Causal thinking and causal language in epidemiology: a cause by any other name is still a cause: response to Lipton and Ødegaard

I have great sympathy with the thoughts of Lipton and Ødegaard [1] – the assessment and communication of "causal" associations is a source of continual frustration for epidemiologists. The authors' lucid account of the use of causal language in epidemiology can essentially (if rather unflatteringly) be simplified to the following: it is impossible to prove that X causes Y; the statement "Smoking causes lung cancer" is thus no more informative than the statement "Smoking two packs a day for N years increases your risk of lung cancer ten-fold". In fact, it is less informative and even misleading. The authors argue that such causal statements are redundant, logically indefensible and should be avoided in favour of more detailed descriptions of the process by which such associations are established (the "story", as the authors put it). The latter are, in themselves, sufficient causal statements (the notion of "letting the data speak for themselves") and nothing is gained by making subjective attributions of causality.