Kilmer17s non confrontational election prediction-

[DjTj]

I realize that. But you're making the claim that polls are finding more Dems than Rs, and then using THAT as the sample to weigh. That's not what they do. They already assume the bias in Ind who are really Rs. What they do next though is the key piece. They take their total sample, and THEN apply a mathmatical formula to accurately affect what the pollsters thinks will be the electorate split.

None of the major polls weights by party ID. They weight by demographics. For example, if there are not enough blacks and latinos in their sample, they will overweight their black and latino responses. If there are not enough young people, they will overweight their young voter responses. This often results in a weighting towards Democrats, but it is not weighted to meet a particular party ID split, only to meet a particular demographic split.

You can see this in the numbers that nonniey posted a while back in this thread:

See below

Florida

LIKELY VOTERS........
Weighted UnWeighted
Percent Frequency
PARTY IDENTIFICATION
Republican 30% 352
Democrat 37 350
Independent 29 333
Other/DK/NA 4 38




Virginia
LIKELY VOTERS........
Weighted UnWeighted
Percent Frequency
PARTY IDENTIFICATION
Republican 27% 290
Democrat 35 349
Independent 35 395
Other/DK/NA 3 40

Ohio

LIKELY VOTERS........
Weighted UnWeighted
Percent Frequency
PARTY IDENTIFICATION
Republican 29% 343
Democrat 37 396
Independent 30 334
Other/DK/NA 4 37

http://www.quinnipiac.edu/institutes...ReleaseID=1812

The Florida sample was heavily weighted towards Democrats because minorities were undersampled, as shown in this word document: http://www.quinnipiac.edu/images/pol...2012_demos.doc

But in Virginia and Ohio, the weighted party breakdown closely matches the actual sample because the polls reached a representative number of minorities.

[TDDance234]

I'm going to go with 273-265, Romney. I think Mitt picks up Virginia AND North Carolina (They are going to add their first GOP governor in 28 years), Florida, and Iowa. Obama will still win Ohio... and carry Nevada, but Romney squeaks by in an 8 point victory.

But it could get ugly quick if Obama wins Virginia, and picks up a Florida victory, we could be looking at one of the "landslide" outcomes that have been predicted here. The only way Obama cracks 300 is if Mitt fails to win FL... I don't see Obama winning both VA and NC.

[Heisenberg]

Not a can of worms at all.

Regardless of whether he changes or not, if he is using polls that already overweigh one side, it will produce false results.

Silver predicted 49 out of 50, but while the polls were right about who won, they were more often than not wrong about the margins.

Weren't Silvers predictions often conservative vs. the actual margins in 2008?

[Predicto]

I realize that. But you're making the claim that polls are finding more Dems than Rs, and then using THAT as the sample to weigh. That's not what they do. They already assume the bias in Ind who are really Rs. What they do next though is the key piece. They take their total sample, and THEN apply a mathmatical formula to accurately affect what the pollsters thinks will be the electorate split.

I'm pretty sure that they don't do that, other than Rasmussen. They adjust for likely demographics (which has a similar effect) but they don't adjust for party affiliation.

[Kilmer17]

Weren't Silvers predictions often conservative vs. the actual margins in 2008?

Yes. Which is the entire point. He was using polls in 08 that overweighed the GOP vote turnout based on models from 00 and 04. That's exactly why polls like that can lead to erroneous results.

Again, if todays voting population resembles 08, then it will be an Obama landslide.

[Tulane Skins Fan]

Weren't Silvers predictions often conservative vs. the actual margins in 2008?

Generally, yes. And the one state he was wrong on was Indiana. He had it going for McCain, and it went for Obama.

[Predicto]

I'm going to go with 273-265, Romney. I think Mitt picks up Virginia AND North Carolina (They are going to add their first GOP governor in 28 years), Florida, and Iowa. Obama will still win Ohio... and carry Nevada, but Romney squeaks by in an 8 point victory.

Err, if that happens, Romney still needs 16 EVs. Are you giving him Wisconsin and Colorado too?

[mjah]

None of the major polls weights by party ID. They weight by demographics. For example, if there are not enough blacks and latinos in their sample, they will overweight their black and latino responses. If there are not enough young people, they will overweight their young voter responses. This often results in a weighting towards Democrats, but it is not weighted to meet a particular party ID split, only to meet a particular demographic split.

Exactly! And thank you for expressing it so well.

The polls are measuring voters' preferences in candidates AND voters' preference in party affiliation. Neither of these is treated as hard-and-fast demographic data to be used later for poll adjustment, because they are the measurements. Neither is demographically solid; both are (correctly) assumed to be fluid.

Imagine taking an Ohio poll and then saying "Well, the results are 52-48 in favor of Obama, but I happen to know that the actual voter preference is 51-49 Romney, so I'm just going to just go ahead and shift 3% over to Romney to make this poll 'correct.'" That would be completely asinine. Voter preference in candidates is what you're out to measure, subject to the constraint of getting the hard demographics correct.

Adjusting the same poll results according to some presumed distribution of party affiliation is every bit as unwise. Party affiliation is not a hard and permanent per-person demographic to be treated as an adjustment knob. It is an output, along with candidate preference.

When a site such as unskewedpolls.com does ridiculous things like putting Romney ahead by 5.6 points in Nevada, it is exhibiting a failure to understand the role of party affiliation in polling. It is an output of the process, not a knob.

[Tulane Skins Fan]

Exactly! And thank you for expressing it so well.

The polls are measuring voters' preferences in candidates AND voters' preference in party affiliation. Neither of these is treated as hard-and-fast demographic data to be used later for poll adjustment, because they are the measurements. Neither is demographically solid; both are (correctly) assumed to be fluid.

Imagine taking an Ohio poll and then saying "Well, the results are 52-48 in favor of Obama, but I happen to know that the actual voter preference is 51-49 Romney, so I'm just going to just go ahead and shift 3% over to Romney to make this poll 'correct.'" That would be completely asinine. Voter preference in candidates is what you're out to measure, subject to the constraint of getting the hard demographics correct.

Adjusting the same poll results according to some presumed distribution of party affiliation is every bit as unwise. Party affiliation is not a hard and permanent per-person demographic to be treated as an adjustment knob. It is an output, along with candidate preference.

When a site such as unskewedpolls.com does ridiculous things like putting Romney ahead by 5.6 points in Nevada, it is exhibiting a failure to understand the role of party affiliation in polling. It is an output of the process, not a knob.

Its a knob alright.

[mjah]

Its a knob alright.

I was wondering who would be first to riff on the last line. Congrats.

[Wrong Direction]

In my opinion, people don't seem to understand the logical connection between 3 and 4.

Polls are finding more Democrats when conducting polls. Why? As it has been reported, more traditional Republicans are calling themselves Independents rather than Democrats in recent years. You see that trend even here on ES, with many strong conservative posters consistently calling themselves Independent.

This has two effects: 1) It makes Democrat margins looks crazy high in polling 2) It makes Romney look like he is kicking ass with Independents.

I think this is partially true. The CNN poll had it a national tie, with D+11 but Romney winning Independents +24. Neither will happen. There is some substitution effect, we just don't know how much. We do know it's not a 1:1 relationship, so some R advantage with Independents looks likely.

The answer is the auto bailout. All indicators are that it has played extremely well for Obama in the rust belt. That Jeep ad brought it right back up to the forefront of the conversation just before the election, and it was big news when the company said the ad wasn't true. If Ohio is different today as far as outperforming for Rs vs the rest of the nation, I think that's the reason. Or at least a big part of the reason.

The early voting thing is interesting. Someone (can't remember who) on POTUS radio on XM this morning was saying that what he saw was indeed a much smaller lead for Obama in early voting in Ohio, but largely owing to increased Republican participation in early voting, not a decrease in Dem participation. If that's true (and it would make sense - McCain wasn't nearly so focused on the early vote operation in 2008 as Obama was, and Republicans learned their lesson after last election) then the more narrow lead might be a wash. I haven't seen anything before about polls of election day voters.

Both very plausible. The R answer on early voting is that they didn't target their base, they targeted unmotivated/on the fence voters. I think early voting is in transition from non-existent to very common, and it's really hard to account for changes by simple comparisons to previous elections.

None of the major polls weights by party ID. They weight by demographics. For example, if there are not enough blacks and latinos in their sample, they will overweight their black and latino responses. If there are not enough young people, they will overweight their young voter responses. This often results in a weighting towards Democrats, but it is not weighted to meet a particular party ID split, only to meet a particular demographic split.

You can see this in the numbers that nonniey posted a while back in this thread: http://www.quinnipiac.edu/institutes...ReleaseID=1812

The Florida sample was heavily weighted towards Democrats because minorities were undersampled, as shown in this word document: http://www.quinnipiac.edu/images/pol...2012_demos.doc

But in Virginia and Ohio, the weighted party breakdown closely matches the actual sample because the polls reached a representative number of minorities.

This is also true, but when both parties are holding their party voters at 93% (ish), the party split comes back into play. The reason is that a white D is nearly (statistically) as likely to vote for Obama as a black D.

It all comes down to turnout. I'm nervous.

---------- Post added November-6th-2012 at 04:36 PM ----------

One more thought re: independents...

What if the new independents aren't traditional Republicans at all? What if they're Reagan Democrat types?

So many unknowns today. I'll be happy tomorrow just because this will (hopefully) all be over.

[Lombardi's_kid_brother]

Prediction 1: Obama wins Ohio more easily than anyone expected.

Prediction 2: I will be drunk and yelling obscenities at Fox News pundits by 8:45 CT.

[Popeman38]

Prediction 1: Obama wins Ohio more easily than anyone expected.

Prediction 2: I will be drunk and yelling obscenities at Fox News pundits by 8:45 CT.

Slacker!

[Wrong Direction]

A buddy of mine who is mildly connected to inside people thinks Obama wins VA, WI, OH and IA. If he's right, it obviously won't even be close. He says his polling place was a lot busier today than in 2008, and full of loud Obama supporters. Implication is the enthusiasm for Obama IS there.

Gotta love election day anecdotes. I'm still nervous.

[Tulane Skins Fan]

Just talking to another GOP friend of mine. What I am objectively finding interesting about this is that he is very confident Romney has won DESPITE the polling, and I am very nervous IN SPITE of the polling. His confidence is also making me nervous. But, he's basically saying the same thing I'm seeing in here. The polls are unanimously overrating the D turnout.