"For C60, 50% of the animals live longer than the controls, and the other 50% die sooner than the controls. The overall increase in longevity seems to be zero."
I disagree with this reduction not least because it ignores the other 200 plus implicit controls of the other 20 groups. As stated, the top 6 or 7 longest lived are c60oo, out of 240 - this isn't massaging data to make c60oo look good. The 224 mice is a far more meaningful comparison as a control group than the study designed group of 12. The 12 controls might be a better comparison group than any other individual group, but not better than the aggregated mass of them all because what the 224 lose on control conditions they make up for in numbers. And too of course once observing how tightly grouped all the others are, bar three or four, it becomes an implicitly more reliable control group. So, what do we believe, that the first 50 percentile is not representative of c60's impact on longevity, but the second group is. Or vice versa, or both are or both aren't? Those are the possibilities.
The key confident statistic which informs on which of those possibilities is likely, is the absurd improbability that c60oo could find the top 6 positions by chance with say 100 mice still going. And more to the point, that the longevity margin is so significant for the top 25 percentile compared to every other group. So that those guys have benefitted from c60oo supplementation would seem to be next to impossible to refute, if the study is being conducted properly, naturally.
So what of the lower 50 percentile? Well, this suggests either the presence of subclasses which c60oo alone is expressing - those amenable to c60 life extension, and those not amenable (say with tumours). Or they were very unlucky.
The symmetry of your graph creates an illusion of normal distribution, a balance of c60's impact either side of the control group, something of a proprosed mean. But that might be the case with say height or weight, but not longevity, not at least when the life extension is exceptional.
Nature finds far easier ways to intervene to dramatically cut short life, than to extend it: its easier to exit at 60, than 100, say. So in that sense, the LHS of the chart requires a far larger sample to be convinced its an expected effect than the one on the RHS, providing the RHS is at a suitable extreme. Hence it is far more likely the underachieving first 50 percentile is down to bad luck, than the over-achieving second percentile is due to blessed fortune - it is just too improbable to achieve such a result. I would bet for example, if we had just 20 control groups, and ran the study millions of times, the first 50 percentile mortality numbers of the c60oo group would be mirrored thousands of more times by control groups, than the second. That is why, I say it demonstrates a longevity effect, both samples appear outliers compared to the small sampled control but they are not at all equivalent outliers, even though they appear so visually. And more to the point to describe the LHS as an outlier to the control group is not fair because the control group would be a poor benchmark due to its sample size. If drawing on the other 20 groups as a quasi-control, as I suggest, then c60oo is not doing so poorly at the halfway stage and the suggestion might be that the control group is overperfroming, which at that sample is more than possible.
My view is that possible there are certain risks the mice need navigate, in order to expereince c60's wind in their sales. But to be confident in that effect, rather than luck or variance, we'd need a considerable increase in sample size. Not so for the second percentile, which is a highly confident result.
Edited by ambivalent, 05 September 2022 - 02:58 PM.
, as I didn't know if I am going to collapse. I drank 2 liters water to dilute the mix.
