WASHINGTON -- If ever there were a time when polls should be certain about something, it is that Mitt Romney will win New Hampshire's Republican presidential primary on Tuesday. At least 53 times over the past two years, and at least 24 times in the last two months, media pollsters have measured the preferences of likely voters in New Hampshire, and found Romney leading the Republican race every time, usually by large, double-digit margins.
The final round of tracking polls taken over the past weekend shows Romney leading Ron Paul and the rest of the candidates by margins of 15 to 24 percentage points. By the metrics of polling, Romney's victory in New Hampshire appears virtually assured.
Yet the same surveys are also full of uncertainty, particularly as reported by the voters themselves, and that margin of doubt leaves open questions about the size of Romney's likely victory and the identity of the candidates who finish second, third and fourth.
Consider these results, all measured over the campaign's final weekend:
- The WMUR/University of New Hampshire (UNH) poll finds that less than half the likely voters (44 percent) say they have "definitely decided" who they will vote for, 27 percent are "leaning to someone" and 29 percent are "still trying to decide."
- Nearly a third of the voters on the News7/Suffolk University tracking poll conducted Saturday and Sunday nights are either completely undecided (12 percent) or say it is very likely (2 percent) or somewhat likely (20 percent) that they might change their minds.
- The Public Policy Polling (PPP) automated survey finds more than a quarter of the likely voters are either totally undecided (4 percent) or say they might end up supporting someone other than their first choice (23 percent).
Self-reported uncertainty alone does not foretell polling surprises, though it may be an important warning sign. The final UNH tracking poll before the 2008 primary found almost exactly the same level of uncertainty as this year among the Republicans and slightly less among the Democrats. Yet multi-poll averages of the final polls on the Republican race came very close to correctly predicting John McCain's six percentage point margin over Mitt Romney, while the final polls on the Democratic race understated Hillary Clinton's support by 8 to 10 points, leading to one of the most infamous polling snafus in recent memory.
In addition to voter reports of uncertainty, the polls are also producing somewhat contradictory stories about late trends. The 7News/Suffolk University tracking poll shows Romney's support dropping 10 percentage points in a week to just 33 percent of the vote. Meanwhile, the University of New Hampshire poll shows a five percentage point drop over the same period with Romney's support at 41 percent. The final polls by other organizations that have not tracked as often peg Romney's support somewhere in between.
Polls are also projecting different levels of support for Jon Huntsman. The Suffolk and UNH surveys, for example, show Huntsman running a clear third, with 11 and 13 percent of the vote respectively. Meanwhile, the final surveys from PPP and the American Research Group give Huntsman 16 and 18 percentage points respectively, enough to be in a virtual tie for second with Ron Paul.
Do polling methods explain these discrepancies? The polls cited here are certainly not created equal. PPP samples from voter lists and uses an automated, recorded-voice methodology. The other surveys use live interviewers and sample telephone numbers generated by a random digit method in order to ensure representation of listed and unlisted landline numbers. Based on the information provided in the survey reports, the Suffolk University poll appears to be the only one of the final weekend polls that dialed a separate sample of mobile telephones (drawn from a commercial list of working cell phone numbers).
But the connection between the varying methodologies and the differences in the survey results is far from obvious. The more important point may be simply that the apparent "house effects" are a symptom of the underlying voter uncertainty. Some voters are having a hard time making up their minds, so subtle variations in how pollsters ask their questions, how hard they push undecided voters for a choice and how they define the likely electorate may be producing different results.
The director of the Suffolk University Political Research Center told The Huffington Post that he considers most of the polls are "pretty much within the same ballpark," showing the same rough order of finish and projecting results for each candidate "with the error rate."
If that proves to be the case, then an average across all polls should produce the clearest picture of the outcome. The chart below, featuring the HuffPost Pollster trend lines based on all public polls in New Hampshire, does just that. As of this writing it shows Romney's support declining slightly (to 36.8 percent) followed by Ron Paul (at 17.6 percent), with Huntsman just a point and a half behind (at 16.0 percent) and rising fast, followed by Rick Santorum (11.6 percent), Newt Gingrich (10.0 percent) and Rick Perry (0.9 percent). Huntsman's momentum in on a track to catch Paul, though who will finish on top is one of those things about which polling simply cannot be certain.

Note: The News7 and Suffolk University continued to poll on Monday night and planned to release a final sample of interviews on Tuesday morning. We will update this post when those results are available.