MANCHESTER, New Hampshire — Pollster Andy Smith of WMUR is pushing a poll that shows former Sen. Scott Brown down 8 points, contradicting the results of most other polls that have come out.

The poll results were announced on local television here on Thursday afternoon just a couple hours before the final debate between Brown and incumbent Democratic Sen. Jeanne Shaheen (D-NH). The poll shows Shaheen with 50 percent and Brown with 42 percent. The poll was conducted from Oct. 22 through Oct. 26 and includes a sample of 555 likely voters with a margin of error of 4.2 percent.

The poll is definitely an outlier, as Brown and Shaheen are within the margin of error of each other in polls from just about every other pollster in business.

If he ends up being wrong, it wouldn’t be the first time Smith has done polling that’s drastically inaccurate right before elections. On January 10, 2010, while working for the Boston Globe, Smith released a poll that showed Brown–then running for Senate in the special election for the late Ted Kennedy’s seat–down 15 points to Democrat Martha Coakley.

“Democrat Martha Coakley, buoyed by her durable statewide popularity, enjoys a solid, 15-percentage-point lead over Republican rival Scott Brown as the race for US Senate enters the homestretch, according to a new Boston Globe poll of likely voters,” Matt Viser and Frank Phillips wrote for the Globe, citing Smith’s poll for the newspaper. 

“Half of voters surveyed said they would pick Coakley, the attorney general, if the election were held today, compared with 35 percent who would pick Brown. Nine percent were undecided, and a third candidate in the race, independent Joseph L. Kennedy, received 5 percent. Coakley’s lead grows to 17 points – 53 percent to 36 percent – when undecideds leaning toward a candidate are included in the tally. 

“The results indicate that Brown has a steep hill to climb to pull off an upset in the Jan. 19 election. Indeed, the poll indicated that nearly two-thirds of Brown’s supporters believe Coakley will win.”

Smith, who was quoted in the story, was even bold enough to chide Brown for seemingly failing, according to his results. “She’s simply better known and better liked than Brown,” Smith said of Coakley. 

Smith worked for the University of New Hampshire Survey Center, which conducted the poll for the Boston Globe. 

“If there ever was a time for a Republican to win here, now is the time,” Smith continued. “The problem is you’ve got a special election and a relatively unknown Republican going up against a well-liked Democrat.”

Of course, nine days later, Smith was proven incredibly wrong. Brown beat Coakley for the seat by five points, winning 52 percent to Coakley’s 47 percent. Smith was off by 20 points nine days before an election.

In 2008, Smith conducted a poll for CNN, WMUR, and UNH that showed then-Illinois Sen. Barack Obama winning the Democratic primary in New Hampshire by 9 points. Smith’s survey of 599 likely voters had Obama at 39 percent, Hillary Clinton at 30 percent, and the other candidates well below that. However, Clinton’s 39 percent beat Obama’s 36.4 percent by about 3 points–meaning Smith’s poll was off by a 12-point swing, despite being conducted just a few days before the primary.

Smith’s failure with that poll was even highlighted by the Denver Post; the paper interviewed him over it, and he blamed evolving technology for his misreads.

“Technology has made it easier not to respond,” Smith said. “Before answering machines, everyone picked up the phone and didn’t want to miss calls. Now there’s caller ID and call blocking and call waiting, which makes it easier to screen calls, which led to the decline in the response rate.”

Smith also likened accurate polling to having ice cream. “It’s like walking up to an ice-cream stand,” he said. “What flavor do you want?”

Then there are a couple examples of wacky polling Smith’s done just this year. A poll he conducted three weeks ago showed New Hampshire’s Second Congressional District Republican candidate Marilinda Garcia up 4 points, leading incumbent Democratic Rep. Annie Kuster (D-NH) 41 percent to 37 percent.

Then a poll released on Tuesday of this week shows Kuster beating Garcia by 23 points. The new poll shows Kuster with 53 percent and Garcia with 30 percent–meaning that somehow over three weeks, the race took a 27-point swing, if one were to believe Smith’s polling. That doesn’t happen in politics unless a candidate is caught up in a drastic scandal–on the magnitude of being found with a hooker in the trunk of their car–especially in the final weeks of a campaign.

Then, in several polls over the course of this year that Smith conducted in Maine, Smith used the wrong language to poll the public to gauge interest on a hunting ban in the state. When Smith was caught, he refused to retract his statements on the matter despite protests from local reporters.

“The polling organization that conducted a statewide poll on Maine’s bear baiting referendum for the Portland Press Herald/Maine Sunday Telegram says there is no need to retract the results even though pollsters mistakenly asked prospective voters if they wanted to make it a crime to hunt bears with bait, traps, or dogs,” the Portland Press Herald‘s Dennis Hoey wrote on Wednesday. “Polls conducted in June, September, and October by the University of New Hampshire Survey Center each included questions that used language from the 2004 bear baiting referendum.”

The question from the 2004 measure asked voters: “Do you want to make it a crime to hunt bears with bait, traps or dogs, except to protect property, public safety or for research?”

The question on the upcoming ballot in Maine, Hoey wrote, “uses slightly different language.” It asks voters: “Do you want to ban the use of bait, dogs, or traps in bear hunting except to protect property, public safety, or for research?”

Smith wouldn’t retract his polling when asked, because, in the words of the local reporter, “the words used in the survey don’t really matter now.”

“We’ve found that the knowledge around this question is quite high for a referendum,” Smith himself said. “People are not likely to change their minds (when they vote). Despite the differences in language, the way people vote won’t change.”

But a different polling firm, Pan Atlantic SMS which is based in Portland, Maine, did announce it was retracting its polling data because it made the same mistake.

“Pan Atlantic SMS, a Portland polling firm, said it was retracting poll results released Tuesday on the bear-baiting question because the poll used the language from the 2004 referendum,” Hoey wrote.

After taking all that into consideration, comparing the polling from respected pollsters in New Hampshire to Smith’s indicates his poll coming out is an outlier. New England College, the most accurate pollster in the 2012 election according to Real Clear Politics as its final three polls were the closest to the final result, has Brown up one point at the beginning of this week. Several other polls released over the past few days have Brown down anywhere from one to three points. A poll released earlier on Thursday by American Research Group (ARG) shows Shaheen and Brown tied at 49 percent, after previous polls from that firm showing Shaheen leading by one or a few points.

Shaheen herself–who certainly has access to her own campaign’s internal polling–has been saying over the past several days at campaign stops that this election will come down to turnout. Sure, it’s still a tossup and could go either way–to Shaheen or Brown. But this poll is an outlier from a pollster with a checkered history, and if the media reports it aggressively, it could be used inaccurately and unfairly as a momentum killer against Brown.

For what it’s worth, bloggers at the liberal Daily Kos came to the same conclusions as Breitbart News about the accuracy of Smith’s polling in a post before this new Shaheen-Brown race poll was announced.