The fact that you think this way is the thing that prompts me to continue replying.
Because you think this, but it is actually, to use your own words, a delusion.
No.
While, yes, the difference is there and, yes, dividing the population by the data we’re also evaluating is weird and likely to reduce the visibility; if the difference was hugely important then it would not nearly vanish when divided.
The fact that it turns from a valley to a pothole is not because we’re looking at the same thing that we’re splitting by, it’s because the thing you’re talking about isn’t very real. (It is somewhat real, obviously. The data indicate to us quite suspicious things.)
Nobody’s saying it is the same.
Consider that the whole thing with the admissions data kerfuffle started because of “the entire data” saying one thing, whereas when you subdivided the data it says something completely different.
The fact that, in the textbook example, we have the luxury of dividing the groups by a second actual metric, while here we’re dividing by the metric we’re investigating, does not diminish the importance of how the act of dividing the data reveals something about the whole data set that means that considering the whole of the data has an inherent flaw.
In the case of the department admissions, looking at the overall data says that males’re almost a third more likely to be accepted; but most department’s data says you’re more likely to be accepted if you’re female.
In the case of here, looking at the overall data says that Terran has lower average MMR and that there’s a low number of high rank Terrans compared to its population; but their MMR doesn’t lead or lag significantly if you look at just the top or bottom halves.
What that says is that by only dividing our groups to control for the absolutely phenomenal number of low ranked Terran players, the problematic observation we make from the overall data nearly vanishes.
Do you see how that, in turn, is the prompt leading to the other arguments - about Terran’s familiarity, approachability, new-player friendliness, that learning the game also implies playing other races, about relative match up frequency - as suggestions for why that’s the case, because nobody thinks that it’s that Terrans are comparatively worse, but there must be some underlying reason that the facts are what they are?