On the eve of Election Day in Alabama, SurveyMonkey’s ongoing tracking in that state continues to show widely varying estimates of the outcome depending on the methods used to model or select the likely electorate.
On Friday, we published findings of interviews conducted over the previous week, with alternative estimates showing comfortable leads for either Republican Roy Moore or Democrat Doug Jones. An update based on those same data plus interviews conducted since (Thursday November 30 through Monday morning, December 11) yields very similar results. Alternative models of the potential likely electorate yield everything from an 11-point lead for Jones lead to an 8-point lead for Moore.
These models fall into two categories. Two are based on past turnout as reported by our respondents (the two columns on the right in the above table). Two others do not. The models based on past turnout yield demographic profiles that are more white and male than recent off-year elections, as measured by U.S. Census surveys. In contrast, ours shows Democrats -- who are more likely to be female and African-American -- reporting that they are following the election more closely and are more likely to vote than are Republicans. That combination argues for being more skeptical about the models that incorporate self-reported past turnout.
Let’s take a closer look. As reported this past Friday, our tracking survey shows that Alabama Democrats are far more unified in support of their candidate and fired up about voting than Republicans, a pattern that is consistent with actual turnout in other elections elsewhere nationwide this year.
Thus, the polls are telling us that turnout in Alabama is likely to advantage the Democrats next week. What we cannot know or calibrate with precision is whether that advantage will be large enough for a Jones win.
In response to Friday’s report, readers have asked for a judgment about which of the various turnout models tested is most accurate or otherwise preferred. While the unique nature of Alabama’s special election makes it hard to apply lessons from polling elsewhere, we can check the demographic composition produced by our various models against past elections using the demographic profiles from the Census Current Population Survey.
The CPS provides one of the best measures of the demographics of past elections. It is a very high-quality, random sample survey (with a response rate near 90 percent), that includes questions in November of even-numbered years asking Americans whether they voted.
More specifically, in Alabama, the CPS shows that voters in off-year general elections since 2006 have been 52 to 56 percent female, 24 to 27 percent black and 18 to 20 percent age 35 or younger. In higher turnout presidential elections, the electorates are younger, though in 2010 and 2014 the racial composition was very similar to the 2012 and 2016 presidential elections.
If we look at the alternative models applied to our Alabama survey, the two based on past self-reported turnout yield electorates that are slightly less black (22 to 23 percent) and less female (48 to 51 percent) than the recent off-year elections.
Given the indication in our survey that turnout is likely to favor Democrats more than usual, we would expect, if anything, to see the opposite demographic pattern. While it is very much a judgment call, these data argue for discounting the two past-vote models in the Alabama election in this case, and focusing more on the models based on self-reported intent to vote, both of which are more favorable to Jones.
Methodology: SurveyMonkey’s poll on the Alabama Senate election was conducted online November 30 through December 11, 2017 among a national sample of 2,203 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. The modeled error estimate for this survey is plus or minus 4.5 percentage points.