Thursday, February 10, 2011

Correlation and causation

Here’s a short lesson in Epidemiology 101. The bottom line first: correlation is not inevitably the same as causation. Two factors, say A and B, may be correlated – i.e. when A occurs B occurs too, or A increases (or decreases) when B increases (or decreases). But this doesn’t necessarily mean A causes B – or B causes A. Of course cause may exist – but not necessarily so. An example? Ten days ago the Journal of Neurology Neurosurgery & Psychiatry published a paper [1] reporting a prospective Scottish study of mortality after head injury. It followed 767 head-injured people, plus groups of matched controls, for 13 years. In this period 40% of the head-injured group died. And their death rate was much higher than the control groups’ death rates. The abstract’s Background section says: ‘Increased risk of death years after injury might be explained by factors associated with, but not a consequence of, the head injury. This unique prospective study investigates mortality over 13 years after injury’. Unique and prospective this study may be, but it sheds no light on whether head injury and subsequent high mortality are causally related (i.e. head injury causes high subsequent mortality) or whether they’re just correlated (i.e. merely present together). In the latter case there may be a third factor associated with both head injury and increased mortality, e.g. smoking, participation in extreme sports, depression. So be on the lookout for claims that A causes B when in reality A’s merely correlated with B. Ho hum.

No comments: