October 22, 1999 (Ira Pilgrim)

Correlations

Statistics are like a bikini: what they reveal is suggestive but what they conceal is vital,

Anonymous(1986)

Hardly a week goes by when we don't hear or read about a correlation. A few examples are a correlation between eating nuts and longevity, living close to power lines and leukemia, taking tamoxifen and breast cancer, eating fish and longevity, and many others. Often the story is presented to imply that one thing causes the other; eating nuts or fish causes you to live longer, tamoxifen prevents breast cancer and that power lines emit something that causes leukemia.

A correlation doesn't mean that two things are causally related. Nor does a correlation mean that two things are related at all. Wearing tuxedos correlates with drinking champagne, but wearing a tuxedo does not cause a person to drink champagne.

The correlation reported by a group of scientists may not be a correlation at all. It may just be a coincidence; a fluke. When the study is repeated, no correlation will be found. This is very common in medical research and the finding of a correlation between two things may be contradicted by another study of the same kind which shows no correlation at all.

Sometimes further investigation shows that the correlation is not only real, but that one thing actually causes another. This is nowhere near as common as we are led to believe by the scientists and the press.

If we postulate that A causes B, then as more evidence accumulates, the postulate becomes either more or less likely. Sometimes it turns out that C causes both A and B, and sometimes that A and B are unrelated.

In 1919 lung cancer represented between one and two percent of all cancers in the U.S.. By 1965, it was 43%. Several surgeons published their observations that almost all of their patients with lung cancer were heavy cigarette smokers and postulated that cigarette smoking caused lung cancer. It was plausible since the tobacco smoke is forcibly inhaled into the lungs, in contrast to pipe or cigar smoking, where the inhalation was not deliberate. However, two skeptical statisticians(Hammond and Horn) didn't believe it and they set out to do a thorough investigation. Half way into their study, both quit smoking. They found that not only did the incidence of lung cancer correlate with cigarette smoking, but that the incidence was proportional to the amount smoked as well as the length of time that a man smoked. Later, tobacco tars were shown to cause skin cancer in mice. Laboratory scientists were unable to find a non-human animal that was so stupid that it would force smoke into its lungs. Despite what you might have heard suggested by one anti-smoking commercials on television, living in a smoke-filled environment is nowhere near as dangerous as actual cigarette smoking.

The case of leukemia and power lines is an interesting one. People who live next to power lines are generally poor. No one who had money would live next to power lines. There are all sorts of things that go with poverty, particularly exposure to all sorts of carcinogens. Besides, no one has been able to cause cancer in animals with electromagnetic fields. So the issue remains in limbo as a correlation that is not necessarily related to cause.

When eating nuts and fish correlated with longevity, I chose to make believe that it might be true and that people who ate nuts were healthier. Why? Because I like to eat nuts and fish.

Next column

Return to the Science Home Page

Return to Ira's Home Page