Question: do you have an actual position on polls?

[brooklynite]

It seems that whenever they're bad we hear "I don't trust any polls" and "the media is skewing the results to help the Republicans", but when they're good, nobody questions them or the pollsters' motives.

[SYFROYH]

1. I like polls but they have limitations.

[Celerity]

9. +1

[hlthe2b]

2. Healthy skepticism and I read methodology.

[GreenWave]

3. We need an anti-poll!

Do you answer the phone? I wait for the recording to avoid scammers.
Do you know personally anybody who has been polled recently?
Were GOP talking points featured predominantly?
Do you know of any GOP defectors?

[House of Roberts]

7. I don't answer the phone,

unless I know the number. I then put the number into my browser to see if it belongs to a legitimate entity. It seldom does.

[Genki Hikari]

22. I've been polled several times over the years

My household was even a "Nielsen Family" back in the 80s.

They don't need to poll you to get an idea of certain probabilities. They only need to ask enough people, which is 1500-2000 people. The results from that sample won't be far off from asking 10,000 or 100,000 or more.

This was established in stats, long ago.

[Sympthsical]

4. They're rough ideas and can capture trends

They're not ironclad predictors, but they can give an inkling of things, particularly if many polls are saying similar things.

[Meadowoak]

5. I have NO trust in polls, they have been wrong more often than they've been right.

Just vote like we're 10 points behind, even if the polls tell us we're ahead.

[Genki Hikari]

23. No, they haven't been wrong that often.

They're right, more often than you realize. Why things are in your grocery store, and how they're positioned--that's all based on statistics. Your grocery store stays in business through getting and heeding good polling. Same as other businesses--they know what you want, when and how, better than you do.

Really.

The reason so many people think polls don't work is because they don't understand what polls are, or how they work.

Stats are about probability, which is always about the future from the vantage point of the present. They're not iron-clad predictors. They're about the chances of something happening in the future, not what will happen.

Once you wrap your head around that, then stats/probability makes more sense.

[ProfessorGAC]

6. I Question Their Methods Quite Often

I knew people that worked at Pew for a long time.
The pollsters are managed more by political science folks, than social statisticians.
As a result, I believe there is cause to cause how questions are formulated, how their stratification is constructed, and how they weight segments based on demographic categories & location.

[Hermit-The-Prog]

8. Polls ain't much good without a line and a hook.

[dpibel]

10. What's your actual position?

As long as we're sharing info.

[iemanja]

11. They don't seem to be very accurate in recent years.

That’s my position. I don’t think they are part of a nefarious media plot against Biden.

[keithbvadu2]

12. There are many polls. Pick the one you like and swear by it. Swear at the others. .............

There are many polls.

Pick the one you like and swear by it.

Swear at the others.

.................... That's why candidates sometimes refer to mystery polls that no one can locate to check.

[TheProle]

13. Polls as accurate as they have ever been, study says

It seemed to be a hat trick of polling catastrophes: Brexit, the 2016 US presidential election and the 2017 British general election. But researchers now say that despite popular perceptions, polls are as accurate as they have ever been.

(Snip)

Writing in the journal Nature Human Behaviour, Jennings and coauthor Christopher Wlezien, from the University of Texas at Austin, describe how they conducted a series of analyses based on almost 31,000 polls covering 351 national elections in 45 countries between 1942 and 2017.

Looking at 286 elections, chosen because polling began at least 200 days before election day, the team found that errors decreased overall from an average of four percentage points at that stage to under two points by the eve of an election. Errors were larger further away from election day for presidential compared to legislative elections (such as a general election).

“Often in presidential elections voters are still learning about candidates, they are still acquiring information, they don’t necessarily know very much about their policies and characteristics and so forth,” said Jennings. “Whereas in [legislative elections] where people are voting for parties, party loyalties tend to be more stable and durable.”

Looking at polls for the last week of campaigns for 220 national elections in 32 countries over the decades to 2017 – with data for a handful of countries stretching back to 1942 – errors have held steady at about 2%. When the team looked only at the 11 countries that had regular polling over several decades, polling errors were found to have dropped over time.

https://www.theguardian.com/politics/2018/mar/12/polls-as-accurate-as-they-have-ever-been-study-says

[WVGal1963]

14. Polls have changed a lot

Polls used to rely on calls to land line calls. Remember?

Land lines don’t much count any longer. Very few people have them.

The actuarial scientists are super busy, scrambling, trying to figure it out. Predicting….calculating risk…..etc.

So. Polls?

Not a believer.

[Talitha]

15. I don't put much faith in polls.

[Ace Rothstein]

16. I had much more faith in them pre-2016

Trump broke them and now I just kind of ignore them whether good or bad.

[LakeArenal]

17. I think some polls are just a bit more dependable than my horoscope.

[Metaphorical]

18. Polls go from being very good to very bad

There are a lot of bad actors out there, but they are not the rule but the exception. If possible, I like to get my hands on private polling done for candidates, as these are likely to be more rigorous. I also have found that people have become much less inclined to speak to pollsters or to be accurate in how they're being polled.

That being said, when you aggregate several polls, you look not only at the mean (the average) but also the variability in the data - how likely the poll is close to or far from the actual mean (or how divergent different poll results are from one another). The higher the variability, the more uncertain the mean represents anything meaningful. In 2020 by this time, the variability was fairly low, and the polls turned out to be fairly accurate. On the other hand, the variability in polls this year is higher - fewer polls being done at the national level, a lack of consistency from one poll to the next by the same pollster, and a bigger uncertainty about how Trump plays in the decision-making process (his influence has waned, but this is a hard thing to measure).

This means that the polls will likely not be accurate this cycle, though whether that favors the Dems or the Republicans are difficult to say. What makes this even more difficult to predict is that this will be the first election cycle since the 2020 redistricting efforts. Many models developed before 2020 are likely to be no longer accurate.

I think the election will break slightly in the Democrat's favor in both the House and the Senate, but it will likely not be a blowout. Inflation is a problem, but unemployment remains tight, and people are more likely to be pessimistic about the economy if they are the ones who have lost their jobs. Biden generally gets good marks for handling the war in Ukraine, and the Russians are not popular. Inflation is high, but people are also seeing wage increases after many years of not seeing them. Gas prices are lower, and Biden releasing gas from the Strategic Reserve will keep them low through the election. I also think pollsters are oversampling Republicans when the Republican party is facing some serious defections. The Jan 6 committee and the documents have both had an impact. Roe has opened the floodgates on anti-abortion legislation, and there's a sense from everyone I interact with that the Republicans have overreached. Finally, the election is in three weeks, so it is unlikely the ground game will change radically between today and then.

This year, take the polls with several handfuls of salt. That's likely to change as more data is gathered over the next decade, but they are now fairly noisy, primarily because of redistricting.

[IcyPeas]

19. every poll had Hillary Winning in 2016. they are meaningless to me since.

[2naSalit]

20. I don't take them...

Too seriously no matter who they favor. I think of them more like the horoscope in the newspaper.

[Genki Hikari]

21. The Long Answer

They can (and often do) work, but accurate polls depend on following good methodological practices to yield good results, such as:

1) A decent representative sample of the population. Finding that perfect cross-section of people to poll is the most difficult task a pollster has, but when you nail it, you get scary-accurate results.

2) Pollsters don't have to ask everyone in, say, the United States or Alabama or wherever what they think to get an accurate result. They only have to ask ENOUGH people. Margin of error is why.

The formula for MOE is well-established; I won't bore anyone with the math and geek terminology involved. Non-stats people don't need any of it, because there's a shortcut to understanding what particular MOE will yield the best polling result vs a bad one. Know this number: +/-3% MOE. That's considered the gold standard for an accurate poll.

For politics, it's also not so necessary to use the formula early on in a voting cycle as it is close to an election. "General" polls can get a good idea of macro trends at +/- 5%. These are often used early in elections because polls are expensive, and the more people you have to contact for a poll, the more expensive it gets. That's why you'll see numbers all over the place in the early months of elections, and then those numbers start getting more consistent over time. This is because the pollsters start increasing their samples over time, thus the MOE numbers shrink and consistently hover around the "magic" number of +/-3%, the closer the polling firms get to the actual day of voting.

Note: Margin of error is important to keep in mind for close races, and why people tend to think polls are rigged rather than operating as designed. Here's why:

If a poll with a +/-3% MOE has Candidate A at 51% chance to win, while Candidate B is at 49%, then that race is effectively a tossup. Why? Because the 3% applies either way to each number in that matchup. A 51-49% split with a +/-3% MOE means that Candidate A chances to win the vote are 48% - 54%, while Candidate B has a 46% - 52% chance. That means that, even if on the morning of the election, the polls show Candidate A up by 51%-49%, the actual votes may tilt the chance of Candidate B winning to 52-48%. The polls weren't wrong in that case, because that result is well within the +/- 3% margin of error.

This is exactly what happened to Hillary in 2016. She was consistently ahead in the polls, but she was never outside the margin of error for most of September through November. That meant for all of those months, that...thing...could still win the election, because it was within the margin of error for that...thing...to do so.

If you want to rest easier, then you want your candidate polling 55% or above on election day. That's outside a+/-3% margin of error.

3) Asking the right questions. This is another place where pollsters can cut their own throats if they're not careful.

Pollsters need to know what to ask, which means paying attention to events and general moods that can affect results. If a pollster in 2022 doesn't ask participants about abortion and how it will affect their vote, then the polling results may be hopelessly skewed.

The other thing pollsters have to be careful of here is how they word/frame questions. My favorite example of how changing a question can change poll results is the famous church vs Sunday approach to polling. If you ask people, "Did you go to church on Sunday/or a religious service for your faith?" then the respondents feel the judgment involved there, and they will say that they did attend a service that week. Even if they didn't. Because they don't want to look bad to the polling firm rep. This even has a term in polling, "The Halo Effect."

Pollsters lately have started asking another question instead: "What did you do this weekend?" Respondents are more likely to reply with what they actually did, because no judgment is involved. They're telling the pollster what they did, not what they think the pollster wants to hear.

This is how we know that church attendance in the US isn't anywhere near 40% every week. It's more like 15-18%.

Asking loaded questions of participants in a sample will create crappy results. This is why hyper-partisan push polls are worthless: They ask questions to get a result the person paying the pollster wants, rather than getting an accurate glimpse of what's going on so that a (sane) campaign can recalibrate to counter a bad polling number.

I can guarantee that the RNC knew that...thing...was in trouble all along in 2020. They knew Biden was flirting with a 54-55% chance to win the election, if not doing better than that. Rather than seeing those numbers and saying, "So what do I need to change about my campaign to get the numbers more in my favor?" like a normal candidate, that...thing...instead started plotting to overthrow the election.

---

The cool thing about polls is that you can see for yourself whether or not a poll is any good. Every reputable polling agency subscribes to an industry standard to provide their methodology to the public. Most of the time, news stories will provide a link to the full poll. Follow it, and you can see not only all of the pollster's results, but also all of the nuts and bolts of:

* How many people were in their sample
* Demographic info about the sample
* Method of contact (cell phone, online, landline, in person--no reputable firm relies on landlines alone or even as a majority of their polls anymore)
* The exact questions asked
* Polling results for each question, often broken down by demographics. Want to know how married black women 18-35 responded? It'll probably be in there.
* And, of course, margin of error.

You can see for yourself if the poll is one you can trust.

[Just A Box Of Rain]

25. A long answer

and an excellent one!

Well done.

[uponit7771]

24. YES !!! In the US they're criminally horrible