Í Wás Wrong. No Mittmentum. Obama Re-elected.

[djAdvocate]

Oct 10, 2012 11:39:10 GMT -5 thyme4change said:

the Tea Party has basically zero reason to vote for Romney, imo.

They have one reason - and that is the lessor of two evils. Even if Mitt isn't what they want, they might think he would be better than the big O.

i didn't say anything about Obama. but since i was apparently unclear, let me highlight it:

the Tea Party has no reason to vote FOR Romney.

[Virgil Showlion]

but to simply say that aggregating is necessarily less accurate than an individual poll (especially one with a KNOWN bias) is categorically false.

If it has a known bias, then it's a biased estimator. The natural question therefore is: of what possible use is a biased estimator?

Considering there's no real theory on how to composite biased estimators (which is equivalent to figuring out the best way of making garbage salad), I'm assuming that each polling company is basing its results on the presumption that its estimates are unbiased. I don't buy into these Paulian conspiracies that pollsters fudge the data in order to manipulate the public.

Thus the only "known bias" comes from the inherent flaws in the modelling, and the best evidence of an unbiased predictor is the long-term performance of the model.

If and only if conditions i and ii in my Reply #40 are met, the estimate of the best-performing model may be improved. If not, the best course of action is to stick with the best known model in terms of past predictive performance. If this model has consistently delivered biased estimates, the logical course of action is to correct for that known bias. I see no reason why a polling company would not already have done this.

Most likely, any biases are because the models are constantly evolving, and the polling companies simply don't know with statistical certainty that their models have any bias. In short, I don't think "known bias" is as known as you seem to think it is.

[AgeOfEnlightenmentSCP]

Polling unit pulls out of Florida, North Carolina, and Virginia because the race is over in those states. Romney has won them.

How long ago was it that I told you Romney had Florida wrapped up?

thehill.com/blogs/ballot-box/polls/261189-pollster-pulls-out-of-fla-nc-and-va-says-obama-cant-win

Suffolk University pollster David Paleologos, whose polls are aggregated into mainstream averages to show where the presidential race stands in the swing states, said he’s finished polling in Florida, North Carolina and Virginia because President Obama has no shot of winning those states.

“I think in places like North Carolina, Virginia and Florida, we’ve already painted those red, we’re not polling any of those states again,” Paleologos said Tuesday night on Fox’s "The O’Reilly Factor." “We’re focusing on the remaining states.”

[djAdvocate]

Oct 10, 2012 12:06:37 GMT -5 Virgil Showlion said:

but to simply say that aggregating is necessarily less accurate than an individual poll (especially one with a KNOWN bias) is categorically false.

If it has a known bias, then it's a biased estimator. The natural question therefore is: of what possible use is a biased estimator?

no, i was talking about "house bias". it has nothing to do with biased estimators. it has to do with bad modeling and bad polling practices that yield consistently faulty results. this is generally not a human error problem, this is a METHODOLOGY problem.

[djAdvocate]

Oct 10, 2012 12:11:11 GMT -5 AgeOfEnlightenmentSCP said:

Polling unit pulls out of Florida, North Carolina, and Virginia because the race is over in those states. Romney has won them.

How long ago was it that I told you Romney had Florida wrapped up?

i think that Suffulk is making a mistake in at least two of those cases. but their decisions are up to them. they have limited resources, and have to deploy them as they see fit.

[Virgil Showlion]

it has to do with bad modeling and bad polling practices that yield consistently faulty results.

Then I ask: if the results are consistently biased in a predictable way, why would a polling company not adjust their model (their "methodology", in your terms) heuristically to compensate?

You've said yourself that these companies are in the business of providing the most accurate estimates possible, and I agree with you. What possible benefit is there in using models with known bias?

[usaone]

As DJ said earlier the only state that matters is Ohio.

[Gardening Grandma]

Oct 10, 2012 11:29:23 GMT -5 Value Buy said:

Guys, I do not want to take anything away from PBP, but please remember, I started a thread on April 21st, called "199 days to go" (currently re-named "27 days to go", where I said Romney would win.
I had the faith.

And the Tea Party is definitly part of this movement.
He may not have been their choice, but we are listening to him.

So how do you feel about his recent swing towards the middle? Are you OK with that since he's just doing it to win votes and you are confident that he really is with you?

[djAdvocate]

Oct 10, 2012 12:19:13 GMT -5 Virgil Showlion said:

it has to do with bad modeling and bad polling practices that yield consistently faulty results.

Then I ask: if the results are consistently biased in a predictable way, why would a polling company not adjust their model (their "methodology", in your terms) heuristically to compensate?

you would have to ask them. Nate Silver has, many times. his calls are never returned. i have my own theories about it, but i am not going to post them, here. they are pure speculation.

the short answer to your question is that most companies do. i will have to assume that what is driving the polling agencies that don't is something other than the relentless desire to produce accurate polls. cost, for example.

You've said yourself that these companies are in the business of providing the most accurate estimates possible, and I agree with you. What possible benefit is there in using models with known bias?

ultimately, none. but there are other factors that drive polling agencies that engender bias. i can think of three without even trying.

[Virgil Showlion]

sure. but there are other factors that drive polling agencies that engender bias. i can think of three without even trying.

I see.

So I'm speaking to "something other than the relentless desire to produce accurate polls" DJ this week as opposed to "there is nothing fudged, Paul" DJ, or "polling agencies actually live and die by accuracy. to think that they are intentionally slanting the results is absurd." DJ.

I admit I have keeping track sometimes.

[AgeOfEnlightenmentSCP]

The polls don't have to be biased to be flawed. As I've pointed out, the chief problem with the legitimate polls in this election is that they do not acknowledge 2010. Even well-respected polling organizations like Rasmussen are weighting their polls to reflect the turnout in the last presidential election. This is a time-tested and reliable method under normal circumstances, but to regard 2008 as "normal" is foolish. It wasn't. The turnout for Obama won't be near what it was in 2008; it'll be closer to a normal election. The accuracy problems are further compounded by the fact that voter party identification is almost the inverse of what it was in 2008, and you have pollsters weighting the polls at Democrat +9 and +11 or even +13. Even in NORMAL election years, it's typical to see Democrat +3 to +6. But this isn't a normal year, as I've mentioned. The polls ought to be even to GOP +1 or +2 to reflect what's really going to happen.

So, when you see Obama up +4 and you subtract the 11 to 15 point flaw in most polls, you can see where we're really at and you start to understand why Suffolk and others are just leaving Florida, Virginia, and North Carolina-- they know it's over.

[Gardening Grandma]

I think it's hysterical that the Right Wing was claiming the polls were flawed when they showed Obama leading. Now they are totally giddy citing the same polls..... ;D

[Virgil Showlion]

Why do you put "Right Wing" in title case?

[djAdvocate]

Oct 10, 2012 12:46:47 GMT -5 Virgil Showlion said:

sure. but there are other factors that drive polling agencies that engender bias. i can think of three without even trying.

I see.

So I'm speaking to "something other than the relentless desire to produce accurate polls" DJ this week as opposed to "there is nothing fudged, Paul" DJ, or

i back that comment up 100%, Virgil. i don't think that any legitimate polling agency INTENTIONALLY fudges numbers. now, when it comes to NOT adjusting for house bias when it is relatively easy to do so, you could ask if this is a distinction without difference, and i can't really answer that.

"polling agencies actually live and die by accuracy. to think that they are intentionally slanting the results is absurd." DJ.

answered below

I admit I have keeping track sometimes.

there are exceptions to everything, Virgil. but i am not even sure that there is an exception, in this case.

just because an agency has the goal of making accurate results DOES NOT MEAN that it will not eschew that cause when it undermines secondary or tertiary goals. imo? it SHOULD. but it doesn't always. i can think of several polling agencies that have serious problems, and one in particular that, even when these problems are pointed out by experts in the field, seem to find no wherewithal to correct it. since they are not commenting on WHY that is the case, i am inclined to think that it is for reasons that would not reflect well on them.

[djAdvocate]

Oct 10, 2012 13:09:51 GMT -5 AgeOfEnlightenmentSCP said:

So, when you see Obama up +4 and you subtract the 11 to 15 point flaw in most polls, you can see where we're really at and you start to understand why Suffolk and others are just leaving Florida, Virginia, and North Carolina-- they know it's over.

rubbish. polls are almost never that inaccurate, with a few notable exceptions, like the 2010 senate race in Hawaii.

[EVT1]

FL and VA are not anywhere close to wrapped up- NC maybe.

11-15 point flaw in most polls

Like another poster said- since it is already in the bag- go ahead and stay home election day and watch some Palin speeches.

[AgeOfEnlightenmentSCP]

Oct 10, 2012 14:07:11 GMT -5 djAdvocate said:

Oct 10, 2012 13:09:51 GMT -5 AgeOfEnlightenmentSCP said:

So, when you see Obama up +4 and you subtract the 11 to 15 point flaw in most polls, you can see where we're really at and you start to understand why Suffolk and others are just leaving Florida, Virginia, and North Carolina-- they know it's over.

rubbish. polls are almost never that inaccurate, with a few notable exceptions, like the 2010 senate race in Hawaii.

We don't know how inaccurate the polls are three and five weeks out. The only way to know how f***ed up they are would be to hold the election THAT day.

Of course, pollsters tighten up when they get down to the short strokes- which is why you're seeing the so-called "Romney surge" now, but I don't think Romney was ever as far behind as they said he was to begin with. I don't think we're seeing a "surge". I think Romney has always been tied, or in striking distance and I think he's started to slowly inch ahead now, and I don't think we're going to see polls bouncing around much anymore. Romney has the momentum, and he will continue to pull ahead.

[AgeOfEnlightenmentSCP]

Oct 10, 2012 14:59:53 GMT -5 EVT1 said:

FL and VA are not anywhere close to wrapped up- NC maybe.

11-15 point flaw in most polls

Like another poster said- since it is already in the bag- go ahead and stay home election day and watch some Palin speeches.

Maybe you should stop laughing and start paying attention:

www.westernjournalism.com/obamas-phony-polls-exposed/

Last Friday the Leftist cheerleaders at the Third Way “think tank” made what had to be a very sad announcement for them. They told the world their researchers had found Democrat registration has fallen by 490,000 in just Ohio alone and without going into finer numbers had also declined by significant numbers in other important states. Florida Democrat registration is down 4.9%, Iowa Democrats have lost 9.5% and remember that “this close” swing state of New Hampshire, the one that flipped its entire government just two years ago – Democrat registration is down 19.7%!

More than this Third Way also found Republican registration was down only a 10th of the Democrat decline and that Independents (who favor Mitt Romney by 14 points) had grown in numbers.


The full story from Ohio must be chilling for Democrats because 44% of these Democrat registration drops come from the Cleveland (Cuyahoga County) area cutting deeply into their lead over local Republicans. Hamilton and Franklin counties, both huge Democrat strongholds have seen steep declines in the number of registered Democrats as well. The combined decline of all three counties accounts for about 79% of the now “missing” Ohio Democrats.

I'm right about this. If you support Obama, you have 27 days to figure out a way to help him. Burying your head in the sand and ignoring the facts won't help.

[djAdvocate]

Oct 10, 2012 15:02:44 GMT -5 AgeOfEnlightenmentSCP said:

Oct 10, 2012 14:07:11 GMT -5 djAdvocate said:

rubbish. polls are almost never that inaccurate, with a few notable exceptions, like the 2010 senate race in Hawaii.

We don't know how inaccurate the polls are three and five weeks out.

you can't poll the future? that's right. this is precisely why your claims of a Romney victory 27 days before the election are rubbish.

[djAdvocate]

Oct 10, 2012 15:11:19 GMT -5 AgeOfEnlightenmentSCP said:

Oct 10, 2012 14:59:53 GMT -5 EVT1 said:

FL and VA are not anywhere close to wrapped up- NC maybe.

11-15 point flaw in most polls

Like another poster said- since it is already in the bag- go ahead and stay home election day and watch some Palin speeches.

Maybe you should stop laughing and start paying attention:

www.westernjournalism.com/obamas-phony-polls-exposed/

this claim is always made by the trailing party, and it is NEVER true. there is no fraud in the polls at all (ok, there are some rare exceptions, but not anything like the CONSPIRACY that you seem to be claiming here), Paul. they are a snapshot of what is happening now. they are not a predictor of the future any more than getting on the Westbound I80 in Sacramento is a predictor that you will end up in Illinois. they ONLY tell you where we are heading.

[djAdvocate]

Oct 10, 2012 15:02:44 GMT -5 AgeOfEnlightenmentSCP said:

Of course, pollsters tighten up when they get down to the short strokes- which is why you're seeing the so-called "Romney surge" now,

totally false. the challenger typically gains during the runup to election because more undecided voters go to the challenger than the incumbent. this has nothing to do with polling methodology, but "human nature" to put off decisions until they are imminent.

but I don't think Romney was ever as far behind as they said he was to begin with.

"as far" Obama never led by more than 4%, Paul. that is CLOSE.

I don't think we're seeing a "surge".

you can think whatever you like. but you are wrong. this is definitely a surge.

I think Romney has always been tied, or in striking distance and I think he's started to slowly inch ahead now, and I don't think we're going to see polls bouncing around much anymore. Romney has the momentum, and he will continue to pull ahead.

doubtful, imo, unless he can capitalize on the last debate. we shall see.

[thyme4change]

I feel the exact same way as when my kids are in the back seat of the car, arguing on whose foot is over the line.

[djAdvocate]

Oct 10, 2012 15:02:44 GMT -5 AgeOfEnlightenmentSCP said:

The only way to know how f***ed up they are would be to hold the election THAT day.

this is true. but we also know that the polls are fairly accurate when the election IS held that day. we ALSO know that methodology of the polls is nearly identical throughout the election season. the ONLY time methodology is adjusted is AFTER elections, Paul- when the data can be matched to the sampling. so your claim about adjustments being made as we inch closer to election is all wet. if you want a pretty rigorous study on this, i can post it.

[djAdvocate]

Oct 10, 2012 15:23:51 GMT -5 thyme4change said:

I feel the exact same way as when my kids are in the back seat of the car, arguing on whose foot is over the line.

sometimes it is pointless to argue. but in this case, i think we can say with absolute certainty that Romney is gaining, and that the gain is attributable to something other than "fudging", as Paul seems to be implying. in fact, he has stated OVER AND OVER again what it is due to, prior to today. the shift in his position is puzzling given the facts, which were clearly on his side before this assertion, and are now wildly different.

[djAdvocate]

there are two things i want to point out here, before i go further.

one. the undecided vote has dropped off considerably since the debate. most of it has gone to Romney. totally normal, except in degree.
two. the estimate swing is 8.5% of the debate margin of victory (also polled) = 4.6%. Obama was ahead 3.2%, according to RCP, before the debate. therefore, based on his performance, we would expect Romney to rise to +1.4%. he is at +1%, as of today.

there is nothing surprising or mysterious or conspiratorial about ANY of this. this is perfectly well established stuff. anyone who is coming unglued about all of this should really get a grip.

[AgeOfEnlightenmentSCP]

Oct 10, 2012 15:21:25 GMT -5 djAdvocate said:

Oct 10, 2012 15:02:44 GMT -5 AgeOfEnlightenmentSCP said:

Of course, pollsters tighten up when they get down to the short strokes- which is why you're seeing the so-called "Romney surge" now,

totally false. the challenger typically gains during the runup to election because more undecided voters go to the challenger than the incumbent. this has nothing to do with polling methodology, but "human nature" to put off decisions until they are imminent.

but I don't think Romney was ever as far behind as they said he was to begin with.

"as far" Obama never led by more than 4%, Paul. that is CLOSE.

I don't think we're seeing a "surge".

you can think whatever you like. but you are wrong. this is definitely a surge.

I think Romney has always been tied, or in striking distance and I think he's started to slowly inch ahead now, and I don't think we're going to see polls bouncing around much anymore. Romney has the momentum, and he will continue to pull ahead.

doubtful, imo, unless he can capitalize on the last debate. we shall see.

The debate is irrelevant to my theory of the election: 2012 is 2010 Part II.

I am still waiting for someone to make a credible argument supported by meaningful data that shows me voters have changed their minds since 2010.

The fact is that this was always going to be an referendum on Obama; and Obama was always set to lose unless something substantially changed. So, if you think Obama is going to win, then by my theory of the election, you have to explain what it is precisely that has changed since the massive, unmistakable repudiation of Obama by the electorate in 2010.

Everything I've seen points to the fact that the anti-Obama trend has picked up steam since 2010.

[Virgil Showlion]

i can think of several polling agencies that have serious problems, and one in particular that, even when these problems are pointed out by experts in the field, seem to find no wherewithal to correct it.

I assume you're talking about Rasmussen, since you've railed against their numbers elsewhere. And you state that their numbers consistently demonstrate bias, which I'll assume is correct.

What sparked our discussion was investorbob's comment that realclearpolitics.com publishes a meta-poll aggregating the results of multiple pollsters. He theorized that this would give him the most accurate results.

I warned him, "You'd think so, but averaging is a poor technique for parameter estimation in many cases." And I suggested, "Your best bet is to find the polling agency with the best predictive track record going back at least fifteen years and stick with their estimate."

Your reply was, "totally. but poll averaging can be every bit as scientific as polling is."

Given the "but", this statement would seem to imply "totally, but if you do the averaging process right, you can get improved results". In this particular case, this statement totally controverts my knowledge of parameter estimation. If an estimator is biased, or if two estimators are assumed to have comparable error and this assumption is false, averaging them can not and will not outperform the superior estimator alone.

I gave the conditions in Reply #40 under which averaging will improve estimator MSE (accuracy), item ii of which is clearly violated unless we assume that all polling agencies have known, significant biases. This, I claimed, is not a reasonable assumption. I believed you would agree with me based on your "polling agencies actually live and die by accuracy" comment, but you apparently don't.

Now you seem to be saying that polling agencies aren't "intentionally" putting out biased numbers, or maybe they are, or maybe they want to publish good numbers but this would undermine secondary goals, or maybe accurate data would embarrass them, but "there is nothing fudged, Paul", except "there are exceptions to everything, Virgil", only "i am not even sure that there is an exception, in this case", and we can soundly conclude that the exception ate cat ate the rat ate the hat that lived in the house that Virgil built.

It sounds to me that you don't believe polling agencies are interested in correcting known biases in their models, that you don't believe they live and die by accuracy, and you do believe that averaging is a useful technique for (literally) averaging out the "we're either too stupid or too embarrassed or too beholden to secret agendas" biases that plague pollsters.

If so, at least this line of reasoning makes sense.

[thyme4change]

So, did we settle this - no need to keep up the charade of campaigning for the next month, because there is only one conclusion, right?

[AgeOfEnlightenmentSCP]

Oct 10, 2012 15:31:59 GMT -5 djAdvocate said:

there are two things i want to point out here, before i go further.

one. the undecided vote has dropped off considerably since the debate. most of it has gone to Romney. totally normal, except in degree.
two. the estimate swing is 8.5% of the debate margin of victory (also polled) = 4.6%. Obama was ahead 3.2%, according to RCP, before the debate. therefore, based on his performance, we would expect Romney to rise to +1.4%. he is at +1%, as of today.

there is nothing surprising or mysterious or conspiratorial about ANY of this. this is perfectly well established stuff. anyone who is coming unglued about all of this should really get a grip.

I agree that there's nothing abnormally conspiratorial about any of this.

I don't claim that the mainstream polling organizations are biased, but I have no doubt that as has traditionally been the case, the media reporting the polls are biased and they tend to pick and choose the salient points that support their position that Obama is going to win the election. The first is that they don't believe Obama will lose, and second because they want to use poll numbers to create a sense of inevitability in order to energize liberals and Democrats and depress conservatives and Republicans.

My problem though doesn't lie primarily with media bias. It goes to methodology.

My theory is simple: the traditional polling methods that look to weight polls according to the turnout in the last presidential election which works in most election cycles. It is traditionally reliable. That's why reputable pollsters use this method. However, it doesn't apply this year. 2008 was an anomaly. Democrat turnout was abnormally high. For a variety of reasons, but abnormal nonetheless. The assumption voter turnout will be similar in 2012 is incorrect. It is safer, in my opinion- my theory of the election- to assume the trend that began to evidence itself in 2010 with more registered Republicans than registered Democrats, and a greater Republican voter enthusiasm and turnout will continue.

I base this primarily on two key bits of data:

Voter identification as Republican vs. Voter identification as Democrat- Advantage GOP. Increased advantage since 2010. Voter enthusiasm GOP vs. Dem- Advantage GOP, by double digits.

The polls are not only not taking this data into account, they are actually weighting their polls with old, irrelevant data that shows the opposite of what is really happening.

[AgeOfEnlightenmentSCP]

Oct 10, 2012 13:13:02 GMT -5 Gardening Grandma said:

I think it's hysterical that the Right Wing was claiming the polls were flawed when they showed Obama leading. Now they are totally giddy citing the same polls..... ;D

The polls always tend to tighten before the race. Before that, they are free to try to sway voters with the false sense of inevitability.

Why do you think that time and again Democrats are always the ones who cannot seem to understand how it's possible for Republicans to win elections when they were behind in the polls for so long?

As we get closer to the elections, pollsters start to become conscious of their own credibility and they tighten up on their methods and predictions.