Conformity in Political Polls
The 2014 midterm election results have got people asking, why were the polls biased towards the Democrats? In fact, for a given election cycle, it’s common for the polls to be consistently biased towards one party, although that party is as likely to be the Democrats as the Republicans.
One possible explanation is that some polling companies copy others. If you run a poll, I can think of two reasons why you might do this, especially if you know your poll is worse than average. It’s likely that all the other polls collectively know something you don’t, and even if the other polls are all wrong, at least you won’t damage your credibility (relative to the other polls) if you follow the crowd.
FiveThirtyEight found that second-rate polls were more accurate in races where a “gold standard” poll occurred than races where no “gold standard” poll occurred, suggesting that second-rate polls copy the best ones where possible. (After collecting their data, the second-rate polls tweak some parameters in their formula so they get similar results as the better polls.) As a result, inaccuracies in a small number of polls can create a consistent bias in the polling averages.
http://fivethirtyeight.com/features/the-polls-were-skewed-toward-democrats/
http://fivethirtyeight.com/features/are-bad-pollsters-copying-good-pollsters/