The general form of this problem is that the correlations, including things like retest and test-to-test repeatability, all come from the fat part of the curve

The general form of this problem is that the correlations, including things like retest and test-to-test repeatability, all come from the fat part of the curve (say +/- 2 sd). But people love to extrapolate those findings to the part of the tail where there is no good data.