I know you're probably not implying regression to the mean is causational, but that was my initial reading so I want to clarify for those who may not be familiar with the concept.
Regression to the mean is simply that any given datapoint is most likely to be the mean, or close to it. This means that any exceptional data point,up or down, can be expected to be followed up by the mean.
The example of this being misinterpreted that I am familiar with is that of a flight instructor's belief on training. When a pilot performed well, they wouldnt comment. If a pilot performed poorly, they would be punished. They believed this was better because when they praised a pilot, they would usually do worse the following run, and when they punish them they do better. This isn't technically wrong, they are just ascribing causation where there is none. I think I saw this example in Signal vs. Noise, but I'm not sure.
Basically, regression to the mean isn't a reason to pick someone who did poorly, it's a reason that that person will do no worse or better than they do normally.
Yeah, I always make an example with coin flips to show how this is true.... lets say heads is success and tails is failure.
Flip 100 coins. Take the ones that 'failed' (landed tails) and scold them. Flip them again. Half improved! Praise the ones that got heads the first time. Flip them again. Half got worse :(
Clearly, scolding is more effective than praising.
Except, coins are not humans with emotions, and neither can they dupe probability to improve their outcome. The 50 heads(success) that you left are not going to do any better than the 50 tails you scolded. You are just changing the sample space in a biased fashion to prove your point.
I think you are totally missing my point. The whole point I am trying to make centers on the fact that the coins are all actually equal. The observed difference in performance is entirely due to chance.
While, of course, human performance is not always equal in the same way our coins are, the fact that performance (or whatever it is you are measuring) will still regress towards the mean still holds true. The coin example just gives an extremely obvious demonstration about what is happening when things regress towards the mean.
Regression to the mean is simply that any given datapoint is most likely to be the mean, or close to it. This means that any exceptional data point,up or down, can be expected to be followed up by the mean.
The example of this being misinterpreted that I am familiar with is that of a flight instructor's belief on training. When a pilot performed well, they wouldnt comment. If a pilot performed poorly, they would be punished. They believed this was better because when they praised a pilot, they would usually do worse the following run, and when they punish them they do better. This isn't technically wrong, they are just ascribing causation where there is none. I think I saw this example in Signal vs. Noise, but I'm not sure.
Basically, regression to the mean isn't a reason to pick someone who did poorly, it's a reason that that person will do no worse or better than they do normally.