Can We Trust the Polls?

Nearly everyone, including Clinton campaign insiders, thought she would lose the New Hampshire primary. Some people (not naming any names) even thought it would be a blowout. This was based on wide, and consistent polling showing Obama with a big lead. This morning, it sure looks like those polls were way off base. Does this discredit the polls? Can we ever trust again?

I think the first thing to highlight is that the polls were only partly wrong. The weighted average support for Obama in the post-Iowa polls was 37.1%. Obama's actual vote share: 36.4%. The polls were spot on for Obama's support. This should put to rest any suggestion that Clinton's victory was a product of the "Bradley Effect." Essentially, the Bradley Effect is when voters tell pollsters that they're willing to vote for a Black candidate, but then when it comes time to actually pull the lever in the privacy of the voting booth, that support melts away. It's named after LA mayor Tom Bradley, and examples of it include, of course, Bradley himself, as well as NY mayor David Dinkins. It should also be noted this "effect" was absolutely absent in Obama's 2004 primary victory over several well-funded, establishment, white Democratic opponents.

So if the polls got Obama right, then how come they got Clinton's support wrong? As I noted in the pre-New Hampshire post, polls were also showing that about 9% of the likely Dem voters were undecided. Now obviously, not every undecided ultimately voted for Clinton, but exit polling does suggest that voters who decided very late broke more heavily for Clinton than for Obama (although, again, not by enough to explain the entire Clinton boost). Another likely source of additional Clinton voters came from Edwards and Richardson, both of whom underperformed by about 2% each (compared to the polls). Those voters, like the undecideds, seem to have gone mostly to Clinton.

So, if Clinton gained a 4 or 5 point net bump (meaning that some undecideds went to Barack, but some Barack voters also changed their minds and went to Clinton) from the undecided break, and then another 2 or 3 net bump from Edwards and Richardson supporters, that explains somewhere around 7 or 8 points of the 10 point gap between polling and actual. The rest could have been ordinary sampling error.

Yes, the polls were wrong. But not quite as wrong as it first appears.