I’m interested in historical sea-voyages of discovery, and I was intrigued to read that on Columbus’s 1492 voyage, his crew were ill at ease about the proposed journey through uncharted waters for an indefinite period to an unknown destination. To reassure the sailors, Columbus concealed the true length of the journey by keeping two logs: one the “real” distance that he calculated using dead-reckoning; the other giving fictitious, shorter distances that he shared with the crew, so they would think they were closer to home than they really were. The irony was that the falsified figures were more accurate than the ones Columbus kept in his “true” log.
This is an unusual case of one systematic error (through Columbus’s dead-reckoning techniques) being exactly cancelled out by another (his falsification factor) to arrive at the correct figure.
One way to reduce this sort of error in project delivery is to break the work into a large number of small, easily-estimable tasks. If each task is estimated in the same manner, the large number of small estimation errors are more likely to cancel out than by estimating a small number of large tasks.
Update: an associate of mine, Jon Turner, has pointed out that the elimination of the root cause of systematic error – rather than just fixing the resultant problems – is one of the central tenets of lean.