The New Hampshire polls were stunningly wrong, of course. ABC polling director Gary Langer calls it "unprecedented." So what was the problem?
Pollster's Mark Blumenthal isn't sure, but thinks it has something to do with methodological decisions regarding the college-educated women subgroup particularly. Polls showed Obama leading Hillary Clinton among New Hampshire women by a few percentage points and she in fact beat him by double digits among female voters. Blumenthal also (and relatedly) notes that today's exit polls show 37% of Democrats only "finally" decided who to vote for within the last three days. So perhaps people weren't as sure as they thought they were when they talked to pollsters over the weekend.
Some random thoughts:
- Public opinion pollsters (as well as the media pundits that love to cite them) often forget to consider the actual effect of the polls themselves. Polling isn't merely an academic exercise. Poll results enter public discourse and help frame the debate. Independents in particular might very well have seen Obama leading by a solid margin and decided to vote for John McCain in the Republican primary instead, thinking (perfectly rationally) that an Obama victory was a sure thing. That said, Matt Zeitlin is right that Obama and McCain have no obvious similarities except cross-partisan appeal and having the media love them. On the issues, it's hard to find starker policy differences in the current race.
- Pollsters and pundits thought most of the remaining undecided voters would break for Obama over Clinton based on the assumption that people tend to remain undecided because they're considering the relatively riskier "insurgent" candidate over the safer "incumbent" one. And the media loves to claim Clinton is running as essentially an unelected incumbent. But maybe that's not voter perception. Particularly after Obama's stunning Iowa victory and the media's assertion that Clinton had no chance in New Hampshire, her position as "incumbent" crumbled and voters didn't want to coronate Obama.
- The gender gap really is key here and has to be explained by something. Was it Clinton's display of emotion? Her proclaimation in the debate that electing a woman as president would constitute quite a bit of "change"? The charges Obama's record isn't pro-choice enough? This woman? Obama did so well in Iowa at least in part because he neutralized an obvious Clinton base. If he intends to win the nomination, he needs to find a way to do it again.
- It probably wasn't racist lying white people (the Bradley/Wilder effect). As Matthew Yglesias notes, the polls were actually entirely accurate in predicting Obama's performance (same for John Edwards). They simply under-predicted support for Clinton, so it seems like they were wildly off-base. It's just not the case that white voters said they'd vote for Obama and didn't. They basically did. The problem is no one predicted Clinton's potential.
In some ways, this simply shows the limits of polling. No methodological tool is perfect. While polling can explain a lot, it can't explain everything. Pollsters need to start being more self-aware of how polls themselves become part of the campaign, in addition to second-guessing assumptions about where undecided voters will eventually break. The media is perhaps the real culprit though, because polling results necessarily explain the scene before those poll results become part of public discourse. If the way the poll results are debated in public discourse in itself changes public opinion, the value of public opinion research on time-specific things like primary elections becomes moot.
Comments