Allow Michael McDonald to say what everyone has been thinking about the waves of polls on the Alabama Senate race.
“When I look at the polling in Alabama, it’s just all over the place,” said McDonald, an associate professor at the University of Florida who studies American elections.
And so why is that?
Why is it that we have arrived at election day with absolutely no reason to be surprised at the outcome Tuesday night between Democrat Doug Jones and Republican Roy Moore?
Yes, the allegations of sexual misconduct against Moore have thrown the race into chaos – an almost certain Republican victory in perhaps the nation’s most Republican state was maybe all but conceded.
But in the 33 days since those allegations – repeatedly denied by Moore – were first reported by The Washington Post, polling on the race has veered to whatever extreme you could imagine.
And then, the day before the election, a fitting end to the polling unpredictability: Five polls were released – two with Moore leading, two with Jones leading and the tiebreaking poll was, appropriately enough, a tie.
What in the Republican red/Democratic donkey is going on?
“A lot of times, you just see the score 49-46, 50-43 and you think of it as a basketball game or a football game,” said Spencer Kimball, the advisor for the Emerson College Polling Society that has conducted numerous polls during the Senate campaign. “If somebody is winning 14-7, they are winning by seven points. But in surveys, that’s not the case.”
As AL.com interviewed polling and political experts over the past week, two clear themes emerged about polls: Defining the electorate and a poll’s margin of error.
Those are the details that go beyond the headline of one candidate leading another candidate by, say, 5 points.
“Polls can be accurate or inaccurate based on their sample composition,” said John Couvillon, the pollster for JMC Analytics and Polling – another frequent polling firm in the Alabama race.
And therein lies the predictive aspect of polling – not forecasting which candidate will win or lose but accurately anticipating who will be going to the polls.
Pollsters said they work to create a polling sample base that will show up at the polls on election day. And that’s the tricky part.
How many black voters? How many white voters? How many female voters? How many male voters? How many Democrats? How many Republicans? How many rural voters? How many urban voters?
“So in your poll if it’s 38 percent Republican, 31 percent Democrat and 28 percent independents, that’s going to create different results than if you have 50 percent Republican sample, 30 percent Democrat, 20 percent independent,” Kimball said.
Who will show up for Alabama’s special election?
McDonald said that in presidential elections, there is enough history and enough research that a representative sample pool can be developed that will reflect who will show up on election day.
“But in an Alabama special Senate election, we don’t have a lot of past history,” he said. “We don’t know what the electorate is supposed to look like.”
Black voters are expected to be a key demographic in Tuesday’s Senate election – a higher black turnout, conventional wisdom goes, the better for Democrat Jones.
“For a special election like this, my thought is it’s going to be 23-24 (percent) will be the black composition of those who show up,” Couvillon said. “Two or 3 percent matters in a close race like this because blacks vote almost unanimously Democrat.”
In polls released Monday, even that demographic varied widely. Change Research had 27 percent of its sample pool composed of blacks while Emerson College had 17 percent blacks. But perhaps further muddying the polling water, Moore led both polls – although his lead was larger in the Emerson poll with smaller sample of anticipated Jones-aligned blacks.
A Monmouth University poll that had the race tied on Monday had 28 percent blacks in its sample pool.
In the Fox News poll that had Jones leading by 10 points Monday, the percentage of blacks in the sample pool wasn’t clear.
Couvillon pointed to a recent Washington Post poll that didn’t make clear the breakdown of the sample pool’s demographics beyond party affiliation – which he described as a “red flag to me.” That poll had Jones leading by 3 points.
“I did some calculations based on the data that was available and it showed something like a 28 percent black electorate,” Couvillon said. “I would not believe that for a millisecond for a special election. So in other words, the sample composition was tilting Democratic and therefore, in my opinion, was not a representative sample.”
The predictive aspect of polling is not the winner or the loser but who will show up at the polls.
“In special elections, the biggest dilemma you face is estimating who will actually go vote,” said Jess Brown, a retired political science professor from Athens State University and a close observer of Alabama politics. “Estimating who will vote and who won’t can be a real problem.”
Sometimes, pollsters have to make “their best guess,” McDonald said.
“I don’t think intentionally – yeah, there’s probably some shady pollsters out there putting their thumbs on the scale – but for the most part, they’re honestly trying to do their best in tough circumstances,” McDonald said. “So they may make a mistake when guessing as to what the electorate looks like.”
The matter of margin of error
Once pollsters have defined their reflective sample pool and collected the data, the “basketball score” – as Kimball put it – emerges for all to see.
Except it’s often not accurately seen, Kimball added.
“I think the most important thing to know is what the margin of error is,” he said.
It’s that phrase associated with every poll – the mathematical leeway that may, to some, add a little fuzzy to what may appear to be crystal clear numbers.
Say Moore is leading Jones by 50-45 in a poll that has a margin of error of 3 percent. Applying that margin of error could lead to Jones leading by 1 point – Moore’s support could be as low as 47 percent and Jones’ support could be as high as 48 percent.
And suddenly, the apparent leader with a comfortable advantage is now a narrow loser.
“Any time a campaign or an election is within three to five points, a survey is not going to be able to statistically predict who is going to win because their margins of error are going to be too large,” Kimball said. “I think a lot of people don’t apply that margin of error.”
Considering the margin of error perhaps makes the results of the polls released Monday all the more remarkable. Previously, polling between Jones and Moore was essentially a toss-up when factoring in the margin of error. Five December polls have the race within the margin of error – statistically, a tie.
“You shouldn’t look at a polling number and say you know with a certainty who is going to win this election,” Brown said. “No, you don’t. A poll, even well done, a poll is nothing more than a snapshot of opinion at the time the poll is done. So the only real poll that ought to be accurate as far as election day goes is a well-done exit poll. Actual voters who have already voted. But the press and the public instinctively want to use polls as predictive devices. A poll well done, the closer you get to election day, the more predictive it should be.”
At least two dozen public polls have been conducted on the Senate race since Moore defeated Luther Strange in the GOP primary runoff in September.
And what do we actually know at this point?
“The polls tell us the election is close,” McDonald said. “But beyond that, I don’t really know what to say. It could go either way.”