National polls show wide variations. Some, such as the Bloomberg poll, have Democratic nominee Hillary Clinton up by 9 percentage points over Republican Donald Trump. Others, such as the Rasmussen poll, show the candidates essentially tied. Are the pollsters sampling different populations, or is there a simpler explanation?
One explanation for some of the differences in results is the weighting of Republicans and Democrats who are expected to turn out to vote. The pollsters pick the percentages of Republicans and Democrats who they expect to vote, and these percentages affect the poll results.
Pollsters do not simply take a random sample of individuals and then present the unweighted results of how they plan to vote. In order to obtain a more accurate result, pollsters weight survey responses by a number of factors, including party affiliation.
Most pollsters assume that more Democrats than Republicans will turn out to vote because there are more Democrats than Republicans registered, and because more Democrats voted in 2008 and 2012. Of course, no one knows how many voters of each party will show up for this election, and all polls depend critically on assumptions about turnout by party.
Although more Democrats voted in 2012 and 2008, the numbers were equal in 2004. During the 2016 primaries, Republicans saw a 35 percent relative increase in turnout over 2008, as record numbers turned out to vote for Donald Trump. In contrast, relative turnout fell 26 percent for the Democrats.
Consider a few recent polls.
In a Fox News poll, conducted from October 10 to 12, Clinton leads Trump by 7 percentage points. The poll assumes that 45 percent of voters will be Democrats, and 36 percent will be Republicans, a difference of 9 percentage points. Twenty percent of voters are assumed to be unaffiliated.
In a recent Quinnipac poll, where Clinton leads by 7 percentage points, and the Selzer/Bloomberg poll, where she leads by 9 percentage points, it is assumed that Democratic voters top Republicans by 6 percentage points.
In a recent NBC-Wall Street Journal poll, Clinton leads by 11 percentage points. The share of Democrats voting was assumed to top Republicans by 9 points.
In contrast, consider the Rasmussen poll, taken October 19, where the candidates are tied. That poll assumes that 37 percent of voters will be Democrats and 33 percent will be Republicans, a difference of 4 percentage points. Thirty percent are unaffiliated.
Naturally, the larger the share of Democrats assumed to vote, the greater the win for Clinton. Rasmussen has the candidates equal because its weighted share of Democrats is lower than the other polls.
The big question is how many more Democrats will turn out on Election Day. A 9 percentage point difference in Democrats’ favor is very different from a 4 percentage point difference.
This reweighting can lead to a self-fulfilling prophecy, as larger numbers for one candidate discourage supporters of the other candidate from showing up to the polls. The message often is that the election is over before it has begun, so there is little point in voting.
Weighting by party affiliation is not the only mystery in polls. The following questions also contribute to the outcome. Are the weighted samples of voters representative of the electorate? Is the rate at which people respond controlled in some way? Do people accurately self-identify their demographic and political affiliation—or do some pretend to be of another group or party? Do people accurately reveal their preferences?
Trying to gain a glimpse of the outcome on November 8, dozens of polling organizations are canvassing the American electorate. Polling results can and do vary widely. Although pollsters often are good at predicting outcomes, there have some recent notorious failures, such as in several state primary elections earlier this year, as well as the Brexit vote in Britain and the cease-fire referendum in Colombia.
The best way to interpret polls is not as statements of certain outcomes but rather as informed views about future outcomes. Individuals with informed views are often right, but they are sometimes wrong. So too with pollsters.
Diana Furchtgott-Roth, former chief economist of the U.S. Department of Labor, is an unpaid advisor to Donald Trump.
Interested in real economic insights? Want to stay ahead of the competition? Each weekday morning, E21 delivers a short email that includes E21 exclusive commentaries and the latest market news and updates from Washington. Sign up for the E21 Morning Ebrief.