It’s one of the most closely watched Senate campaigns in years — and one of the most confounding. The battle in Alabama between Roy Moore and Doug Jones has pollsters thoroughly flummoxed.
Other recent polls include a Washington Post survey showing Jones ahead by 3 points, a CBS News/YouGov survey showing Moore up by 6 and a Gravis poll putting Moore’s lead at 4 points.
Here’s a closer look at what’s going on.
Why has it been so hard to poll the race?
The central problem is predicting who will vote.
For polling organizations, calling potential voters and counting their responses is often the easiest part of the job. The real challenge is to predict who is actually going to show up to cast a ballot.
In the last two off-year elections in Alabama, somewhere between 40% and 50% of registered voters cast ballots. The Alabama secretary of state has estimated that this year’s special election will draw a smaller turnout, more on the order of 25%.
As the electorate shrinks, it usually doesn’t do so evenly. One side or the other usually benefits. Pollsters need to adjust their results to take into account who is most likely to turn out. Polling organizations have a variety of techniques to try to forecast what the electorate will look like, and right now, those different approaches are giving very different answers.
How much difference can a turnout model make?
One approach simply asks people how sure they are that they’ll vote. When SurveyMonkey did that, counting as a likely voter those who said they were certain to vote or probably would, they found Jones had a 9 point lead over Moore.
Another approach looks at people’s vote history: Rather than ask them about their future behavior, ask them about the recent past. Did they vote in the last couple of elections?
When SurveyMonkey defined as likely voters those who reported that they had voted in 2014 (plus those who were too young in 2014 but said they were certain to vote this time), Moore led by 3 points.
Testing other approaches, the pollsters came up with results ranging from an 8 point Jones lead to a 9 point lead for Moore — all using the same underlying survey data.
Monday afternoon, Monmouth University released a survey showing three different turnout models. Under a standard midterm turnout similar to 2014, Moore would lead, 48%-44%. A higher turnout, more similar to a presidential year, would yield a lead for Jones. And a turnout model based on the midterm election last month in Virginia showed the two candidates tied, with 6% still undecided, they found.
When combined with the potential for purely random error inherent in all surveys, the findings make a projection of the outcome virtually impossible.
Mark Blumenthal, SurveyMonkey
Why is the variation in results bigger than usual in this election?
Most of the time, pollsters have a pretty good idea of who will show up: Elections tend to follow predictable patterns.
Those patterns don’t exist in this case. It’s a special election, with nothing else on the ballot in much of the state, being held in December, which is not a typical election time. So past voter history isn’t as useful as usual.
That alone would greatly complicate predicting who will vote. On top of that, Moore is a polarizing candidate with some deeply committed followers, but also many Republicans who disapprove of him. And the multiple accusations by women who say he pursued them when they were teenagers add a whole additional layer of complexity.
Are there other significant differences between the polls?
The two most recent polls conducted by telephone with live interviewers, who called both cellphones and landlines, were the ones done by Fox and the Washington Post. Both showed Jones ahead.
By contrast, automated polls, which by law cannot call cell phones, have tended to yield results more favorable to Moore. Pollsters who use automated techniques, which are much cheaper than live-interviewer polls, have a variety of ways to adjust for the limitations of their methodology, but it’s possible that this time around, they are missing a significant part of the electorate.
The Fox poll found, for example, that Jones led by 30 points among people reached on cellphones, who tend to be younger than people who usually use a landline.
Could there be a ‘hidden vote’ for one side or the other?
Perhaps some people don’t want to admit to an interviewer that they plan to vote for Moore, given the air of scandal surrounding him. In that case, the automated polls could be more accurate in showing Moore leading.
On the other hand, Moore’s supporters don’t seem shy about stating their position, as reporters on the ground in Alabama have found. And it’s also possible that some longtime Republicans don’t want to admit to a pollster that they’re planning to vote for a Democrat.
If the result is so uncertain, why bother with polling?
Everyone wants to know who’s going to win, but forecasting the outcome isn’t the only reason — or even the main reason — for polling.
Polls also tell us a lot about which voters are backing which candidate — younger Alabamians, especially younger women, favor Jones by a big margin, for example. Repeated polls can show how a race develops over time.
And polls can also tell us a lot about which issues have had an effect on voter opinions. As a result, they serve as a check on the self-serving spin that voters inevitably will hear starting Tuesday night, once the actual results are counted.