To clarify: suppose one figures that a lineup of 10 people has a 33% chance of identifying the right guy, a 33% chance of identifying the wrong guy, and a 33% chance of identifying nobody. If the police have a suspect and two people ID him, that increases the likelihood that they in fact have the correct suspect. On the other hand, there's a 1 in 27 chance that two people will ID the same wrong suspect if their mistakes are independent; if the wrong suspect has some trait that increased his likelihood of being picked, the odds of erroneous double-ID go up substantially.