps, parents will now add an extra dollop of cheese in their children’s sandwiches in the hopes that their cheese consuming offspring will grow up to get a doctorate in engineering. Whether these facts are taken seriously or not, surely there is a line to be drawn here in using these facts and interpreting them to design, suggest or support a policy- surely.

While Asad Umar highlighting the need to relax the lockdown has several merits, the data used to rationalize said lockdown was confounding. The comparison made was: in one month, deaths from coronavirus are projected to reach 720 (based on Asad’s projection from the last 6-day mortality rate for Pakistan due to COVID-19). Comparatively, in an average month, traffic accidents across Pakistan lead to over 4,800 deaths. Asad argues that while there is such a higher mortality rate for traffic accidents (comparably), never has a case been made to stop traffic, restrict mobility of people and/or forbid cars from continuing to occupy and dominate roads.

The argument here is about trade-offs. That the loss to people if traffic was banned and mobility was restricted would be much greater. It is probably a good-natured comparison, except that it makes no sense. Traffic accidents are not contagious like COVID-19, for one, which remains the most compelling argument made in favor of a lockdown while facing the ravages of this pandemic. Traffic accidents are also preventable by taking precautionary measures that do not necessarily have to involve closing down roads or restricting mobility.

Yes, traffic accidents will decrease if there is no traffic but there are other strategies that can be adopted to lower traffic-related mortality rate. Traffic accidents are also far more predictable since there is so much historical data available to make reliable projections. On the other hand, it is substantially harder to predict the rise in numbers of cases and mortality if the virus is not stopped. But as little as the world knows about the coronavirus, there is a clear understanding that physical distancing and contact will flatten the curve i.e. number of positive cases will decline if folks were staying at home.

Of course, physical distancing does not naturally translate to a lockdown. There are several strategies to choose from given country-specific constraints and capabilities. In Pakistan, it seemed lockdown was the right choice but as circumstances have evolved and new evidence has come in, perhaps the lockdown is not the best option. There is certainly a huge burden to bear here for the lockdown to continue in its current form—loss of employment and wages, more people sliding in poverty, loss of revenues for big and small businesses and the economy shrinking by miles. Furthermore, while flattening the curve will lower the spread, the disease would stay for longer.

But the point here is much simpler: this is the time for the government to think carefully before communicating critical information and making sweeping comparisons to give weight to an official stance. The use of data here to compare mortality rates between traffic accidents and COVID-19 is at best superfluous, and at worst can have dangerous implications. Lockdown or no lockdown, it undermines government’s own efforts to keep unaffected people safe and practice physical distancing wherever they can. It also opens doors for the acceptability to other—what we call in economics—spurious correlations.

One example is the rumor about a positive link between 5G usage and positive coronavirus cases (later proven to be incorrect). Another oft-quoted example across South Asia is where researchers found a lower incidence of coronavirus cases among countries that had a national BCG-vaccine program. Despite the correlation, there may not be a causal link and there may be a variety of other unrelated reasons (lesser testing, country specific social distancing measures, or public health interventions etc.) for the observed pattern to exist. This research needs to be explored by researchers, but not relied upon.

Needless to say, data needs to be carefully interpreted, and the interpretation needs to be carefully verified before any conclusions or comparisons can be drawn lest we unwittingly give people a false sense of security or hope. That itself can be dire.