In New Hampshire, poll errors hide the real state of the race
© Getty

What is it about New Hampshire that attracts so much attention during presidential elections?

ADVERTISEMENT

It has only four electoral votes, the fewest of any “swing state,” and has reliably voted Democrat for president since 1992, with the notable exception of the 2000 election in which George W. Bush defeated Al GoreAlbert (Al) Arnold GoreThe Hill's Morning Report - In Nevada, bets on Sanders, eyes on Bloomberg Mellman: Primary elections aren't general elections Overnight Energy: Green groups to sue over Trump rollback of Obama water rules | GOP climate plan faces pushback from right | Bezos launches B climate initiative MORE by a mere 7,200 votes.

Bush would likely have lost the state had Ralph Nader not received more than 22,000 votes, surely enough to have put Gore over the top. Gore might have been able to win New Hampshire and therefore the presidency, regardless of what happened in Florida, if his campaign had focused more on the Granite State, perhaps visiting a time or two in the waning weeks of the campaign.

Fear drives politics, and a look at the campaign schedules of the candidates and their surrogates show that both the Clinton and Trump campaigns don’t want to wake up Wednesday morning with déjà vu from the Gore campaign.

Hillary ClintonHillary Diane Rodham ClintonClinton asked if she'd be Bloomberg's vice president: 'Oh no' Trump launches three-day campaign rally blitz Free Roger Stone MORE visited New Hampshire Sunday, President Obama will be at the University of New Hampshire Monday, and Trump visited Friday and will hold his final rally in Manchester Monday night.

The actions of the campaigns say that both believe the state to be too close to ignore. But how do we really know? Polls, of course.  

If the 2000 campaign taught the public and pundits anything it’s that there are 51 elections for president, not a single national election. This led to an increase in state-level polls, and that increased focus coincided with a decline in response rates in traditional polls and an explosion of new, less expensive polling methodologies.  

Between 2008 and 2014, the percentage of live interviewer polls conducted in battleground states in the final two months of the election decreased from 53.4 percent in 2008 to 43.5 percent in 2014 while IVR polls (robo-polls) increased from 39.8 percent to 43.5 percent. Over that same time, online polls increased from 6.8 percent to 13.0 percent.  

Most readers and members of the media will surely ask, “so what?”

Increasingly, polls are reported without consideration of the methodologies used and the impact those methodologies are known to have on poll results. This can seriously impact interpretations of the state of the race in any given state.

MORE STORIES FROM THE HILL:

What's going on in these final week election polls?

Election 2016: Like it or not, you still need to vote

Will blacks swing the election to Trump?

Live interviewer polls almost always start with a sampling frame (the list that potential respondents come from) that ensures all landline and cellular telephones (with some caveats) are included and respondents are randomly selected.

Interactive voice response (IVR) polls (also known as "robo-polls") are prohibited by federal law from calling cell phones. Consequently, the samples in IVR polls are biased toward people with landlines (older, wealthier, and those who have lived in their home longer) and against households who only have cell-phones (younger, lower income, urban and minority).  

It’s easy to see how this can lead IVR polls to have more Republican samples as older wealthier voters are more likely to be Republicans while younger, lower income and minority voters are more likely to be Democrats.

IVR polls have tried to compensate for this bias by supplementing their samples with web-based survey samples. These web-based samples are, to a greater or lesser extent, not selected at random and likely cannot be used to infer opinions of entire populations of people.

Another sampling frame that is often used is lists of registered voters. But many of these lists suffer from the same problems that landline samples have — they have limited household coverage in that they systematically exclude all people who are not registered to vote, as well as people who may be registered, but have only cell-phones or unlisted landlines. Again, the samples tend exclude younger, lower income, new arrivals as well as minority voters — in other words, Democrats.

So what does this have to do with New Hampshire?  

New Hampshire has one of the highest turnovers in its population of any state in the country.  Approximately one-third of the potential voters in 2016 either did not live in New Hampshire in 2008 or were not old enough to vote in the 2008 election.

IVR surveys and surveys using lists of registered voters systematically exclude significant numbers of Democratic voters causing them to, in my opinion, overestimate Republican support.

While polling provides important insights into how states will vote, there are important implications that should be considered when examining them. It is critical to examine the methodology used, the sponsor of the poll, the voting history of the state as well as significant changes to the state population in recent years.

Polls have become the thermometer of elections, but campaign junkies should understand the limitations of the imperfect measure that polls provide.

So while pundits try to suss out the impact on the polls of campaign events, such as the recent announcement by FBI Director Comey, they pay far less attention to understanding more basic issues of survey research.

Smith is the director of the University of New Hampshire’s Survey Center and an Associate Professor of Practice in the Department of Political Science.


The views expressed by Contributors are their own and are not the views of The Hill.