Upshot/Siena Voter File FL Poll Clinton +1

[NOVA Green]

Is this actually a real poll?

I read the article, which is quite interesting, and got the impression that this is more a data analytical software forecasting/modelling as used by professional major campaigns.

Did I miss something?

Yes, it's a real poll, but using some of those analytical techniques to select the sample of voters polled, instead of calling random telephone numbers.  The theory is that this will produce a more accurate picture of the electorate.

Well it is an interesting variation of "traditional" polling models, and if that's how they're doing this, it will be interesting to see more of this model, especially closer to the election and see how they score on accuracy.

[heatcharger]

Hillary clearly has a higher ceiling here. Most GOTV efforts won't be captured by this poll because it's based on the voter file from June.

[HillOfANight]

https://twitter.com/Nate_Cohn/with_replies

Reading replies to the poll, I think campaigns tend to poll from the voter file, while cheap (CNN) or national pollsters do random dial (though Fox/Monmouth uses voter file).

Random digit dial could be useful during primaries when lots of new people are entering the process. For November though, I think polling who has been registered since June 2016 is a pretty safe bet (it would probably miss mostly Hillary friendly groups).

RDD polls tend to rely on people's professed intent to vote, while voter file polls tend to do a lot of guess work based on past voting history.

http://www.slate.com/articles/news_and_politics/victory_lab/2011/12/_likely_voters_lie_why_private_campaign_polls_get_such_different_results_from_public_media_polls_.html
This is an old article that goes into it. 55% of people that say they won't vote do, and many that say they will, don't.

I don't think there's much different from Upshot's poll from say Monmouth's, but the maps, analysis, transparency they provide is very welcome.

[afleitch]

Given the above article, one thing I find unusual about US polling is how little pollsters pry into a respondents past voting behaviour, despite it being a very good indicator for future voting behaviour.

[‼realJohnEwards‼]

Given the above article, one thing I find unusual about US polling is how little pollsters pry into a respondents past voting behaviour, despite it being a very good indicator for future voting behaviour.

The thing is, many people don't want to admit that they voted for the loser. This is why LA Times is skewed, for instance.

[SirMuxALot]

Only on Atlas can I learn such things as the fact that being on the wrong side of an enthusiasm gap prior to election day is actually an advantage.

[heatcharger]

Only on Atlas can I learn such things as the fact that being on the wrong side of an enthusiasm gap prior to election day is actually an advantage.

Buddy, she's still leading in the poll you know.

[ApatheticAustrian]

Only on Atlas can I learn such things as the fact that being on the wrong side of an enthusiasm gap prior to election day is actually an advantage.

depends on the size of your potential electorate i guess.

if you are tied and your side is in fact not energized at all...you have room to grow.

[henster]

Clinton only getting 30% of whites in 4-way 35% in 2 way, warning signs for Ds for future election that we haven't reached bottom with whites yet. Obama got 37% in '12 how low can Ds go with whites?

[dspNY]

Clinton only getting 30% of whites in 4-way 35% in 2 way, warning signs for Ds for future election that we haven't reached bottom with whites yet. Obama got 37% in '12 how low can Ds go with whites?

There are almost 20% undecided in the 2016 comparison with Clinton getting 35% of whites whereas the 37% Obama got is forever locked in. It actually shows that if Clinton gets 37% of whites, she wins FL

[HillOfANight]

Given the above article, one thing I find unusual about US polling is how little pollsters pry into a respondents past voting behaviour, despite it being a very good indicator for future voting behaviour.

http://www.nytimes.com/2016/01/26/upshot/bernie-sanders-is-very-dependent-on-infrequent-voters.html

It depends on the kind of election. For example, voter file only polls would have missed Obama's 2008 victory. Sanders/Trump did well in random digit dial polls, while Clinton/Cruz did better among other polls. In the end, Trump won the nomination, and Sanders often outperformed polls.

The advantage voter file polls do have is that you know exactly the demographic of who you're calling and you have a higher success rate of getting polled. And knowing who is who, you can easily oversample to get a low margin of error (rather than poll 10 Blacks, have 1 pro Trump and report it as 10% Blacks for Trump).

In this specific Florida poll, the oversample is useful in getting the true preference of Hispanics and younger voters.

[HillOfANight]

http://www.nytimes.com/interactive/2016/09/20/upshot/the-error-the-polling-world-rarely-talks-about.html?rref=collection%2Fsectioncollection%2Fupshot

Upshot gave 4 pollsters the data they got, and let them reweight it.

Marquette Law (weighted to Census and vote intention): Clinton +3
Echelon Insights (weighted to voter file, history/intention): Clinton +1
Penn Shoen Berland Research (voter file, intention): Clinton +4
Corbett-Davies, Gelman, Rothschild (vote history): Trump +1

A net five-point difference between the five measures, including our own, even though all are based on identical data. In general, the participants who used vote history in the likely-voter model showed a better result for Mr. Trump.

[dspNY]

http://www.nytimes.com/interactive/2016/09/20/upshot/the-error-the-polling-world-rarely-talks-about.html?rref=collection%2Fsectioncollection%2Fupshot

Upshot gave 4 pollsters the data they got, and let them reweight it.

Marquette Law (weighted to Census and vote intention): Clinton +3
Echelon Insights (weighted to voter file, history/intention): Clinton +1
Penn Shoen Berland Research (voter file, intention): Clinton +4
Corbett-Davies, Gelman, Rothschild (vote history): Trump +1

A net five-point difference between the five measures, including our own, even though all are based on identical data. In general, the participants who used vote history in the likely-voter model showed a better result for Mr. Trump.

So Nate Silver pretty much wants clicks when he reweighed this poll to Trump +1. In reality, when you average those out, the poll should have been reweighed to Clinton +2 on 538

[dspNY]

Patrick Ruffini looked at the numbers from the undecided/uncommitted pool of voters in the Upshot poll and found that 52% of them were Dem, 31% indy and less than 20% GOP

https://twitter.com/PatrickRuffini/status/778289855981842433

Far more room for Clinton to grow than Trump in FL

[NOVA Green]

Patrick Ruffini looked at the numbers from the undecided/uncommitted pool of voters in the Upshot poll and found that 52% of them were Dem, 31% indy and less than 20% GOP

https://twitter.com/PatrickRuffini/status/778289855981842433

Far more room for Clinton to grow than Trump in FL

This could well explain the wide variation in recent Florida polling numbers....

[Alcon]

http://www.nytimes.com/interactive/2016/09/20/upshot/the-error-the-polling-world-rarely-talks-about.html?rref=collection%2Fsectioncollection%2Fupshot

Upshot gave 4 pollsters the data they got, and let them reweight it.

Marquette Law (weighted to Census and vote intention): Clinton +3
Echelon Insights (weighted to voter file, history/intention): Clinton +1
Penn Shoen Berland Research (voter file, intention): Clinton +4
Corbett-Davies, Gelman, Rothschild (vote history): Trump +1

A net five-point difference between the five measures, including our own, even though all are based on identical data. In general, the participants who used vote history in the likely-voter model showed a better result for Mr. Trump.

So Nate Silver pretty much wants clicks when he reweighed this poll to Trump +1. In reality, when you average those out, the poll should have been reweighed to Clinton +2 on 538

You're equivocating two different types of adjustments.  The type that the Upshot's experiment does is taking the raw data, and seeing what weighing technique different pollsters do.  The other type is looking at a pollster's past results relative to other pollsters, and adjusting a poll based on what the "house effect" of that pollster seems to be relative to the average poll.  Putting aside that these two types of adjustments are basically apples-to-oranges, it's quite possible that a pollster might have a "house effect" for some reason besides the choices they make during weighting.  It's also possible that a given weighting methodology comes up with more Republican-leaning results in this case, but more Democratic-leaning results in other cases.  Finally, it's possible that the Upshot ended up with three participants (of four) who have weighting methodologies that generate an unusually pro-Democratic result for this poll.  We don't know.

But FiveThirtyEight's "house effect" calculation is entirely comparing a pollster's aggregate results to the aggregate results of the average pollster, adjusted for which states the pollster polled in, and when.  It's not something that involves discretionary input by Nate Silver, as far as I know.  He didn't look at this and manually decide to make it Trump +1.  My assumption is that the pollster's results, when adjusted for time polled, and the "house effect" of the pollster, were predicted to be two points more Democratic than the average pollster.  And that calculation is based on a sample of information larger than the four adjusters the Upshot brought in.  So accusing this of being a bullsh**t clickbait attempt seems unjustified to me.