解释：路透社/益普索美国民意调查是如何开展的？

Explainer: How are Reuters/Ipsos US public opinion polls conducted?

2026-08-07T14:41:25.112Z / https://www.reuters.com/world/us/how-are-reutersipsos-us-public-opinion-polls-conducted-2026-08-07/

Summary

• Polls collect respondents’ views at a particular moment in time
• All polls are subject to margin of error
• Congressional elections present unique challenges

WASHINGTON, Aug 7 (Reuters) – Reuters/Ipsos conducts polls of national U.S. public opinion on a variety of political, business and social topics including a regular reading on the approval of the president and on Americans’ views of which party they want ​to run Congress.

WHAT DO PUBLIC OPINION POLLS MEASURE?

Public opinion polls capture respondents’ views at a particular moment in time. By reaching out to groups ‌of people designed to represent the broader U.S. population, they provide a snapshot of national opinion.

The Reuters Daily Briefing newsletter provides all the news you need to start your day. Sign up here.

HOW ARE THE REUTERS/IPSOS POLLS CONDUCTED?

Ipsos uses its proprietary “KnowledgePanel,” a representative sample of Americans aged 18 and over.

Participants are selected through a postal address-based sampling method that includes all U.S. households. They are polled online. Respondents who do not already have internet access are provided with internet service and a laptop at no ​cost.

The data is weighted to reflect U.S. Census data on how the broader U.S. population breaks down by factors including gender, age, race, Hispanic ethnicity (people of ​Hispanic descent are the nation’s second largest ethnic group), education, household income and census region. Data may also be weighted to account ⁠for political composition using partisan self-identification (benchmarked to the Pew Research NPORS study) or self-reported 2024 presidential vote choice (benchmarked to election results).

HOW MANY PEOPLE ARE POLLED?

Six times a year, ​Reuters/Ipsos conducts a poll of at least 4,000 respondents over roughly a week. These polls have a margin of error of about 2 percentage points for the full sample, with ​higher margins of error for smaller sub-groups of respondents.

Over the course of the year, Reuters/Ipsos conducts at least 24 polls of presidential approval and other topics. Those polls reach at least 1,000 people over a few days and typically have a margin of error of about 3 percentage points for the full sample.

WHAT IS A MARGIN OF ERROR?

Margin of error is a statistical term intended to express ​the possible variation between a poll’s finding and what the result would be if the full population were polled. If a representative poll of the U.S. population finds that ​75% of respondents prefer ice cream to Brussels sprouts and has a 3% margin of error, that means that in the full population somewhere between 72% and 78% of Americans hold that ‌preference.

WHAT IS ⁠A ‘GENERIC BALLOT’ POLL?

Generic ballot polls are national polls that provide an overall view of which party poll respondents want to see in charge in Congress. While these polls provide a view of national public sentiment, they do not capture all the nuances of the state-by-state and district-by-district contests that determine control of the U.S. Senate and House of Representatives.

House elections in particular are heavily influenced by partisan redistricting, also known as gerrymandering, which has resulted in contests where just about three dozen of the 435 House seats are expected to be competitive ​in the November midterm elections.

Senators serve for overlapping ​six-year terms, which mean that just ⁠a third of the seats in that chamber are up for grabs in a congressional election. This year, 35 U.S. states will hold elections for Senate, with eight states — Alaska, Georgia, Iowa, Maine, Michigan, North Carolina, New Hampshire and Ohio — seen as most competitive.

WHAT ​ABOUT REGISTERED VOTERS AND LIKELY VOTERS?

Reporting on polls will often refer to the findings from registered voters and likely voters, as ​these are the people ⁠who decide election outcomes. The sample of registered voters represents the widest group of people eligible to cast ballots, and excludes responses from people who aren’t eligible because they aren’t citizens or have criminal records. Just over 60% of eligible voters cast ballots in 2024, according to the University of Florida’s Election Lab.

Pollsters try to measure the views of those who will ⁠cast ballots by ​using models of likely voters, the people who seem most likely to show up on Election Day. ​These models are by nature subjective, and pollsters take multiple factors into account in determining who to include, including respondents’ self-reported level of enthusiasm about the upcoming election and past voting behavior.

Like any attempt to predict ​future behavior, those models can have flaws that cause them to overestimate or underestimate which groups of voters do cast ballots in the end.

Writing by Scott Malone; Editing by Alistair Bell