WASHINGTON (AP) — Chances are, you have never been contacted for an election poll. But the dozens of high-quality election polls that will be released before Election Day, Nov. 5, represent a reasonable estimate of the opinions of all Americans.
The best pollsters do that by ensuring they can randomly select the group of people who respond. That means each household in the United States has an equal chance of being included. Pollsters cannot reach every single household or even come close, so they assemble a group of people with the same range of political affiliations, ages, genders, educational backgrounds and locations as Americans overall.
In other words: You may not have been contacted to participate in the latest poll by the Associated Press-NORC Center for Public Affairs Research at the University of Chicago, but someone else who shares your background and outlook likely was.
High-quality pollsters select people randomly to take surveys
It is the concept of random selection that allows a relatively small group of survey participants to represent the country as a whole.
Top-quality pollsters often start with lists of possible home addresses or telephone numbers, and then people are randomly selected from within that group. This is the kind of method that the AP uses in its polls conducted through the AP-NORC Center.
Some pollsters use a different technique, where anyone who wants to participate in their panel can join it. But with that approach, there is less certainty that the group of people responding to any given poll — a “sample,” pollsters call it — is randomly representative of a broader population.
If the initial sample does not look like the country as a whole, some views could be overrepresented or underrepresented, making it harder to accurately capture the attitudes of the entire U.S. population.
An individual’s chance of being selected to participate is low
Polls conducted by the AP through the AP-NORC Center use the AmeriSpeak panel, where households across the U.S. are randomly selected for the sample and then contacted to tell them about the panel. If the household agrees to participate, people complete an introduction survey that collects basic information and participate in polls between two times to three times each month.
For this kind of poll, the odds of being randomly selected to participate are extremely low. There are about 130 million households in the U.S., so to start with, each individual household has only the tiniest chance of being chosen. Even once a household has been selected to participate, there is a relatively small chance of being selected for the surveys that are conducted by media organizations such as the AP-NORC poll.
Pollsters make adjustments to make sure they’re reflecting the population as a whole
It is not a perfect system. Some groups are harder to reach or are less inclined to take surveys, such as nonwhite adults or people without a college education.
To correct for that, pollsters magnify the responses of people who are part of those underrepresented groups to make sure the population percentages in the survey reflect the overall population and they lower the impact of people who are part of groups that are more likely to take surveys.
This process is called “ weighting.” The goal is to make some responses count for more if their demographic characteristics are underrepresented in a survey and some count for less if people like them are overrepresented. To figure out which participants should get more weight and which should get less, pollsters use findings from the most accurate surveys out there, such as ones by the Census Bureau, to get a baseline for what the U.S. population actually looks like.
Even this extra step cannot ensure that the group of people who are being surveyed is fully representative. That is why all high-quality pollsters will tell you about the margin of sampling error, which helps you understand how much the response could vary.
Pollsters do not talk to every single person in the country, so the results have some amount of error. The margin of error is a reminder that each finding is not exactly precise. It also is a guide for understanding how big the range of responses could be.
