How can polling be saved in 2024?

[AGA]

Can they fix their methodology and become accurate again? Or are we in an age where good polling is nearly impossible?

[charcuterie]

Have Sezler do it all!

The real answer is we have to see what sort of polling outcomes come out of 2022, and the adjustments made from that. Polling was really mixed in this year's election, and I'm not sure the answer is immediate as to why it was so much off (have we just gotten unlucky two general election cycles in a row?)

[Gone to Carolina]

Something I am curious about, and I'll have to find the study for it because I saw it a few months ago though it didn't register at the time. It basically asserted that people with low degrees of social trust (i.e. stated people are generally untrustworthy) swung massively towards Trump in 2016 and polling missed this because low trust people do not answer the phone (this latter bit makes sense to me I suppose).

So I guess where polling was wrong, and where I was wrong defending polling is the idea that weighting by education did a lot to fix the issue (and to be fair this did seem to be the case in 2018). However, it was clearly not 100% of the fix this year looking at the drastic polling average misses. So I'm thinking about ways pollsters could experiment with controlling for this lack of social trust.

I don't know if anyone has it on hand, but I'm wondering if "How are your neighbors voting?" as a line of questioning would've performed better this year. Could be a potential way to control for poll respondents being more likely to be high trust (and therefore more likely to be Democratic supporters at the moment and skewing poll results). 

Just one example of this was the final USC Dornsife Poll (14 Day Window):

Traditional Head to Head Poll (Rounded): Biden +12

"Expectations about social contacts voting" (Rounded): Biden +4

Just a thought.

[Benjamin Frank]

Have Investors Business Daily/TIPP do it.

[BlueSwan]

Once all of the data is in, we need to do a serious audit of just how wrong the polls were and where. Until then, I'm not sure we can really know what to do.

Also, these systematic polling errors may very well be caused by Trumps uniqueness as a candidate. If we have a Generic Republican vs Generic Democrat election, I bet all the conspiracy nuts who cannot normally be reached by pollsters, will just stay home.

[Adam Griffin]

Measurement of "social contacts" (i.e. "how are your neighbors voting?") seemed to be quite accurate this cycle. On a massive scale, it might prove to be a useful methodology going forward.

[Amenhotep Bakari-Sellers]

TX isn't a battleground anylonger, the D's thanks to HEGAR and Beto lost the gains the made in 2018, and D's are expected to lose the redistricting seats in TX since the Ds didn't pick up the House.

FL is still a Tossup, but Rubio isnt losing his race in 2022.
Tester and Manchin are probably gonna lose in 2024, both states are going the way of MO

[Panda Express]

It's entirely possible polling may revert to being more accurate once Trump is booted from office.

Regardless, my policy from here on out is to add 3-4 points to the Republican candidate on any poll.

[Amenhotep Bakari-Sellers]

It's entirely possible polling may revert to being more accurate once Trump is booted from office.

Regardless, my policy from here on out is to add 3-4 points to the Republican candidate on any poll.
[/quote

D's aren't running candidates that liberal again like HEGAR, McGath and Bollier, they did bad and cost D's downballot races, so you don't add 3-4 pts to R candidates

[McGarnagle]

Polling's dead, and television's the box they're gonna bury it in.

[pbrower2a]

Change the wording of questions now made, at least after early voting is done, probably after a question of eligibility; then one for approval or disapproval of the incumbent

1. In the upcoming Presidential election this November, do you plan to vote for

a. Smith, the Republican
b. Jones, the Democrat
c. someone else
d. I don't know
e. I am not voting

2.   Have you already voted, or do you plan to vote later

a. I have already voted
b. I expect to vote by mail or otherwise before Election Day
c. I expect to vote on Election Day
d. I do not intend to vote

[Amenhotep Bakari-Sellers]

TX and IA dropped off the list as battlegrounds, Ds are at 2016 levels of support in TX thanks to ultra liberals HEGAR and Beto

OH, FL, NC and GA are still battlegrounds and ME 2 due to fact Sherrod Brown and Angus King are up for reelection in 2024. Rick Scott is gonna bectargetred too

There will never be a 413 map again, after 2020 has passed by.

[Admiral Stockdale]

It may be that the issue is reaching a representative sampling of voters.  If you call 50000 people only to get 500 responses... ugh.  There's a real risk there.

The pollsters will argue that the Trump name and COVID may have impacted polling.

Let's see what they do with the 2 GA Senate runoffs... 

[Radical Leftist]

2016 and 2020 really have destroyed the credibility of polls, and I think that there needs to be some major changes to the way that pollsters conduct polls, even more than this last election cycle.
I for one trusted the polls this year, and was completely blindsided by Trump's holds in traditionally red states like Texas, Ohio, and Iowa; and I probably won't be trusting them in 2022 and 2024.

[pbrower2a]

Donald Trump is not the normal Presidential candidate. He breaks the mold, as he cares more about his ego and winning than he does about reshaping the world to fit his image.

The one thing that he is good at is getting a significant number of people to vote for him even if they loathe his conduct, personality, and agenda. Truth means nothing to him and Will is everything. Trump's appeal is analogous in a way to that of an abusive spouse: you will regret leaving me.

[SN2903]

Polling is irrelevant

[It’s so Joever]

I think one trend is that the generally more #populist (I love that meme) candidate has a tendency to overperform their polls.
I don’t know how pollsters can change that affect in their actual methodology, although I’m sure some ideas said here may be helpful.

As for how how Atlas should look at polls... I don’t know, looking at who YouTube/internet comments support more in a race and then adjusting for that?

(Interestingly, Obama had more YouTube comment support than Romney, so it’s not just a partisan thing it seems)

Really though, let’s wait for 2022 midterms to gauge polling in the new political era.

[Hope For A New Era]

(Interestingly, Obama had more YouTube comment support than Romney, so it’s not just a partisan thing it seems)

I've noticed that. Go back to political ads from 2012 and it's all people cheering on Obama. 2016 and 2020 ad comment sections are full of MAGAts.

[Amenhotep Bakari-Sellers]

D's need to nominate pragmatic candidates Bollier, Greenfield, HEGAR, were socialistic females that lost badly. If Beto, Castro, JD Scholten ran instead, of HEGAR, GREENFIELD and Bollier, we would of seen a different outcome

Gov, Senate races are different in TX where males win and Congressional races where females win. HEGAR got the same treatment as Valdez and was blanched not Abbott but by Cornyn, 10 pts, and we were told she raised so much money that went to paid consultants

All the Rs females that won were pragmatic enough, not tea partiers

[Bootes Void]

- the polling industry should take a break and obsess poll over every little thing like it did under trump and ask for various questions to gauge public opinion including "what do your neighbour thinks?" to get honesty

[ElectionsGuy]

Traditional live-caller polling is now terribly flawed. It can't be saved, and those in the industry aren't terribly concerned about 'saving it'. They continue to call it the gold standard when it clearly isn't anymore. Polling is only going to get harder, it will only get worse from here on out (not necessarily saying it'll always undercount Republican support, but it will under the current coalitions) unless they take notes from the pollsters that did get this election relatively accurate (not just talking about Trafalgar, cause they had some R-leaning boo-boos this year) You need mixed-modes to reach all demographics, and you need to oversample certain subgroups to make sure the sample is representative. Weighting by education clearly was not the single most important thing everybody thought it was. It's about geography (even when polls get rural regions, they're getting disproportionately dense areas), it's about makeup of party ID (most polls have too many John Kasichs, and not enough Brandon Strakas). This is all due to non-response bias by R-leaning groups which is making this more difficult and tedious but that's the kind of granular detail you have to go into to make sure it's right.

[ElectionsGuy]

Measurement of "social contacts" (i.e. "how are your neighbors voting?") seemed to be quite accurate this cycle. On a massive scale, it might prove to be a useful methodology going forward.

Wait a minute, you're telling me the methodology Trafalgar uses has been 'quite accurate'? Wow, I'm shocked I tell you!

[The Ex-Factor]

Something I am curious about, and I'll have to find the study for it because I saw it a few months ago though it didn't register at the time. It basically asserted that people with low degrees of social trust (i.e. stated people are generally untrustworthy) swung massively towards Trump in 2016 and polling missed this because low trust people do not answer the phone (this latter bit makes sense to me I suppose).

So I guess where polling was wrong, and where I was wrong defending polling is the idea that weighting by education did a lot to fix the issue (and to be fair this did seem to be the case in 2018). However, it was clearly not 100% of the fix this year looking at the drastic polling average misses. So I'm thinking about ways pollsters could experiment with controlling for this lack of social trust.

I don't know if anyone has it on hand, but I'm wondering if "How are your neighbors voting?" as a line of questioning would've performed better this year. Could be a potential way to control for poll respondents being more likely to be high trust (and therefore more likely to be Democratic supporters at the moment and skewing poll results). 

Just one example of this was the final USC Dornsife Poll (14 Day Window):

Traditional Head to Head Poll (Rounded): Biden +12

"Expectations about social contacts voting" (Rounded): Biden +4

Just a thought.

Don't know if this is the study you are referring to, but this graph is striking in that it shows that even after you adjust for education the people with low social trust still aren't answering polls. I don't know what the polling industry can do about that.

https://twitter.com/davidshor/status/886618278931705857

[Asta]

FOX Exit polls showed a landslide victory for Trump in OH and dead heat in MI, WI and PA.

So were these voters not comfortable telling pollsters till Election Day? If you have low trust in institutions, then you wouldn't have trust in it before or after the Election day.

[Crumpets]

There were two things polls were generally accurate on this year that I think can be something to build off of and something to pay more attention to going forward:

1) They were pretty good at guessing the Democrat's final vote share, as they were in 2016. There are obvious exceptions, but the "is Joe Biden above 50% in X state?" metric turned out to be a reasonably good predictor of how he did. In Florida, for example, Biden's final poll number on the RCP average was 47.9. His final vote share there was 47.8. In Michigan, his final poll number on RCP was 50.0. He currently has 50.6 there. In North Carolina, he was at 47.6 in the average, and he's currently at 48.6 there.

2) They were reasonably accurate in terms of which states voted to the left/right of which other states. This is notably an improvement since 2016, where nobody was predicting Texas would vote to the left of Iowa. Here, even if the margins were off, the order of "most Democratic" to "most Republican" state was pretty accurate, as borne out by 538's snake chart.