Stung by Recent Misses, Pollsters Try to Improve Accuracy in 2022

[question everything]

After big misses by pollsters in 2016 and 2020, each election now brings an inevitable morning-after question: Did the pollsters get it right?

(snip)

But even when multiple polls are averaged together, the results can be off. In 2020, the final public polls, as averaged by the Real Clear Politics website, suggested that President Biden was sailing into Election Day with a lead of roughly 6.7 percentage points in Wisconsin. He squeaked by with a win of less than 1 point. The final merged polls of Florida found Mr. Biden to be narrowly ahead. He lost by more than 3 points. Overall, national polls in 2020 were the most inaccurate in 40 years, a study by the main association of survey researchers found, and state-level polls in 2016 were significantly off the mark.

This year, pollsters are trying a range of tactics to make sure they are interviewing a mix of people who fully represent the electorate. Some are inviting voters by text message to take surveys, adding to the interviews they conduct on cellphones or landline phones. The goal is to expand the number and type of people agreeing to share their views.

(snip)

But pity the poor pollster. In the 2020 presidential election, five states were decided by a margin of 1.3 percentage points or less: Arizona, Georgia, Pennsylvania, Wisconsin and North Carolina. Four of those states are on a knife’s edge again, with Senate candidates separated by 1.5 percentage points or less in Arizona, Georgia and Pennsylvania, and by about 3 points in Wisconsin, the 538 aggregate of polls finds. Nevada’s Senate race is also close, with the GOP candidate ahead by 1.4 points.

Polling is usually too blunt an instrument to detect which party is really ahead in a 1.4-point race. But even if a pollster points to the correct winner in a race, the poll can be inaccurate. Predicting a narrow win, only to find that the winner racked up a big margin of victory, also shows that the pollster didn't accurately detect which types of voters were truly going to participate in the election.

https://www.wsj.com/livecoverage/election-midterms-2022/card/stung-by-recent-misses-pollsters-try-to-improve-accuracy-in-2022-WYaPTOPPQX1pPG0NixUF (subscription)

[MontanaMama]

1. I won't ever pay attention to polls going forward.

I don't answer phone calls from numbers I don't recognize and I believe that's true for a whole lot of people. If phone calls are how the pollsters attempt to reach people, they'll never get a good cross section of Americans for an accurate poll. GenZ will never answer their phones...maybe if pollsters texted them...kidding not kidding. Younger people don't want to answer the phone.

[iemanja]

2. They were posted constantly here

Usually to complain about them. I wouldn’t have even known about most of them if not for DU.

[marble falls]

3. First they need to admit: they are pretty close to political horoscopes.

[PortTack]

5. This..or the 1950's weatherman

[uponit7771]

9. Not in other G20 countries, they are pretty accurate there. M$M repeats MAGA stupidity here is more

... of an issue than anything especially when MAGA media starts a thought stream that rates.

[Bev54]

4. As stated by the data guys that got it right, polls are a predication and do not

tally votes. Early voting data can tell a lot more so they could get it right and polling sites were not paying attention to the actual data or chose to ignore it. The media and polls like Nate Silver's know better but chose to parrot the republican narrative from republican pollsters. They all should be ashamed of themselves and apologize to the public.

[iemanja]

8. They aren't even a prediction

They are the survey of views of a certain sample at that moment. Also, most everyone ignored the margin of error, and very few results this cycle were outside the margin of error.

[Bev54]

11. They truly did ignore the early vote and what it was showing as well.

Both pollsters and media really were willing suckers in this deception.

[Yavin4]

6. Voter participation is difficult to predict because there are more options for voters to participate

A voter participation model that counts if a voter voted in the past two or three election cycles may miss a voter that will be voting in this cycle perhaps because participating in the current cycle is easier than it was in the past. Maybe there's early voting, drop off ballots, mail in ballots, etc.

[BlueIdaho]

7. I really gave the polls zero attention this election

Seems my decision worked out just fine.

[honest.abe]

10. Political polling may become a dead science given the number of people who refuse to participate.

To make it even more difficult the response rates are very different depending on age groups which then further skews the results. I wish there was no polling at all except the poll on election day.