Recent punditry on Alabama’s upcoming special election for U.S. Senate has a distinct theme: The race between Republican Roy Moore and Democrat Doug Jones is “impossible to poll” because “nobody has any clue what turnout is going to look like.” That difficulty, combined with the controversies surrounding Moore, put Jones “just a normal polling error way from a win” in heavily-Republican Alabama, though “it’s all but impossible” to predict the winner.
Data collected by SurveyMonkey confirms these arguments. Minor differences in the methods used to model or select the likely electorate produce wildly varying estimates in Alabama. Data collected over the past week, with different models applied, show everything between an 8 eight percentage point margin favoring Jones and a 9 percentage point margin favoring Moore.
The same survey data also reveal the underlying tensions behind the volatile results: Alabama Democrats are angry and energized, while a significant but critical minority of Republicans are conflicted between a nominee they dislike and a President they support.
“Special elections are extremely hard to poll,” political scientist Jonathan Bernstein recently tweeted, “even when they’re not as screwy as this one.” This particular election has two especially unique elements: First, it involves a special election in mid-December, so historical turnout data from previous comparable elections does not exist. Second, Republican nominee Moore has been mired for weeks in allegations over his alleged sexual misconduct, prompting prominent Republicans to either throw their support to Jones or, like Alabama Sen. Richard Shelby, pledge to write-in another name. The combination makes the pollster’s biggest challenge – modeling the likely electorate – that much tougher.
Modeling Likely Voters
SurveyMonkey has been polling around the 2017 elections in Virginia, New Jersey and now Alabama to help refine our techniques for representing likely electorates. Though much of the data has not yet been published, in November we described how the Democrats’ lead in Virginia varied based on different methods of identifying likely voters in data gathered in partnership with The Washington Post.
In Alabama, the differences are far bigger.
Among all Alabama registered voters interviewed in the past week and weighted using our standard methods to match to U.S. Census estimates of registered voter demographics, we find Democrat Jones leading Republican Moore by eight percentage points (53 to 45 percent). That margin is similar to the findings among all registered voters in two recent live-interviewer telephone polls conducted by Fox News and Washington Post/Schar School.
We know the actual electorate will be a fraction of all registered voters – roughly 40 to 50 percent of Alabama’s registered voters participated in the last two off-year general elections, in 2010 and 2014 – we just don’t know how small that fraction will be or who exactly will turn out.
Yet applying several commonly used techniques to narrow the electorate in Alabama produces very different results. The five techniques we used are discussed below:
Jones leads by wide margins when we simply ask voters if they plan to vote: Using our standard weighting methods, he leads by 9 points (54 to 45 percent) among those who say they are certain or probable to vote and by six points (53 to 47 percent) among the smaller subset of those who say they are absolutely certain to vote.
An alternative, which favors Moore, defines likely voters as those with a self-reported history of voting in lower turnout elections. For example, among those who say they voted in 2014 (plus younger voters, age 18–20, who were not eligible in 2014 but say they are absolutely certain to vote this year), Moore leads by 3 points (50 to 47 percent).
As in Virginia, we also applied a more sophisticated model based on self-reported past voting, that included all voters but weighted down voters who say they vote less often in non-presidential elections. That model gives Jones a very slight edge, 50 to 47 percent.
When we examined the data weighted using standard methods, however, we noted an anomaly absent from SurveyMonkey polling conducted in October in Virginia and New Jersey. The Alabama registered voters who reported voting in 2016 favored Donald Trump over Hillary Clinton by a 50 to 39 percentage point margin. Trump’s actual margin was significantly larger (62 to 34 percent).
As a result, we also tabulated alternative estimates where we also weighted by 2016 self-reported vote preference. SurveyMonkey has not previously released election data weighted by past vote, but other pollsters, including the CBS/YouGov survey in Alabama, do incorporate self-reported 2016 vote in their weighting.
Including 2016 vote in the weighting narrows Jones’ advantage among all registered voters to just 2 percentage points (49 to 47 percent). It results in an exact tie (49 to 49 percent) among both subgroups of self-reported likely voters, and gives Moore a 10-point lead (53 to 43 percent) when narrowing to those who report having voted in 2014 and an 8-point lead (52 to 44 percent) or when using the model that includes all registered voters but gives less weight to those who say they vote less often.
Thus, different approaches to estimating the likely electorate in Alabama produce widely varying results. The 2016 past vote anomaly is especially puzzling. It may indicate either a big underlying Democratic turnout advantage seeping into the response patterns of our respondents, or it may be a sign of sample bias in need of correction. Either way, when combined with the potential for purely random error inherent in all surveys, the findings make a projection of the outcome virtually impossible.
Responses to questions beyond the ballot help illustrate why turnout projections are difficult and why voter preferences may be quite erratic (the data cited in the remainder of the report are based on results among all registered voters weighted by demographics and self-reported 2016 vote preference).
Democrats Fired Up
First, Democrats are angry about President Trump and Roy Moore, enthusiastic about voting and united in support of their nominee.
Better than nine out of ten Democrats in Alabama disapprove of President Trump (94 percent) and prefer a candidate who takes different positions than Trump (92 percent). Almost as many (81 percent) strongly disapprove of Trump. Their perceptions of Roy Moore – 83 percent unfavorable, 73 percent strongly unfavorable – are nearly as negative. Virtually all (98 percent) support Doug Jones over Roy Moore (1 percent).
They are also more fired-up about the special election – 79 percent of Democrats say they are absolutely certain to vote compared to 67 percent of Republicans. Nearly two-thirds of Democrats (65 percent) say they are following the election very closely compared to 55 percent of Republicans.
These data imply that Democratic turnout will be higher, but Democrats plus Democratic leaners are still a minority in Alabama. Jones does have a significant advantage (55 to 33 percent) among independents who lean to neither party, but a Jones victory also requires that significant numbers of Republicans either vote for him or skip the election.
Some Republicans Conflicted About Moore
Many Alabama Republicans express some discomfort with their nominee. While 60 percent rate Roy Moore favorably, nearly a third (31 percent) rate him unfavorably. Just 28 percent of Republicans give Moore a very favorable rating, less than half of the intensely favorable rating Democrats give Doug Jones.
Still, 81 percent of Republicans support Moore, while 15 percent say they are ready to vote for Jones. Support for Jones is slightly greater (17 percent) among independents who lean Republican. While a modest fraction of Republicans, these defectors to Jones – who are nearly two-thirds female and nearly half aged 44 or younger – are the source of his overall edge among all registered voters. Without their support, Jones would trail in every likely voter scenario.
The survey also shows considerable uncertainty about these vote preferences. Better than a third (38 percent) of the Republicans for Jones say they could still change their minds. A smaller but significant chunk (17 percent) of Republicans supporting Moore also say they could still change their minds..
These data tell us both about potentially shifting preferences and the chance that many conflicted Republicans may skip the election or choose to write in a third name.
Republicans Not Conflicted About Trump
While some Alabama Republicans feel conflicted about Roy Moore, very few express reservations about Donald Trump. Nearly nine out of ten approve of Trump’s performance (88 percent) and want a U.S. Senator who will take positions similar to Trump (90 percent).
Perhaps more relevant: Better than half (61 percent) of Republicans and Republican leaners who support Doug Jones also approve of the job Trump is doing as president, and 65 percent want a U.S. Senator who takes positions like Trump. These data suggests the potential power of the president’s recent endorsement of Moore.
Thus, while the data do not lend themselves to an easy prediction of a winner, they do tell us quite a bit about why the race has been so close and about the conflicting sentiments confronting many Alabama Republicans.
Full topline results and a detailed demographic breakdown from this survey can be viewed here.
Methodology: SurveyMonkey’s poll on the Alabama Senate election was conducted online November 30 through December 7, 2017 among a national sample of 1,559 registered voters living in Alabama. Respondents for this survey were selected from the nearly 3 million people who take surveys on the SurveyMonkey platform each day (more details on our methodology here). The data have been weighted for age, race, sex, education, and geography using the Census Bureau’s Current Population Survey to reflect the demographic composition of registered voters in Alabama. Some results were also weighted based on 2016 presidential vote as described in the article. The modeled error estimate for this survey is plus or minus 5.5 percentage points.