American Scientist

To the Editors:

In “What Everyone Should Know about Statistical Correlation” (January–February, Ethics) and in a subsequent Letter to the Editors, Vladica Veličkovic and Ted Grinthal repeat what I think is a virtually ubiquitous misstatement: They said it is an error to think that “correlation implies causality.” Setting aside the issue that a sample correlation may not mean it exists in the population, a correlation does imply causality—it’s just that the pattern of causality explaining the correlation is ambiguous. The correlations between a country’s chocolate consumption and its Nobel Prize winners, mozzarella cheese consumption and engineering doctorates, and countless others more or less frivolous than these all derive from some real causality. These measurable relationships in and of themselves tell us little about why they exist. Determining causality requires logic, research design, or subtle statistical modeling (usually with additional data).

I agree with both authors that false conclusions about causality based on the existence of simple statistical relationships is a widespread and pernicious error of critical thought. But the proper phrase to teach every schoolchild should be something like “correlation is causality, but ambiguously so.”

Dan Montello
University of California, Santa Barbara
Santa Barbara, CA