Another thing from software development: I know companies that try to decrease this effect when making estimates in planning, by giving every team member a stack of cards with possible guesses (1 day, 2 days, 3 days, 5 days, etc). Each member chooses a card in secret, then everybody turns theirs over at the same moment. If everybody more or less agrees, then that value is chosen. If there's disagreement, they talk more (typically people who chose outliers get to explain why), then the procedure is repeated.

First step is clearly good, in the second step conformity is probably a factor again.

Another interesting point (not for this discussion, but for software estimates :-)) is that the numbers on the cards are the fibonacci series, and they don't actually have hours or days, but "points". The project leader has historical data that tells him how long one "point" actually took in past projects.