It’s possible that there are different classes of responders, and it might be possible to detect this from the data they have. I doubt that they’ve done the right statistical test, though- you’d be looking for a mixture of three Gaussian peaks, rather than one Gaussian which has a large enough standard deviation to be represented both around zero and below zero. (Without a link to the paper, I don’t care enough to check.)
It also seems like you could check the test-retest reliability of this- record stats, have people not exercise for X amount of time, then exercise for Y amount of time, then not exercise for X amount of time, then exercise for Y amount of time. Are the people who get worse/better the first time the same as the people who get worse/better the second time? Or do the groups mix like you would expect from it being random noise?
The distribution is not multimodal, but very smooth. Timmons mentions test-retest reliability, but it’s not clear that he means trying the intervention repeatedly vs just measuring VO2Max repeatedly. And his citation is not very precise, but to the Heritage family study. Here they say that response is heritable, which requires that it be individually real.
I am also a little skeptical that they’ve done a good job of separating within person variation and between person variation. However I think a multimodal distribution is the wrong way to look for this (I think multimodal distributions are rare). Your second suggestion is much better.
However I think a multimodal distribution is the wrong way to look for this (I think multimodal distributions are rare).
I suppose; I was taking the view that there were different classes of responders, when it could be the case that each person has some “response to exercise” number that smoothly varies from -.2 to 1, or whatever, in which case you wouldn’t see a multimodal distribution (but you would still see high test-retest reliability).
It’s possible that there are different classes of responders, and it might be possible to detect this from the data they have. I doubt that they’ve done the right statistical test, though- you’d be looking for a mixture of three Gaussian peaks, rather than one Gaussian which has a large enough standard deviation to be represented both around zero and below zero. (Without a link to the paper, I don’t care enough to check.)
It also seems like you could check the test-retest reliability of this- record stats, have people not exercise for X amount of time, then exercise for Y amount of time, then not exercise for X amount of time, then exercise for Y amount of time. Are the people who get worse/better the first time the same as the people who get worse/better the second time? Or do the groups mix like you would expect from it being random noise?
The distribution is not multimodal, but very smooth. Timmons mentions test-retest reliability, but it’s not clear that he means trying the intervention repeatedly vs just measuring VO2Max repeatedly. And his citation is not very precise, but to the Heritage family study. Here they say that response is heritable, which requires that it be individually real.
I am also a little skeptical that they’ve done a good job of separating within person variation and between person variation. However I think a multimodal distribution is the wrong way to look for this (I think multimodal distributions are rare). Your second suggestion is much better.
I suppose; I was taking the view that there were different classes of responders, when it could be the case that each person has some “response to exercise” number that smoothly varies from -.2 to 1, or whatever, in which case you wouldn’t see a multimodal distribution (but you would still see high test-retest reliability).