Polling statistics at a glance
Polling statistics are easiest to understand when you place response rates, coverage, and accuracy next to each other. The numbers below show how survey participation changed over time, how polling methods shifted, and where the biggest accuracy gaps appeared in major election and issue polling work.
Key takeaways
- Telephone survey response rates fell from 36% in 1997 to 9% in 2012 (Pew 2012 representativeness study).
- Standard survey household reach fell from 90% in 1997 to 62% in 2012 (Pew 2012 section 1).
- Among contacted households in 2012, 14% yielded a completed interview (Pew 2012 section 1).
- AAPOR found more than 60% of all pre-election polls in 2020 were done online (AAPOR 2020 pre-election polling report).
- Pew found the average 2022 topic estimate differed from government benchmarks by 4 percentage points across 26 topics (Pew 2022 polling accuracy).
- The 2020 national-popular-vote polling error was the largest in 40 years (AAPOR executive summary).
Fast facts
- The public survey environment changed from mostly live-phone work to a mix that includes online polls, IVR, texting, and probability panels (Pew 2023 polling landscape).
- Pew found 61% of national pollsters used different methods in 2022 than they did in 2016 (Pew 2023 polling landscape).
- AAPOR found 64% of national presidential polls were conducted online in 2020, up from 38% in 2016 (AAPOR 2020 pre-election polling report).
- Pew reported that live interviewer phone polls are now a minority share of U.S. polling, while online polls and IVR polls together are more common (Pew 2017 low response rates).
Polling response rates over time
Polling statistics often start with the easiest number to misunderstand: response rate. A response rate is not the same thing as accuracy, but the Pew figures in the dataset show why pollsters pay attention to it. When fewer people answer, survey teams have to work harder to reach a usable sample, and the interview pool becomes harder to manage.
Response-rate trend table
| Year | Typical telephone survey response rate | Household reach in the standard survey | Completed-interview share among contacted households |
|---|---|---|---|
| 1997 | 36% (Pew 2012 representativeness study) | 90% (Pew 2012 section 1) | 43% (Pew 2012 section 1) |
| 2003 | 25% (Pew 2012 representativeness study) | 79% (Pew 2012 section 1) | 34% (Pew 2012 section 1) |
| 2009 | 15% (Pew 2012 representativeness study) | 72% (Pew 2012 section 1) | 21% (Pew 2012 section 1) |
| 2012 | 9% (Pew 2012 representativeness study) | 62% (Pew 2012 section 1) | 14% (Pew 2012 section 1) |
The table shows a steep decline on every measure Pew reported. The typical telephone survey response rate fell by 27 percentage points between 1997 and 2012, while the household reach rate dropped by 28 points over the same span (Pew 2012 representativeness study; Pew 2012 section 1).
That matters because response rate and household reach describe different parts of the same problem:
- Response rate shows how many sampled cases actually participate.
- Household reach shows how many households the standard survey can get to at all.
- Completed-interview share shows how much of the contacted pool becomes usable data.
Seen together, those figures explain why modern polling statistics focus so much on method design rather than just raw sample size.
What the decline looked like in practice
Pew’s data show a steady downward path, not a one-time shock.
- 1997: the typical telephone survey response rate was 36%, and the completed-interview share was 43% (Pew 2012 representativeness study; Pew 2012 section 1).
- 2003: the typical telephone survey response rate had already fallen to 25%, and the completed-interview share slipped to 34% (Pew 2012 representativeness study; Pew 2012 section 1).
- 2009: the typical telephone survey response rate was 15%, and the completed-interview share was 21% (Pew 2012 representativeness study; Pew 2012 section 1).
- 2012: the typical telephone survey response rate was 9%, and the completed-interview share reached only 14% (Pew 2012 representativeness study; Pew 2012 section 1).
That pattern is one reason polling statistics are often discussed alongside representativeness, not just cooperation. A survey can still be useful with a lower response rate, but the data show that the contact and completion pipeline got narrower year after year.
Why polling coverage matters
A polling dataset is not only about how many people answer; it is also about who can be reached in the first place. Pew’s standard survey figures show that the household reach rate fell from 90% in 1997 to 62% in 2012 (Pew 2012 section 1).
That decline is important because it means the survey frame itself was changing while polling teams were still trying to estimate the public. The same Pew analysis also reported that among contacted households, the completed-interview share fell from 43% in 1997 to 14% in 2012 (Pew 2012 section 1).
At a glance
- 1997: 90% household reach; 43% completed-interview share (Pew 2012 section 1).
- 2003: 79% household reach; 34% completed-interview share (Pew 2012 section 1).
- 2009: 72% household reach; 21% completed-interview share (Pew 2012 section 1).
- 2012: 62% household reach; 14% completed-interview share (Pew 2012 section 1).
When readers compare polling statistics across years, this is the kind of sequence that matters most. The biggest lesson is not that polling became impossible. It is that the old contact model covered less of the population and turned fewer contacts into completed interviews.
Why it matters for modern polling
Pew’s later analysis makes the operational shift even clearer. In 2017, Pew said live interviewer phone polls were now a minority share of all polling conducted in the U.S., and that online polls and IVR polls were collectively more common than live interviewer phone polls (Pew 2017 low response rates).
That tells a straightforward story:
- Live phone polling stayed important, but it no longer dominated.
- Pollsters diversified methods as response rates fell.
- Coverage and mode choices became part of the accuracy conversation.
Polling methods changed by 2022
Polling statistics from 2022 show a market that looked very different from the one in the 2010s. Pew reported that 61% of national pollsters used different methods in 2022 compared with 2016 (Pew 2023 polling landscape). That is a large share of the polling field changing approach in a single cycle.
Pew also found that 17% of national pollsters used at least three different methods in 2022, up from 2% in 2016 (Pew 2023 polling landscape). The shift suggests more experimentation and more mixed-mode work.
Method-shift snapshot
| Measure | 2016 | 2022 | Source |
|---|---|---|---|
| Pollsters using different methods from 2016 to 2022 | - | 61% | Pew 2023 polling landscape |
| Pollsters using at least three methods | 2% | 17% | Pew 2023 polling landscape |
| Pollsters using texting | 1 pollster | 9 pollsters | Pew 2023 polling landscape |
| Texting share of active national pollsters | - | 13% | Pew 2023 polling landscape |
| Pollsters using probability-based panels | roughly 10% during most of the 2010s | 33% | Pew 2023 polling landscape |
| Pollsters using live phone alone or in combination | - | 32% | Pew 2023 polling landscape |
| Pollsters relying on live phone as the only method | - | 10% | Pew 2023 polling landscape |
| Online opt-in samples matched to registered voter files | 3% to 5% in 2018, 2020, and 2022 | 3% to 5% | Pew 2023 polling landscape |
The table shows that the mix of methods became broader, not narrower. A few useful patterns stand out:
- Texting moved from one pollster in 2020 to nine in 2022 (Pew 2023 polling landscape).
- Probability-based panels reached 33% in 2022, well above the roughly 10% level Pew said was typical for most of the 2010s (Pew 2023 polling landscape).
- Live phone remained in use, but only 10% of national pollsters relied on it as their only method in 2022 (Pew 2023 polling landscape).
Those numbers suggest polling did not simply abandon traditional methods. Instead, many firms layered methods to reach different audiences or to improve sample balance.
Polling accuracy and benchmark error
Low response rates make people worry about accuracy, but the accuracy data in this dataset are more nuanced than a simple “polls are wrong” claim. Pew’s 2022 polling accuracy analysis covered 26 topics and found the topic estimates differed from government benchmarks by an average of 4 percentage points (Pew 2022 polling accuracy). That is a useful broad benchmark because it shows typical issue-poll error without reducing the whole field to one election result.
What the 2022 accuracy analysis found
- Pew surveyed 10,606 U.S. adults for the analysis (Pew 2022 polling accuracy).
- 67% of adults said they had received at least one COVID-19 vaccine dose by June 2021, versus 66% in CDC data (Pew 2022 polling accuracy).
- 15% of adults favored decreasing policing spending in their area, and Pew noted that estimate could range from 11% to 19% and still fit the illustrated margin-of-error logic (Pew 2022 polling accuracy).
- 77% of polled adults said they voted in the 2020 general election, while the actual turnout was 66% (Pew 2022 polling accuracy).
- 53% of adults said they had a retirement account, compared with a matching benchmark of 32% (Pew 2022 polling accuracy).
- 19% of respondents said their household received SNAP benefits in 2020, while the nationwide SNAP benchmark was 11% (Pew 2022 polling accuracy).
Why the benchmark comparisons matter
These figures show that polling statistics can land close to benchmarks on some measures and far from them on others. The vaccine-dose estimate was nearly exact, while self-reported voting and retirement-account ownership were much farther from benchmark levels (Pew 2022 polling accuracy).
That difference is important because it shows error is not evenly distributed across topics. Polls may perform differently depending on whether the question is about a recent behavior, a sensitive outcome, or a status variable that respondents may misreport.
Pew also said issue-poll errors of the magnitude seen in some of the least accurate 2020 election polls would alter issue opinion measures by less than 1 percentage point on average (Pew 2022 polling accuracy). That is a sharp reminder that the most visible election misses do not automatically translate into equally large issue-poll misses.
Election polling statistics in 2020
Election polling is often treated as the public test of survey methods, and the dataset includes several figures that show both the scale of the challenge and the limits of the criticism. Pew said national polls in 2016 indicated Clinton would win the popular vote by a 3-point margin, while the final popular vote margin was 2 points for Clinton (Pew 2018 trust polls). Pew described those national polls as basically correct on the popular vote outcome (Pew 2018 trust polls).
That does not mean every election poll was equally successful. AAPOR reported that the 2020 national-popular-vote polling error was the largest in 40 years (AAPOR executive summary). That is one of the most important figures in the dataset because it frames 2020 as an unusually difficult cycle rather than a normal polling year.
Final-week state polling details
AAPOR said 317 state-level presidential polls in the final two weeks reported statistical adjustments and represented 72% of all polls in that period (AAPOR executive summary). That alone indicates how common weighting and adjustment practices had become.
AAPOR also reported that:
- 92% of those 317 polls accounted for education level in their final results (AAPOR executive summary).
- Only 4% of poll respondents on average answered with something other than Biden or Trump in final-two-week state-level polls (AAPOR executive summary).
- There were 23 state-level presidential polls in the final two weeks that reported how respondents said they would vote (AAPOR executive summary).
The signed-error figures are also useful because they show that polling misses were not random in direction. AAPOR found that polls in states Trump won by more than five points in 2016 had an average signed error of 5.3 percentage points too favorable for Biden (AAPOR executive summary). In states Clinton won by more than five points in 2016, polls had an average signed error of 3.5 percentage points too favorable for Biden (AAPOR executive summary).
That means the average miss was systematically tilted, not just noisy.
National and state poll method mix in 2020
AAPOR’s method data help explain how quickly polling formats evolved:
- More than 60% of all pre-election polls in 2020 were done online (AAPOR 2020 pre-election polling report).
- 64% of national presidential polls were conducted online in 2020, up from 38% in 2016 (AAPOR 2020 pre-election polling report).
- In competitive battleground states, online polls rose from 38% in 2016 to 50% in 2020 (AAPOR 2020 pre-election polling report).
- 60% of all 2020 state-level presidential polls were conducted online (AAPOR 2020 pre-election polling report).
- Human-interviewer national phone polls using RDD fell from 36% in 2016 to 6% in 2020 (AAPOR 2020 pre-election polling report).
- In competitive states, human-interviewer and RDD or RBS polls moved from 18% in 2016 to 19% in 2020 (AAPOR 2020 pre-election polling report).
- Exclusive IVR use in state-level polls fell from 14% in 2016 to almost none in 2020 (AAPOR 2020 pre-election polling report).
What the election data suggest
The polling statistics do not support a single simple conclusion like “online is better” or “phone is better.” They show a field in transition, with more method diversity and more statistical adjustment at the same time.
A practical reading of the dataset is this:
- Polling methods diversified because the old response environment weakened.
- Online polling became common in both national and state pre-election work.
- Adjustments, especially education weighting, became standard practice in many final-week state polls.
- Even with modern methods, some election cycles still produced large, directionally consistent misses.
Polling statistics that matter most for readers
For readers who want the shortest possible digest, the most informative figures in this dataset are the ones that connect method, coverage, and accuracy.
Most important numbers
- 36% to 9%: the drop in typical telephone survey response rate from 1997 to 2012 (Pew 2012 representativeness study).
- 90% to 62%: the drop in standard survey household reach from 1997 to 2012 (Pew 2012 section 1).
- 43% to 14%: the drop in completed-interview share among contacted households from 1997 to 2012 (Pew 2012 section 1).
- 61%: share of national pollsters using different methods in 2022 versus 2016 (Pew 2023 polling landscape).
- 33%: share of national pollsters using probability-based panels in 2022 (Pew 2023 polling landscape).
- 4 percentage points: average difference between Pew’s 2022 topic estimates and government benchmarks across 26 topics (Pew 2022 polling accuracy).
- 60%+: share of 2020 pre-election polls done online (AAPOR 2020 pre-election polling report).
- Largest in 40 years: AAPOR’s description of the 2020 national-popular-vote polling error (AAPOR executive summary).
These are the polling statistics that most clearly explain the current debate. They show a field with lower participation, more method experimentation, and a mixed record on accuracy that depends on the metric and the election cycle.