@charliehall (quoting and criticizing me saying) “there seems to be a tendency”
of course, I though I qualified enough with “seems”, passive voice, and “tendency” 🙂
but we need to treat uncertainty appropriately: there is current estimate and, then, there is an estimate of a chance that this path can lead to success, that is E [P(X>t)] v E(X), pardon my French.
So, when allocating resources (and hopes), you need to look where success is possible.
Look up literature on multi-armed bandits.
In this particular case, I see similar age-related pattern repeating in 3 cases, so it is worth investigating – and it does correspond to the rumors floating around. So, we should not reject “un-scientific” experiments, collected without controls, we just need to look for ways to validate them.
With this data, do you know if more detailed data is publicly available, it would much more productive to review it in more detail, say, separate by country, use train/test to generate hypotheses on one set, and test on the other.