Can We Still Trust Polls?

Overview

The video argues that while public confidence in polling was shaken after major election surprises—especially the 2016 U.S. election—the overall evidence suggests polls are still broadly reliable when understood correctly.

2016 U.S. Election: Why the “Polling Failure” Narrative Doesn’t Fully Hold

National polls were largely accurate

The speaker notes that national polling in 2016 closely predicted the popular vote, including:

• Polls had Hillary Clinton leading by about ~3 points
• She ultimately won by about ~2 points

The bigger errors are framed not as a systematic failure of polling overall, but as issues concentrated at the state level.

Why 2016 went wrong: state-level polling weaknesses

A commission-style review is cited as identifying key state-polling problems, including:

• Missing late movement among undecided voters toward Donald Trump (late “swing” not captured)
• Under-representation of people without college educations, contributing to unexpectedly large results in the Midwest

Bias Accusations Don’t Consistently Explain Poll Misses

The video addresses claims that pollsters favor one party by looking at historical patterns:

• The direction of polling misses (whether polls lean Republican or Democratic) looks roughly like a coin flip, not a consistent partisan bias.
• It also claims that national presidential polling accuracy has improved or remained strong over time.

Measured Accuracy Remains Strong

Evidence cited includes:

• Average absolute polling errors of about 2–3 percentage points for national presidential polls since 2000
• A cross-national study (covering 45 countries over 75 years) suggesting polling accuracy has a good overall track record, with no broad decline despite some recent misses

Brexit as an Example of “Misses,” Not Total Failure

Brexit is used to illustrate that polling inaccuracies don’t automatically mean total breakdown:

• Betting markets leaned strongly to Remain, but polls showed a close race for much of the final month
• “Leave” was often ahead
• This is presented as a challenge caused by uncertainty, not a catastrophically wrong polling system

Key Concept: Polls Are Hard Because Voting Behavior Is Future-Focused

The video emphasizes that election polls must predict two things about future behavior:

1. Who will actually vote
2. How they will vote

Because respondents are being asked about future actions, behavioral prediction is inherently difficult.

Polls Are Especially Valuable for Tracking Attitudes and Trends

Even if polls struggle more in very close elections, they can reliably map shifts in public opinion over time.

Same-sex marriage polling is highlighted as an example:

• Different organizations asked different questions
• Yet they showed the same general trend:
  • opposition in the 1990s/early 2000s
  • steady gains
  • majority support later
• The polling trend is said to align with real-world evidence, including the timing of statewide referenda successes

Benchmarking and Internal Validation Support Accuracy

The video mentions benchmarking, comparing regularly conducted polls to large, federally funded, high-response surveys using the same questions.

It cites close alignment on measures such as:

• Party identification (Democrat/Republican shares)
• Religious affiliation (Protestant/Catholic/other/no affiliation)

The implication is that accuracy can remain stable even when response rates decline.

Bottom Line

The conclusion is that polls should be trusted with realism about precision limits:

• Pew Research Center is presented as valuing polling as a democratic tool
  • giving citizens a voice
  • serving as a check on politicians’ claims about public opinion
• Polls can be trusted so long as expectations match their real precision
• They should not be treated as unquestionable determinants of election outcomes

Presenters / Contributors

• Will Jennings
• Christopher Wlezien
• Pew Research Center (speaker/organization referenced; no individual name provided in the subtitles)