The refrain from Donald Trump backers of late is that the polls are systematically underrepresenting them, thus making it seem like the real estate mogul is behind Hillary Clinton when he is in fact ahead. What to make of this?
For starters, the idea that pollsters are intentionally overlooking Trump voters—a claim Rush Limbaugh made this week—is nonsense on stilts. In general, pollsters have an incentive to get the correct result. While there are few hucksters out there, a number of pollsters—Gallup, Pew, Hart/McInturff, Langer Research—have well-earned reputations for fairness and accuracy.
Similarly, the idea that Trump voters are being overlooked obscures the basic mathematics behind polling. Anybody who has taken a college course on statistics knows that, when you conduct even a small random sample, you can draw inferences about the broader population.
Still, there are two ways that the polls could be underestimating Trump’s standing. First, most polls are still of registered voters. Most registered voters ultimately vote in the presidential election, but many do not. If those non-voters are less disposed to Trump than actual voters, the polls right now may be understating his position. If this is a problem, it should begin to correct itself after Labor Day, when pollsters begin using likely voter screens.
Second, there could still be statistical bias in the polls. The challenge with polling is that it is taking a sample of voters, a population that does not actually exist until Election Day. Pollsters have to make guesses about what that population will look like. While well-informed, those guesses can be wrong.
This has happened several times over the last couple cycles. The final 2014 midterm polls made the race in key states out to be much closer than it actually was. Senate battles in states like Georgia, Iowa, and Kentucky looked to be nail-biters, but Republicans won them easily. On the other hand, the Democrats overperformed the polls in 2010—for instance, in California and Nevada. The 2012 presidential polls were especially strange. The national polls had the race a virtual dead heat for the final month (and the well-regarded Gallup poll showed Mitt Romney ahead the whole month), but the state polls showed President Obama beating Romney fairly comfortably. The state polls turned out to be correct.
Importantly, these errors have underestimated both Democrats and Republicans over the last couple cycles. It is just not right to say that the polls have a pro-Democratic or pro-Republican tilt. They can be off by a few points in either direction. So—from our vantage point, here in late August—it is just as possible that the polls are understating Hillary Clinton’s standing as they are Donald Trump’s.
One thing is for certain: Crowd sizes, yard signs, and other indications of intensity are irrelevant. Enthusiasm only matters in an election insofar as somebody must be sufficiently interested to come out to vote. After that, the vote of the most diehard supporter counts just as much as the vote of somebody who flipped a coin.
It’s also important to keep in mind the perspective of scale. Barack Obama’s rallies filled whole stadiums in 2008, but he still won a smaller share of the vote than George H.W. Bush in 1988 and even Warren G. Harding in 1920. Approximately 130 million Americans will participate in the November election—a number that dwarfs even the largest political rally, a thousand of times over. Using crowd sizes to gauge general election support is like using a twelve-inch ruler to measure the height of the Empire State building; it’s a bad tool.
