Why Political Predictions Still Contain So Much Uncertainty

In recent weeks, two things have become increasingly clear: Hillary Clinton is beating Donald Trump in the vast majority of scientific national polls, and the reliability of those polls has become more contested than ever before.

Republican nominee Trump has led the way, of course, by charging that Democrats are pushing “phony polls” and that the press has refused to report on the truth. Neither is true. Throughout October, a range of independent public polls has consistently shown Clinton in the lead.

But that doesn’t mean the polls show a certain victory for Clinton. There are many reasons for this uncertainty. Voters can still change their mind. Pollsters can mispredict turnout. The polls themselves have statistical margins of error. And recent technological changes have made polling both more challenging and more experimental.

In a traditional survey, pollsters need to tap a random sample of landline phones. But Americans have become far less likely over the decades to volunteer to participate in polls when called, and more and more Americans are not even reachable by landline. That has forced pollsters to include cell phones in their samples, which cannot be dialed by computers under U.S. law.

At the same time, the Internet has allowed pollsters to measure public opinion quickly and inexpensively. Rather than randomly compile a sample for every new poll, companies like YouGov, Morning Consult and SurveyMonkey create a much larger community of regular poll takers online. The pollsters then sample demographic slices of that community to estimate the electorate.

To date, the two methods have been returning similar results, and past elections have shown each to be decent predictors of the election outcome. But the old saw remains: The only poll that matters is the one taken at the ballot box.

How polling works

A lot rides on the polls. They steer media coverage, help candidates strategize, and determine who participates in the debates. Phone polling is a proven but costly method. Online polls are cheaper but typically do not use a random sample of participants.

TELEPHONE

MAKE RANDOM CALLS

Pollsters dial thousands of random phone numbers, often pulled from directories of likely voters. Robo-dialers call landlines, but cell-phone numbers are dialed manually–a costly process.

INTERVIEW RESPONDENTS

Once the pollsters connect with someone who’s willing to talk (most are not), they read the survey questions and take down the responses.

ONLINE

SOLICIT PARTICIPANTS

Pollsters use web ads and other tactics to recruit millions of people to take online surveys.

PANEL A

PANEL C

PANEL B

SEND THEM SURVEYS

These participants are grouped into panels, or pools. Pollsters send surveys to a sample of the panel that is a proxy for the adult population.

COMPUTE RESULTS

Responses are adjusted based on Census data and past voter turnout to reflect the voter population.