> > People implying that the study shows only correlation really don't seem to understand how we establish causality in science.
> Or perhaps you don’t fully understand the challenges of establishing causality? Just because an intervention causes an improvement in some bio markers that are associated with lower mortality (unfortunately) does not mean that the intervention will cause lower mortality.
> The classic example is vitamin D supplements: Higher vitamin D levels are associated with lower mortality in many medical conditions. Vitamin D supplements increase vitamin D levels. But vitamin D supplements seldom lower mortality.
That is quite a simplification. The research as far as I know is that too low Vitamin D levels are associated with higher mortality, but that supplementation above a certain level is meaningless (and most caucasians can get those levels through normal sun exposure). So yes vitamin D supplements in general don't improve mortality.
Note also the situation here is completely different. The link between mortality and morbidity markers and exercise have been established in other studies. The study here establishes that this directly translates to a correlation between fitness and mortality. So in a way it's the opposite of the vitamin D case.
> Why? Probably because vitamin D is produced in the skin when we are in the sun. The more healthy subpopulation of any study will typically spend more time outside, so they will have higher vitamin D levels. But it’s (relative) health that causes higher vitamin D, not the other way around.
You realise that you are proclaiming causality here?
> The only way to reliably establish causality is really an end-to-end randomized controlled trial (RCT). Stitching together two RCTs is not sufficient.
A RCT does not establish causality. That's essentially my point. A single study/experiment never proofs causality. You need a theory to explain the causality and multiple studies that falsify other possible causality mechanisms. That has been done extensively for excercise and morbity/mortality, the current study just establishes that this also correlates in the bigger picture. So yes the study itself does not "proof" causality, it's just a piece in the bigger puzzle of causality.
> (Not saying that exercise does not lower mortality BTW, just that it’s complicated, and a study such as this is probably picking up two signals: one causal and one purely correlational.)
> > Why? Probably because vitamin D is produced in the skin when we are in the sun. The more healthy subpopulation of any study will typically spend more time outside, so they will have higher vitamin D levels. But it’s (relative) health that causes higher vitamin D, not the other way around.
> You realise that you are proclaiming causality here?
Yes, with emphasis on the word probably. Just like I will happily proclaim that exercise probably lowers mortality.
> A RCT does not establish causality. That's essentially my point. A single study/experiment never proofs causality. You need a theory to explain the causality and multiple studies that falsify other possible causality mechanisms. That has been done extensively for excercise and morbity/mortality, the current study just establishes that this also correlates in the bigger picture. So yes the study itself does not "proof" causality, it's just a piece in the bigger puzzle of causality.
The beauty of an end-to-end RCT is that it effectively neutralizes other possible causality mechanisms. You do not seem to appreciate this. My impression is that your reasoning is more in line with evidence based medicine (EBM), rather than the hypothetico-deductive method that I personally subscribe to. In my way of thinking there will never be definitive “proof” of causality, but I will happily take a drug that has gone through sufficiently powerful RCTs that failed to prove its ineffectiveness (and harm).
> A RCT does not establish causality. That's essentially my point. A single study/experiment never proofs causality. You need a theory to explain the causality and multiple studies that falsify other possible causality mechanisms.
This might be a good heuristic for assigning confidence to results, but in theory, an RCT absolutely does establish causality, assuming internal, external, construct, and statistical validity of the study.
Multiple studies may not be better than one good study (assuming the above), which can be tested by looking at the leverage in a meta-anlaysis.
Having a theory is kind of orthogonal to a study finding a true positive result. Almost every published study, true or false, invokes some kind of theory, true or false. There is a joke in the soft sciences along the lines that you make up a new theory for every study.
> Or perhaps you don’t fully understand the challenges of establishing causality? Just because an intervention causes an improvement in some bio markers that are associated with lower mortality (unfortunately) does not mean that the intervention will cause lower mortality.
> The classic example is vitamin D supplements: Higher vitamin D levels are associated with lower mortality in many medical conditions. Vitamin D supplements increase vitamin D levels. But vitamin D supplements seldom lower mortality.
That is quite a simplification. The research as far as I know is that too low Vitamin D levels are associated with higher mortality, but that supplementation above a certain level is meaningless (and most caucasians can get those levels through normal sun exposure). So yes vitamin D supplements in general don't improve mortality.
Note also the situation here is completely different. The link between mortality and morbidity markers and exercise have been established in other studies. The study here establishes that this directly translates to a correlation between fitness and mortality. So in a way it's the opposite of the vitamin D case.
> Why? Probably because vitamin D is produced in the skin when we are in the sun. The more healthy subpopulation of any study will typically spend more time outside, so they will have higher vitamin D levels. But it’s (relative) health that causes higher vitamin D, not the other way around.
You realise that you are proclaiming causality here?
> The only way to reliably establish causality is really an end-to-end randomized controlled trial (RCT). Stitching together two RCTs is not sufficient.
A RCT does not establish causality. That's essentially my point. A single study/experiment never proofs causality. You need a theory to explain the causality and multiple studies that falsify other possible causality mechanisms. That has been done extensively for excercise and morbity/mortality, the current study just establishes that this also correlates in the bigger picture. So yes the study itself does not "proof" causality, it's just a piece in the bigger puzzle of causality.
> (Not saying that exercise does not lower mortality BTW, just that it’s complicated, and a study such as this is probably picking up two signals: one causal and one purely correlational.)