When correlation and causation coincide

Abstract

Introductory statistics teaches us that correlation does not imply causation. If two variables, A and B, are correlated with high statistical significance, it does not necessarily imply that A causes B (nor that B causes A). Hence, the high correlation between birth rate and number of storks in variously industrialized European countries in the past century is not a proof that storks bring babies (nor that babies attract storks). Even in realistic scenarios, rushing to the conclusion of causation is not an uncommon fallacy: the lower rate of road accidents involving white as opposed to black cars has been explained by the better visibility of the former. But often, the correlation is explained by a third (“unaccounted for”) factor C that causes both A and B: industrialization causes both reduction of wild-life habitat and rise of nuclear families – hence reducing the number of storks and births. Younger drivers account for a disproportionate fraction of road accidents, and they prefer black cars. But in medicine things get more complicated. I regularly encountered patients with back-pain who would stoically refuse to take pain killer, rightly asking: isn’t pain just a symptom and shouldn’t one treat the root cause of the pain? Of course, symptoms are only correlated to, and not the cause of, the pathology. As with so many ills of mankind – e.g. poverty, violence, climate change – we are taught to fix the root cause, not treat the symptoms. This common wisdom breaks down when it comes to the dynamics of diseases or any dysfunction in complex systems. Ironically, perhaps more often than not, treating the symptom actually works. Consider cancer: whenever a protein X is discovered to be expressed