My point is that the data is messy. Smokers aren't a randomized subset of the general population.