The other causality question: estimating attributable fractions for obesity as a cause of mortality

Recent scientific controversy over the accuracy of population attributable fraction (PAF) estimates for obesity as a cause of mortality has made the concept of PAF visible in both scientific and popular news. The PAF is widely thought to provide information about causation, or attribution, of disease and also to provide information on the consequences of interventions to eliminate the exposure of interest. I discuss the methodological and conceptual limitations of the PAF in providing these two kinds of information. Because of these limitations, the PAF does not provide scientists or policy makers with an accurate answer to the question, How much of the disease burden could be eliminated if the exposure was eliminated from the population? Further, these limitations cannot be overcome merely by better statistical modeling; they must be addressed through more rigorous discussion of specific interventions and the causal consequences of such interventions.