A glance at the graph above should explain everything. Though many assume the slope of the "perception graph" is monotonically negative, the slope is actually positive.
Ah, so everyone is pulled toward the average of what people think, not the average of what people are. My reading nerfhammer as saying the latter is, I think, what is responsible for my unease with his characterization.
Not really here or there, but my assumption from what I'd read on the effect was "not monotonically positive", not "monotonically negative" (which would be more surprising).
"Although test scores put them in the 12th percentile, they estimated themselves to be in the 62nd."