Were you aware that there is a 98% correlation between the number of lawyers in Hawaii and the per capita consumption of mozzarella cheese over a 10-year period? Hmm…

Or have you considered the 94% reverse correlation between the cost of a 16-oz bag of potato chips and the starvation death rate in the U.S.? As the cost of a bag of chips has gone up, the incidence of death by starvation has gone down. That being the case, congress should regulate the price of chips ever higher to save people from dying of hunger, right?

Well, not so fast…

Being able to establish statistical correlation does not prove that one action has any bearing on the other even if the statistics support it. Just for the fun of it, Tyler Vigen has set up a computer program that crunches the numbers
on thousands of random statistics and matches the statistic sets to others that have a high correlation.

Mr. Vigen is a “Geospatial intelligence analyst” with the U.S. Army National Guard and an enthusiastic lecturer on the phenomenon of “spurious correlations.” He has posted dozens of these spurious correlations on his website to
make a point about science and statistics.

When you see a close connection between, say, bedsheet deaths and ski revenues, Vigen says, you consider for a moment what the connection might be, but then you quickly dismiss any significant relationship because it is not
supportable by common sense. “We saw data,” Vigen says, “we formed an hypothesis about that data in the form of a causal mechanism, and then we rejected the hypothesis based on our personal experience with the world.”

Vigen says that the critical weighting of the data’s significance is “big deal,” because “that’s the part that computers can’t do.” Good science relies on an objective collection of data, but it also necessarily involves interperetation.
“Statistical data can show correlations, and then it’s up to us as rational thinkers to establish whether there’s actually a connection between the variables, or if it’s merely a coincidence,” he says.

No Comments

Post A Comment