The main reason why polls fail to predict elections, imho, based on

[Mr.Mystery]

an informal search of on-line sources.

There are many reasons why polls fail to predict the outcome of an election:

* a lot of people refuse to answer phone calls from numbers they don't recognize

* to use data correctly, the pollster needs not only the responses, but also personal information such as age, education level, income etc., and people do not want to reveal that to a stranger

* all that information needs to be "weighted," and that is an imprecise guesstimate. For instance, if uneducated white males make up 18 percent of the respondents, but they are predicted to be 26 percent of the voters, their responses would have to be multiplied to reflect that

* sample size may seem large, but it's not that large -- only 38 people out of 1000 need to flip their support to shift Candidate A from an "overwhelming" +12 lead to "almost tied" +4 lead.

the biggest reason polling is inaccurate however is this

* about 40 percent or so of people responding to the poll will not vote on election day. Conclusion, poll respondents can easily say who they prefer in an election when answering a poll, but are those same people motivated enough to go out and vote on election day?

Pollsters have no good way of determining this.

Bottom line: It's all about turn-out.

[BannonsLiver]

1. Be careful. "Dad" will be around to scold you soon, I'm sure.

[we can do it]

7. Ha!

[ColinC]

2. Knock every door.

[iemanja]

3. 40%?

Wow.

[Mr.Mystery]

5. It makes sense, since only about 60 percent vote even in a typical Presidential election cycle

Biden v Trump was not typical with a 67 percent turnout.

https://fairvote.org/resources/voter-turnout/

Also this:

"nearly all of Pew Research Center’s public opinion polling on issues is conducted among the general public and not just among voters. Nonvoters make up a sizable minority of general public survey samples. In our 2020 post-election survey, nonvoters were 37% of all respondents (8% were noncitizens who are ineligible to vote and the rest were eligible adults who reported not voting)."

https://www.pewresearch.org/methods/2021/03/02/what-2020s-election-poll-errors-tell-us-about-the-accuracy-of-issue-polling/

[iemanja]

8. Don't they adjust for likely voters?

To screen out some of those non-voters? They usually have questions geared at how likely someone is to vote.

[Mr.Mystery]

9. Excellent question -- except they have no good way of determining "likely voters."

Aye, there's the rub.

[quote] Every pollster has a different method for predicting who will end up voting, and none offer definitive accuracy. (There’s the Gallup method, the Perry-Gallup method, probabilistic vote-scoring, predictions from machine-learning models, the matching of vote-histories, and so on.) So ambiguous and uncertain are likely voter models that Burns “Bud” Roper, son of the public polling pioneer, said in 1984 that “one of the trickiest parts of an election poll is to determine who is likely to vote and who is not. I can assure you that this determination is largely art.” [unquote]

https://lithub.com/how-pollsters-got-the-2016-election-so-wrong-and-what-they-learned-from-their-mistakes/

[iemanja]

11. Interesting. Thanks. nt

[milestogo]

4. Voting early means less of a crowd on Election Day.

So I voted early. I would wait forever to vote, but I know some people can't.

[LakeArenal]

6. Also:

The polls want a simple yes or no, or, on a scale from one to five…

To complex issues that there is no adequate simple polling.

[W_HAMILTON]

10. I think part of the problem is that they often DO predict the winner...

...when margins of error are taken into account. And the typical layperson does not understand how the margin of error (MoE) works in this regard. Let's look at an individual poll to explain...

The most recent NY Times/Siena poll of the Warnock/Walker GA Senate race shows Warnock at 49% and Walker at 46%. The MoE is 4.8 (relatively huge, actually). This means that this particular poll indicates that the final outcome could be anywhere from:

53.8% - Warnock (extreme situation where entire MoE favors Warnock)
41.2% - Walker

to

44.2% - Warnock
50.8% - Walker (extreme situation where entire MoE favors Walker)

So, technically this poll shows that any final outcome where Walker wins by no more than 50.8% of the vote to Warnock winning by no more than 53.8% of the vote would count as an accurate prediction. Since correct application of the MoE means that you apply the MoE to BOTH candidates' poll numbers, this results in a wide variance. But the typical person viewing a poll does not take this into account. They just see Warnock up by three points and assume that if he wins in a squeaker or Walker ends up winning, the poll was wrong, when that is not the case. Again, taking into account the MoE, Walker could theoretically win by over six points and the poll would still have called the race correctly, even though the average person looking at it would say they blew it big time.

This is why polls are almost meaningless in tight races (which most of the most frequently polled races usually are). Once you factor in the MoE (almost always two to four points), it means the poll is basically showing that either candidate could win, in which case, no shit!

[Mr.Mystery]

12. Good point.

Then there was the Florida election of 2000, in which Gore voters THOUGHT they voted for Gore, but ended up voting for Pat Buchanan (because of the infamous butterfly ballot) or invalidating their Gore vote by writing in his name where the ballot says "write-in candidate."

So . . . the exit polls were more correct than the official count.

Those confusing ballot designs have finally been eliminated as far as I can tell.

[dsc]

13. A few things

even in the early polls they do check for registered voters and at your present address (I was excluded from one poll since I had moved and not reregistered yet even though I always vote). Two the MOE does account for the 'switch' of votes you discuss. The MOE is designed to account for errors generated from a bad sample. Now the modeling for likely voters is an issue. Kansas was way off since the electorate in its primary was vastly more Democratic than it normally would have been. As we get closer to the election this should be less of an issue since many will have already voted and the rest will presumedly know if they will or won't vote. I also have to say the MOE explanation above my post but below yours is a must read.

[gratuitous]

14. Reputable polling firms want to stay in business

Bad numbers mean less repeat business, and if you're wrong too often, you go out of business. For the sham outfits that are trying to flood the zone with discouraging news for us, they don't give a fart in a frying pay if they're totally wrong. That's not the service they're offering to the well-heeled people who pay them.

But a reputable firm analyzes its data over and over again, trying to get the numbers right and taking into consideration the reasons you've enumerated as well as other reasons more obscure and arcane than the average person is aware of. After the election, the good firms will again try to determine where they got it right but more importantly where they got it wrong and correct for that in the future (if they can).

It IS all about turnout, and the sham outfits try to depress turnout by discouraging voters they suspect aren't likely to vote for the preferred candidate of their paymaster.

[SWBTATTReg]

15. I don't participate in polls because I'm sick and tired of hearing about the nonstop republican

voter this, that, etc., all kinds of reasons and nonsense about why they feel the way they do, and there really isn't a 1 for 1 on what democratic voters feel, etc. (at least not a one for one).

Thus, I'll be silent and spring my Nov. surprise when I voted democratic across the ticket as I normally do.

[PortTack]

16. Current poll predictions are really no different than the 1950s weatherman

Sunny…partly cloudy later with a chance of showers. Windy..clam later in the day.