>[Note: as pointed out by comments below, extrapolation to life-years saved is very speculative, since all the studies on this in humans are going to be confounded all to hell by healthy user bias and socioeconomic correlations and the like. That said, it feels like a fairly reasonable extrapolation given the comorbidity of obesity to various extremely problematic medical conditions. Be warned!]
should be sufficient to exempt me from charges of “pretending to know things.”
The confidence intervals thing is probably a good idea, but I have no idea where to start on that, really, since the confidence intervals would be mostly driven by “how confident am I feeling about using correlational studies on health outcomes to make causal claims about the effects of a treatment” more than any objective factor.
I’m not actually sure about whether a study looking at the effects of successful weight loss on mortality would be all that helpful for this conversation, since that would still end up being a totally correlational study with enormous error bars and confounders, and successful long-lasting weight loss isn’t very common (itself which will introduce yet more confounders). Also I don’t think such a study exists.
The confidence intervals thing is probably a good idea, but I have no idea where to start on that, really,
Basically you have no idea about which extrapolations are reasonable to make and do them anyway in spite of having no idea what’s reasonable to do.
If you don’t think you understand the subject well enough to give confidence intervals I don’t think you understand it well enough to give any extrapolations that have any usefulness.
The idea that an “objective” number that’s clearly wrong is better then a more subjective that factors in more knowledge is also flawed.
I feel my disclaimer in the post:
>[Note: as pointed out by comments below, extrapolation to life-years saved is very speculative, since all the studies on this in humans are going to be confounded all to hell by healthy user bias and socioeconomic correlations and the like. That said, it feels like a fairly reasonable extrapolation given the comorbidity of obesity to various extremely problematic medical conditions. Be warned!]
should be sufficient to exempt me from charges of “pretending to know things.”
The confidence intervals thing is probably a good idea, but I have no idea where to start on that, really, since the confidence intervals would be mostly driven by “how confident am I feeling about using correlational studies on health outcomes to make causal claims about the effects of a treatment” more than any objective factor.
I’m not actually sure about whether a study looking at the effects of successful weight loss on mortality would be all that helpful for this conversation, since that would still end up being a totally correlational study with enormous error bars and confounders, and successful long-lasting weight loss isn’t very common (itself which will introduce yet more confounders). Also I don’t think such a study exists.
Basically you have no idea about which extrapolations are reasonable to make and do them anyway in spite of having no idea what’s reasonable to do.
If you don’t think you understand the subject well enough to give confidence intervals I don’t think you understand it well enough to give any extrapolations that have any usefulness.
The idea that an “objective” number that’s clearly wrong is better then a more subjective that factors in more knowledge is also flawed.