NYT LIVE POLL THREAD:

[Former Dean Phillips Supporters for Haley (I guess???!?) 👁️]

Anyway, its clear that Miller will probably finish above Ojeda in this poll, but, as I stated before, this is not some god-tier level polling, and the part where voters actually care is coming up.

I dunno, while surely it isn't "god tier," I think this is probably the best quality, or else close to the best quality, set of individual house race polls we are likely to get.

The methodology they are using is good - it is not typical Siena methodology, and they are calling off of the voter file and weighting by education, two methodological details which 2016 showed can be quite important.

What other house polls do we have?

1) Internals - these are often good quality campaign polls that are done professionally to a high standard for the purpose of things like message testing. Sometimes these are lower quality though (e.g. a pretty decent # of PPP polls), and may be done more for the purpose of a campaign trying to find good numbers to release than for message testing/campaign testing research. In either case though, even though some of the pollsters who do the higher quality internals can be quite good, the internals are not reliable because they are selectively released by campaigns or by supporting committees/PACs/etc.

2) Generic Ballot polls - there are lots of generic ballot polls, many by good quality pollsters, but this is just not the same thing as a poll of an individual house race.

3) Monmouth - The only other pollster I can think of who has done a decent # of House polls in Monmouth. Monmouth is generally a good pollster, however I think these NYT/Siena polls may actually be better methodologically than the Monmouth polls (particularly because NYT/Siena calls off the voter file).

4) Stray house polls done by some random, usually low-rated pollster who hasn't done many other polls, usually done for a local news organization - These are almost invariably of horrible quality, because they are done not so much for accuracy, as just because the news organization that is paying for the poll wants to write an article. These polls often have ridiculous #s of undecided and are basically junk.


I don't think think there is really any other source of good house polls (?). So I don't think we should knock these polls too much. They seem to be basically the best quality polls of individual house races available, the only real competitor being Monmouth.

[Former Dean Phillips Supporters for Haley (I guess???!?) 👁️]

Well in that scenario they'd very likely win the House(even though the Clinton districts don't quite get to 23 districts, if they get say 18 pickups it will be very easy for Democrats to pick up a few more seats to win the majority). And I think the Democrats are underrated with some of these districts, and even if people think they'll win the lion's share that doesn't lead them to expect Paulsen to lose by double digits.

Ideally we'd like to not just get the minimum needed to narrowly control the house, but enough for a reasonable majority with at least a little bit of padding. With only a bare majority, it could very easily be lost even in 2020 (much less 2022), and it could be problematic electing a speaker, much less getting legislation passed while allowing for at least a small number of possible defections.

That is why the races like VA-07 and, NC-09, WA-03, etc. are important - because (while the Clinton districts and especially the suburban districts like MN-03, WA-08, KS-03, etc are likely to fall first), a decent chunk of those other districts are needed to have a majority large enough to provide basic margins of safety.

[Former Dean Phillips Supporters for Haley (I guess???!?) 👁️]

And, also, this is just one pollster, one that has gotten rather odd results so far in many races, so I would take such polls with a grain of salt, especially since, according to their own numbers, Ojeda is actually leading with 2016 voters, but they were screened out.

That's not what the 2016 voters thing means. That alternative scenario showing Ojeda winning is if the electorate were similar to that that voted in 2016 - i.e. if it were a Presidential year size turnout, with the electorate having the same statistical/demographic characteristics as in 2016.

That does not mean that 2016 voters are screened out. There are many 2016 voters included in the turnout scenario that the poll is projecting. Some are screened out, but that is certainly appropriate for a midterm.

[Adam Griffin]

And, also, this is just one pollster, one that has gotten rather odd results so far in many races, so I would take such polls with a grain of salt, especially since, according to their own numbers, Ojeda is actually leading with 2016 voters, but they were screened out.

That's not what the 2016 voters thing means. That alternative scenario showing Ojeda winning is if the electorate were similar to that that voted in 2016 - i.e. if it were a Presidential year size turnout, with the electorate having the same statistical/demographic characteristics as in 2016.

That does not mean that 2016 voters are screened out. There are many 2016 voters included in the turnout scenario that the poll is projecting. Some are screened out, but that is certainly appropriate for a midterm.

The thing is, WV is (perhaps still) one of the few places where Democrats would/should do better with lower turnout and/or an older voter base. Maybe the generational turnover is complete there now, but the broader dynamic has been true for quite some time throughout a good swath of Appalachia (including KY, where lower turnout might have beat Bevin; I think Conway did best with 65+ year-olds).

[levgre]

Ojeda might still win but some of his social media strategies and otherwise aren't great.  He has the energy and charisma, but he's not drawing in conservatives very well.  He doesn't come off as completely genuine, and conservatives are by default extremely suspicious of anyone with D next to their name.

[Co-Chair Bagel23]

I don't think Ojeda could even win this CD

[Politician]

Are we seriously writing off Ojeda based on one poll?

[GeorgiaModerate]

Are we seriously writing off Ojeda based on one poll?

And an unfinished one at that.

[Jeppe]

Are we seriously writing off Ojeda based on one poll?

Well, Republicans have outperformed their summer polling by 15 points on average since 2004 in West Virginia. This sort of lines up with that. The lowest polling drop was the now GOP Governor Jim Justice, who was leading 10% and ended up winning by 7%

[KingSweden]

Are we seriously writing off Ojeda based on one poll?

ATLAS

[wbrocks67]

I really wish NYT was also asking Senate/Gov questions here too in these districts.

[Filuwaúrdjan]

This is a methodological fiasco and a sad, shabby disgrace. Genuinely disappointed that there are any posters here who have taken this farce at all seriously.

[ON Progressive]

This is a methodological fiasco and a sad, shabby disgrace. Genuinely disappointed that there are any posters here who have taken this farce at all seriously.

These polls are perfectly fine methodology-wise. It’s just that idiots here can’t stop going nuts over tiny sample sizes.

[Former Dean Phillips Supporters for Haley (I guess???!?) 👁️]

This is a methodological fiasco and a sad, shabby disgrace. Genuinely disappointed that there are any posters here who have taken this farce at all seriously.

What specifically is wrong with their methodology (as opposed to wrong with how people interpret it before it is actually done)?

[Zaybay]

This is a methodological fiasco and a sad, shabby disgrace. Genuinely disappointed that there are any posters here who have taken this farce at all seriously.

What specifically is wrong with their methodology (as opposed to wrong with how people interpret it before it is actually done)?

their weighing process. Basically, its weighed by 3 factors, census data, education, and voter history. The problem is that mostly the census data, and voter history.
1. The census data means that trends in the district, seen in IL or VA, would not appear, so a district that has seen a shift in demographics would still have its 2010 demographics. Even NYT says that this process isnt even used by most pollsters.
2. Voter history is a big one, as it gets rid of the possible new voters that enter the system. This is a problem that a lot of pollsters have, but usually they move to voters who are likely to vote by now. Siena hasnt.(they have it as a subcategory you can see, which actually shows that WV-03 is even, but its not used in the overall number)

There are other small methology bugs, but these are the two main ones.

And there is still the prevailing problem that all Siena polling has and never bothers to fix, their incumbency fixing. Basically, they will give incumbents an edge in who is sampled because incumbents. This may be fine in neutral years, but in wave years or even favoured years, its incredibly bad, and can lead to the NY polls they have released, which all have a more GOP favoured sample than 2016 and 2014.

[levgre]

You must be logged in to read this quote.

What specifically is wrong with their methodology (as opposed to wrong with how people interpret it before it is actually done)?

their weighing process. Basically, its weighed by 3 factors, census data, education, and voter history. The problem is that mostly the census data, and voter history.
1. The census data means that trends in the district, seen in IL or VA, would not appear, so a district that has seen a shift in demographics would still have its 2010 demographics. Even NYT says that this process isnt even used by most pollsters.
2. Voter history is a big one, as it gets rid of the possible new voters that enter the system. This is a problem that a lot of pollsters have, but usually they move to voters who are likely to vote by now. Siena hasnt.(they have it as a subcategory you can see, which actually shows that WV-03 is even, but its not used in the overall number)

There are other small methology bugs, but these are the two main ones.
[/quote]

I read their explanation of their weighing process on their methodology and poll pages and I don't see census data or voter history mentioned among their factors.

"In general, we follow a pretty standard set of weights and weight categories. We weight by age, party, gender, region, race, likely turnout and education."

"Here, we’re weighting by age, party registration, gender, likelihood of voting, race, education and region, mainly using data from voting records files compiled by L2, a nonpartisan voter file vendor."

"Likelihood of voting" could possibly take into account voting history, but they don't say that it does.  But I would guess it is completely based on the answers to the question ""Likelihood of voting" in the poll.

Also Carol Miller is not an incumbent.  I don't know if by "incumbent" you mean the party or the candidate, usually it refers to a candidate.

[Zaybay]

This is a methodological fiasco and a sad, shabby disgrace. Genuinely disappointed that there are any posters here who have taken this farce at all seriously.

What specifically is wrong with their methodology (as opposed to wrong with how people interpret it before it is actually done)?

their weighing process. Basically, its weighed by 3 factors, census data, education, and voter history. The problem is that mostly the census data, and voter history.
1. The census data means that trends in the district, seen in IL or VA, would not appear, so a district that has seen a shift in demographics would still have its 2010 demographics. Even NYT says that this process isnt even used by most pollsters.
2. Voter history is a big one, as it gets rid of the possible new voters that enter the system. This is a problem that a lot of pollsters have, but usually they move to voters who are likely to vote by now. Siena hasnt.(they have it as a subcategory you can see, which actually shows that WV-03 is even, but its not used in the overall number)

There are other small methology bugs, but these are the two main ones.

I read their explanation of their weighing process on their methodology page and the poll page and I don't see census data or voter history mentioned among their factors.

"In general, we follow a pretty standard set of weights and weight categories. We weight by age, party, gender, region, race, likely turnout and education."

"Here, we’re weighting by age, party registration, gender, likelihood of voting, race, education and region, mainly using data from voting records files compiled by L2, a nonpartisan voter file vendor."

Likelihood of voting is based off of past voting as well as what the person says. If you are to say, "Ill definitely vote" and then respond that you havent voted recently, they wont mark you as a likely voter.

As for the census data one, its what they base the demographics on, so its not weighing, but it would be preferable if they used 2017 census numbers instead.


Edit: forgot to mention this, but if you want to see the weighing in action, look at all the races they have surveyed so far. You may notice a trend in the percentages, one that is statistically impossible. Besides MN-03, all of them are within 1% of each other, which may be believable, if they all were similar districts. But they arent, and getting all of these districts to have a 1% difference has around a 5% chance of occurring normally.

[Devout Centrist]

This is a methodological fiasco and a sad, shabby disgrace. Genuinely disappointed that there are any posters here who have taken this farce at all seriously.

I don't wish to empty quote, but I agree. This entire episode is disgraceful and most of you ought to be ashamed.

[levgre]

You must be logged in to read this quote.

What specifically is wrong with their methodology (as opposed to wrong with how people interpret it before it is actually done)?

their weighing process. Basically, its weighed by 3 factors, census data, education, and voter history. The problem is that mostly the census data, and voter history.
1. The census data means that trends in the district, seen in IL or VA, would not appear, so a district that has seen a shift in demographics would still have its 2010 demographics. Even NYT says that this process isnt even used by most pollsters.
2. Voter history is a big one, as it gets rid of the possible new voters that enter the system. This is a problem that a lot of pollsters have, but usually they move to voters who are likely to vote by now. Siena hasnt.(they have it as a subcategory you can see, which actually shows that WV-03 is even, but its not used in the overall number)

There are other small methology bugs, but these are the two main ones.

I read their explanation of their weighing process on their methodology page and the poll page and I don't see census data or voter history mentioned among their factors.

"In general, we follow a pretty standard set of weights and weight categories. We weight by age, party, gender, region, race, likely turnout and education."

"Here, we’re weighting by age, party registration, gender, likelihood of voting, race, education and region, mainly using data from voting records files compiled by L2, a nonpartisan voter file vendor."

Likelihood of voting is based off of past voting as well as what the person says. If you are to say, "Ill definitely vote" and then respond that you havent voted recently, they wont mark you as a likely voter.

As for the census data one, its what they base the demographics on, so its not weighing, but it would be preferable if they used 2017 census numbers instead.
[/quote]

Gonna need a source on that otherwise you are just making an assumption.

I don't see where they mention where they get their demographics from, but this is how they explain their turnout modeling.

"As the last section implies, our turnout model plays a far bigger role in our survey than it does in most public polls. It determines whom we call. How we adjust our sample. And, of course, it has a big role in determining exactly who we think is likely to vote.

In a lot of ways, the turnout model is very complicated. But conceptually, it starts with three simple observations about turnout in midterm elections dating to 1982, based on available voter file, census and polling data: "

[Former Dean Phillips Supporters for Haley (I guess???!?) 👁️]

their weighing process. Basically, its weighed by 3 factors, census data, education, and voter history. The problem is that mostly the census data, and voter history.

1. The census data means that trends in the district, seen in IL or VA, would not appear, so a district that has seen a shift in demographics would still have its 2010 demographics.

Demographic change is a slow process (not actually that much difference between 2018 demographics and 2010 demographics in most places), and in any case it appears most likely that they are using more recent ACS rather than 2010 data (the methodology page mentions the ACS frequently). In addition, for some demographics like age and sex, those are directly in the voter file on the level of the individual voter, so no census data is needed. In some states, individual race data is also available, so in those cases they will be using that, not census/ACS data.

You must be logged in to read this quote.

Which is a deficiency in the methodology used by most pollsters, not a deficiency in their methodology. Their methodology going off the voter file is more like the higher quality polls that campaigns use internally than most public polls.

You must be logged in to read this quote.

I disagree strongly that it is bad to poll using voter history. Voter history is a much better predictor of actual voting than self-reporting. Granted, one should not automatically screen out an entire category of people (e.g. people who did not vote in 2014), but they are not in fact doing that.

[Zaybay]

You must be logged in to read this quote.

What specifically is wrong with their methodology (as opposed to wrong with how people interpret it before it is actually done)?

their weighing process. Basically, its weighed by 3 factors, census data, education, and voter history. The problem is that mostly the census data, and voter history.
1. The census data means that trends in the district, seen in IL or VA, would not appear, so a district that has seen a shift in demographics would still have its 2010 demographics. Even NYT says that this process isnt even used by most pollsters.
2. Voter history is a big one, as it gets rid of the possible new voters that enter the system. This is a problem that a lot of pollsters have, but usually they move to voters who are likely to vote by now. Siena hasnt.(they have it as a subcategory you can see, which actually shows that WV-03 is even, but its not used in the overall number)

There are other small methology bugs, but these are the two main ones.

I read their explanation of their weighing process on their methodology page and the poll page and I don't see census data or voter history mentioned among their factors.

"In general, we follow a pretty standard set of weights and weight categories. We weight by age, party, gender, region, race, likely turnout and education."

"Here, we’re weighting by age, party registration, gender, likelihood of voting, race, education and region, mainly using data from voting records files compiled by L2, a nonpartisan voter file vendor."

Likelihood of voting is based off of past voting as well as what the person says. If you are to say, "Ill definitely vote" and then respond that you havent voted recently, they wont mark you as a likely voter.

Gonna need a source on that otherwise you are just making an assumption.
[/quote]

You must be logged in to read this quote.

[levgre]

Likelihood of voting is based off of past voting as well as what the person says. If you are to say, "Ill definitely vote" and then respond that you havent voted recently, they wont mark you as a likely voter.

You must be logged in to read this quote.

Your claim doesn't follow from their's.  They don't say how much they weight someone saying they'll "definitely vote" vs past voting history.  For all we know, an answer of "definitely vote" overrides voting history and puts them in the "will vote" category.  They may just look into voting history to augment the "very likely" or lower categories.

[Zaybay]

their weighing process. Basically, its weighed by 3 factors, census data, education, and voter history. The problem is that mostly the census data, and voter history.

1. The census data means that trends in the district, seen in IL or VA, would not appear, so a district that has seen a shift in demographics would still have its 2010 demographics.

Demographic change is a slow process (not actually that much difference between 2018 demographics and 2010 demographics in most places), and in any case it appears most likely that they are using more recent ACS rather than 2010 data (the methodology page mentions the ACS frequently). In addition, for some demographics like age and sex, those are directly in the voter file on the level of the individual voter, so no census data is needed. In some states, individual race data is also available, so in those cases they will be using that, not census/ACS data.

You must be logged in to read this quote.

Which is a deficiency in the methodology used by most pollsters, not a deficiency in their methodology. Their methodology going off the voter file is more like the higher quality polls that campaigns use internally than most public polls.

You must be logged in to read this quote.

I disagree strongly that it is bad to poll using voter history. Voter history is a much better predictor of actual voting than self-reporting. Granted, one should not automatically screen out an entire category of people (e.g. people who did not vote in 2014), but they are not in fact doing that.

You must be logged in to read this quote.

In some districts, you are correct, its too small of a change to actually make a difference, but in others it definitely is. Take the IL districts. These have seen a large influx of more affluent and minority voters, which can be seen in the 2017 census data. But the poll uses 2010 census data, so they are still weighing it by 2010 margins, which is a large problem. Again, it wouldnt be a problem if they just updated the census they were using, but they didnt.

You must be logged in to read this quote.

No, it isnt. Using census data is usually a terrible idea to sumarize an electorate for it leaves out voters who may show up more often, voting age voters, and voters who have passed on and are still in census data. Most pollsters dont use it because it gets a bad result that only works if the electorate is similar to census data, which it never has been. For instance, if we were to use census data in GA, then Hillary should have lost the state by 10 points. But this didnt account for
A. The changing demographics of who was in the state
B. The likelihood of demographics to punch above their weight
Which is why pollsters who didnt use the census were better off than their counterparts, and why most pollsters dont use data that is rather easy to find.

You must be logged in to read this quote.

Sometimes, yes, it is a better idea to use voter history than voter self-reporting. But in midterms, and especially wave years, it isnt. Waves and midterms bring out different groups of people into the voting block, and get rid of others. Using past voter data has proven unreliable in 2006, 2010, 2014, and even this voting cycle. It was voter history that got Northam+3 in the end, and it was voting history that thought Lamb would lose by 5(they switched to a selfID one in the end, which got Lamb+1), not to mention AL, AZ, and OH. Self voting ID was much more reliable in all of these races. Most pollsters know this, and its why most switch over to a self Voter ID after Labour day. Siena doesnt do this, which hurts its credibility.

[Zaybay]

Likelihood of voting is based off of past voting as well as what the person says. If you are to say, "Ill definitely vote" and then respond that you havent voted recently, they wont mark you as a likely voter.

You must be logged in to read this quote.

Your claim doesn't follow from their's.  They don't say how much they weight someone saying they'll "definitely vote" vs past voting history.  For all we know, an answer of "definitely vote" overrides voting history and puts them in the "will vote" category.  They may just look into voting history to augment the "very likely" or lower categories.

Its a Siena poll, and this is what they always do. Im making my assumption based on how they weighed their previous polls, such as NY, and thats how they have done it. Perhaps they switched to accommodate the NYT, but I doubt it.

[Tintrlvr]

The way they are wording the description is strange, but I think what they are doing is taking everyone at their word that they definitely will vote but then also counting a lot of people who say they "probably will"  or "might" vote if they have generally voted in the past. You'll notice that the "definitely will vote" electorate is much smaller than their actual total expected votes for every poll.