Below are the updated data tables and cross tabs for the University of New Hampshire Survey Center/Portland Press Herald poll. Those of you who follow me on Facebook may have seen some of the methodology details and answers to some other questions. The full report should answer the rest.
Of course, since this is a very close race there are going to be people who slam the methodology, sample size, questions, which statistics we used, etc.
Let’s address one anticipated criticism right now.
In Sunday’s story we used the topline of likely voters to show that the gubernatorial horse race is essentially a statistical dead heat between Republican Gov. Paul LePage and Democratic U.S. Rep. Mike Michaud. Those who look at the topline with leaners in the full poll will notice that Michaud actually gains a percentage point, from 40 percent to 41 percent, LePage stays the same at 36 percent and independent Eliot Cutler gains a point, from 15 percent to 16 percent. Michaud supporters might argue that the extra point puts the Democrat outside the polls 4.3 percent sampling error and gives him the lead.
Well, it does, by exactly .7 percent. We didn’t use the number because it’s statistically insignificant, particularly when you considered that the nearly half of the electorate is largely undecided, ambivalent about the candidates or disengaged from the race. In the horse race question, the pollster effectively asked respondents to make up their mind on the spot. The leaner question took the mandate one step further, by essentially asking someone who said they were unsure in the first question to say which way they are leaning. That’s why the undecideds dropped from 7 percent to 5 percent in the leaner question.
That’s why we used the likely voter number. If both numbers show a dead heat five months from Election Day and half the electorate is undecided then there’s no point to use a number that gives one candidate a statistically insignificant lead. Sure, the Michaud campaign would have dug it, but it’s of no consequence to readers.
Andy Smith, the director of UNH Survey Center, said we made the right call. He said there is no “substantive difference” between the two numbers. Additionally, he said, using the leaner result would have taken a somewhat artificial result in the horse race — artificial because the poll made so many undecided people make up their mind on the spot — and make it more artificial.
Some other details from the poll:
In the field between June 12 and June 18, 625 randomly selected adults, including 527 likely voters. Sample error is +/- 3.9 percent for all respondents and +/- 4.3 percent for likely voters. The error rate increases to 6.5 in the U.S. House races because of the smaller sample (222 in CD2 and 227 in CD1).
Live interviews with landline and cellphone only households: Voters on land line only (9 percent), mixed (50 percent) and cell only (42 percent).
The median length of the interview was 11 minutes.