Perhaps a measure way out on the tail of a distribution can revert most of the way toward the mean without significantly affecting other, nonlinearly correlated measures? In that case, even the grimmest nightmares of racists and prison guard unions could still come to pass, just not yet...