Correlation Still Does Not Prove Causality

Oct 2014

There seems to be no limit to the emailed articles from friends or friends-of-friends that contain seemingly alarming economic predictions based on the status of some alleged indicator.  Never mind that more than half of the alarmist predictions use faulty or irrelevant data.  Even when the data are real and seemingly relevant, we again urge caution.  As [we've said], “Given long enough time and deep enough data, you can find a seemingly impressive linkage between almost any two factors.  But correlation isn’t causation.”  To illustrate the limits of such “data mining,” Harvard economist Greg Mankiw noted the following close (but nonsen­sical) correlations:  (1) the divorce rate in Mississippi and the number of murders by bodily force; (2) honey-producing bee colonies and the number of juvenile arrests for possession of marijuana; (3) the per-capita consumption of mozzarella cheese and the number of civil engineering doctorates awarded; and (4) U.S. crude oil imports from Norway and the number of drivers killed in collisions with railway trains.  Each of these pairs … has a statistically high correlation, yet none make any sense.  They are the one-in-a-hundred exceptions that can be found if you look hard enough.  Our advice is to beware of anything that comes in a mass-circulated email and also to beware of correlations that may seem less nonsensical but are still essentially random noise.