Some of it's the regression effect, where the extreme results really are flukes -- and the achievers are aware of it -- while average results are more likely to

Some of it's the regression effect, where the extreme results really are flukes -- and the achievers are aware of it -- while average results are more likely to be statistically representative. Of course, "aware of it" is key, which is why the results sometimes diverge at the other end of the scale.