Hernan, VanderWeele, and others argue that causation (or a causal question) is well-defined when interventions are well-specified. I take this to be a sort of methodological axiom of the approach.
But what is a well-specified intervention?
Consider an example from Hernan & Taubman’s influential 2008 paper on obesity. In that paper, BMI is shown up as failing to correspond to a well-specified intervention; better-specifed interventions include one hour of strenuous physical exercise per day (among others).
But what kind of exercise? One hour of running? Powerlifting? Yoga? Boxing?
It might matter – it might turn out that, say, boxing and running for an hour a day reduce BMI by similar amounts but that one of them is associated with longer life. Or it might turn out not to matter. Either way, it would be a matter of empirical inquiry.
This has two consequences for the mantra that well-defined causal questions require well-specified interventions.
First, as I’ve pointed out before on this blog, it means that experimental studies don’t necessarily guarantee well-specified interventions. Just because you can do it doesn’t mean you know what you are doing. The differences you might think don’t matter might matter: different strains of broccoli might have totally different effects on mortality, etc.
Second, more fundamentally, it means that the whole approach is circular. You need a well-specified intervention for a good empirical inquiry into causes and you need good empirical inquiry into causes to know whether your intervention is well-specified.
To me this seems to be a potentially fatal consequence for the claim that well-defined causal questions require well-specified interventions. For if that were true, we would be trapped in a circle, and could never have any well-specified interventions, and thus no well-defined causal questions either. Therefore either we really are trapped in that circle; or we can have well-defined causal questions, in which case, it is false that these always require well-specified interventions.
This is a line of argument I’m developing at present, inspired in part by Vandebroucke and Pearce’s critique of the “methodological revolution” at the recent WCE 2014 in Anchorage. I would welcome comments.
This is very interesting. Please share the manuscript when it is finished. Thanks!
LikeLike
I don’t think there is a circle here. I think a well-defined causal question is any question that has an intervention specific enough to operate. We can ask any question we want and ask if a causal relationship exists. I think intervention is question specific that as long as the intervention could answer the question, it’s well-defined. We of course could improve our question based on previous research, and it’s more like a spiral, that we kept improving and ask better-defined causal question.
LikeLike