Okay, I can finally prove all you sample weighters are wrong

[Wonkish1]

LOL, so now we're supposed to believe that GOP party ID is as low as during Watergate.

Look analyzing party ID skew in polls is a bit of a crap shoot either way you look at it, but so is expecting Dems to have a D+6 or more spread or GOP a  R +2 or more spread.

Also, the pollsters are admitting that its harder than ever to try to keep these things stable because they're only getting 10% response rates. Polling is very volatile right now with Party ID all over the board and results shooting off in wild areas(all the more reason why the RCP average has been rising in value this election over previous ones).

I think it's reasonable to assess a small shift to at least even(from R) or D +5(from higher D). It shouldn't shift the poll much, but it should make your averages a little more stable(and help to smooth out some noise).

I've mentioned before that I think the self selection in polling responses(10% respond) may be heavily influenced by the moods of certain voters.

[Wonkish1]

LOL, so now we're supposed to believe that GOP party ID is as low as during Watergate.

Look analyzing party ID skew in polls is a bit of a crap shoot either way you look at it, but so is expecting Dems to have a D+6 or more spread or GOP a  R +2 or more spread.

I don't get it.  You have to be trolling.  Did you even read my post?  How can someone read my OP and still try to argue this at all?

Yes, R identity is low because Tea Party members/supporters claim themselves Independent conservatives and because the Rs moved so far to the right to appease these people back in the primaries, a lot of moderates are still staying as Independent.  At the heart of these polls, the ideological background is not only in-line with a 2004 type election, but actually even MORE conservative of a sample compared to 2004.

LOL, so your 'smart analysis' is to try to line up ideology in a single poll? You do realize that ideology is very volatile based upon how the question is asked, right? For example: One of the best pollsters out there Battleground asks that question with a tiered answer system that polls usually ~60% conservative. You can't link ideology. Anybody that has ever studied ideology polling would know this.

What's the term you youngins say again? Oh yeah, epic fail! And its even more hilarious that a few hacks followed you up by clapping.

I'll tell you what, how much do you want to bet that the exit polls will:
A) Feature both parties with party ID in the 30s
B) That the D spread will either be +5 or less.

If you really believe this crap you'll put your money where your mouth is. Loser donates to the other parties national committee and posts the confirmation email(excluding personal info) on the thread. How much do you want to bet?

[Wonkish1]

We need an Intrade market for exit poll DRI breakdown.  I will sell you guys all the D+7 you guys want at 2:1.

That money wouldn't go where those mouths are.

I would sell the hell out of D+7. Apparently I'll have a lot of takers too if we're to believe the Dems on these threads.

And I would likely be willing to risk a ton of capital on it too because I could hedge away any value at risk I didn't want by simultaneously going long Obama win. It would create a perfect arbitrage with the only way I could lose is if Romney won on a D +7 or more.

Come on guys put you're money where your mouth is buy that D + 7 up. This bookie is open for business.

[Wonkish1]

And party self identification isn't volatile? You sir are the fool.

Look at jfern's post to see how consistent that ideology ID has been.

They both are volatile and vary based upon the pollster. Everything inside of these polls are volatile especially when you go from one polling firm/type to another.

It's the ultimate catch 22. It's why you can't even put that much faith into weighting a poll to some pre determined spread either. Some polls have questions asked in such ways as to result in higher party ID or ideology ID spreads and others don't. So you're left wondering what to do. All you do know is that when the exit polls come out it's not going to be D+7 and that both parties will be in the 30s. That's all you do know. You have no idea as to whether or not the party ID in the poll is different because of:
1) Response rates
2) Wording of questions that took another wise good sample and made it seem like a weighted sample
3) Or maybe a different weighting of race or age or whatever ended up boosting the party ID numbers much more than it boosted the actual vote change.
4) You don't know. It could be for 20 different reasons and any of those reasons could have actually impacted the sample itself or just the appearance of the sample.

No matter what you do is going to be a bit of a mess. And polling has only gotten messier over time with cell phones, caller id, net polling, etc.

But at the end of the day, all you can do(if you're going to trust polling) is to hope that by averaging more polls with slightly different methodologies and actively looking into the numbers that you can arrive at maybe some slightly better understanding of the current situation in the race than what any single poll can offer you. That's it!


You're telling me that the annual average for a singular polling outfit polling ideology is 'stable' so that means that a single poll by a different polling outfit has a stable result for that question. Think about that for a second.


But hey what do I know. I guess I must be stupid and easy target to take a bunch of money off of me since your one analysis on one poll is so amazing. So how much will the bet be?

[Wonkish1]

Doesn't this still leave room for discussion of the ideological vote distribution?

What are your numbers for how the 2012 moderate vote splits between Obama and Romney?

Of course it does. The subject matter of analyzing ideology ID is a very valuable addition to study in polls. The only point Im making is that the OP's analysis doesn't prove his thesis for many many reasons. Including volume of past data. The use of only one poll, and then an acknowledgement of its continued limitations of use.

The first thing that needs to be done is to see if any standardization will bring polling al numbers any closer together. If you don't have any evidence that this does happen than the tool would be rendered useless. I have done this with party ID and in the vast majority (but not all) cases the results are brought closer to the mean.

Somebody should perform a similar analysis on this.

[Wonkish1]

Doesn't this still leave room for discussion of the ideological vote distribution?

What are your numbers for how the 2012 moderate vote splits between Obama and Romney?

Yeah, it definitely will just shift over to being the same discussion, just with new substance, so long as everyone is wise enough to read what King has put together here. But those numbers should be much easier to track and speculate on and shift accordingly compared to the highly fickle and unclear Party IDs.

Yup this changes nothing really. Same discussion, now only speculating on moderates instead of independents. It's fancy math-art that really doesn't add much to the discussion although you seem to be peddling this pretty forcefully judging by your links to this thread appearing all over.

For the record, regarding the polling that took place in Alberta in 2012 for our election, and let me say this clearly no amount of reasonable adjustments or corrections to the polling could have predicted the results. None. At all. Adjusting for ideology, party affiliation, demographic turnout, nothing could have predicted the election day results. To match polling with results required taking the highest polling percentage of undecided voters (around 20% although polling ranges were from 3 to 20) and then giving over 80% of those undecided to one party, which no reputable polling firm would do in their right mind.

So bash your heads against the wall with all this nonsense. If anything enthusiasm to vote in my opinion is perhaps the most useful, as I believe it allows for an accurate snapshot of voters. I know many people that give random or deliberately inaccurate information to pollsters and then vote differently or not at all. /endrant

Well, I do honestly believe that the manipulation of these numbers will provide a more accurate picture than does Party ID. I've been posting the link in response to people who keep peddling the obviously useless Party ID crap, looking for their response (to which, I might add, none have had any). Is this perfect? Haaaaaaardly. Is it better? I do believe that it is. But you are certainly free to feel otherwise and I understand why you would. Still, I'm very intrigued to see what the work with these numbers wil produce in the year, and I think we all should be, as at the best it could become a useful and perennial general shifter for better poll accuracy, and at worst we just stop using it.

Again I said that I disagreed with the OPs thesis that ideology ID in a single poll validated party ID sspreads of D +7 or higher. The logic doesn't work. But I do think that ideology ID may be worth looking at. I mentioned a first test above.

[Wonkish1]

Again I said that I disagreed with the OPs thesis that ideology ID in a single poll validated party ID sspreads of D +7 or higher. The logic doesn't work. But I do think that ideology ID may be worth looking at. I mentioned a first test above.

So a single poll is the problem?  Here.  Fine.  I only used one to be nice, but if you'd like to know all the polls I've within the last month who released their Ideology ID crosstabs here:

NPR Poll 9/30
Democratic 37%
Republican 30%
Independent 33%

Liberal 22%
Moderate 37%
Conservative 38%

D+7, but C+16.  Same as NBC/WSJ, 2004, and 2008.

AP-Gfk 9/20
Democratic 31%
Independent 46%
Republican 23%

Liberal 21%
Moderate 34%
Conservative 39%

D+8, but C+18.  In line with NBC/WSJ, NPR, 2004 exit, and 2008 exit.

University of Connecticut 9/18
Democratic 46%
Independent 16%
Republican 38%

Liberal 17%
Moderate 37%
Conservative 36%

D+8, but C+19.  A bad Uni poll, but still in line with NBC/WSJ, NPR, AP-Gfk, 2004 exit, and 2008 exit.

Reason-Rupe 9/21
Democratic 36%
Independent 29%
Republican 28%

Liberal 23%
Moderate 42%
Conservative 35%

D+8, but C+13.  In line with NBC/WSJ, NPR, AP-Gfk, UConn, 2004 exit, and 2008 exit

5 for 5 so far.  I didn't cherry pick either.  These four plus the NBC/WSJ are literally the five most recent national polls on RealClearPolitics which published this data.  I don't have time to continue doing this, but feel free to keep searching for that magical sample of D+7 that doesn't have at least the 2004 split of C+13 on ideological ID.  

Pro-Tip: It probably doesn't exist and if it does it's only one out of so many polls that it's likely an outlier.

A) Those are not all of them in the last month because I know of more.
B) They are not even the 5 most recent that published the data
C) LOL, you say that party ID is more volatile, but look at the numbers you just posted. Party ID was almost near constant in each poll. Ideology was even more volatile(which is what I said it would it would be). There goes the key cornerstone of your thesis.
D) Pollsters have always over polled ideology compared to exit polls and now your doing a direct comparison as if to line them up:
--> Look at the Gallup numbers listed in this thread. They've always consistently year in and year out over polled the conservatives relative to exit polls, and the same is true for Battleground. Now we're to believe that the polling will line up perfectly. Yeah right.
E) There is no question that a ton of Republicans didn't show up to the polls in 2008. Turnout was not the same. So that part of your thesis is incorrect. The country will most definitely be more conservative than the 2008 ideology ID exit polls because every single pollster saw a did in late 2008 for conservatives relative to the past and present...and whether you understand the implications of this or not that was just the knock out to you BS 'thesis'.

[Wonkish1]

I'm not trying to prove that weighing by ideology is a good thing, bro.  I'm saying it's bad just like party ID is.  If Party ID was good, Party ID should shift ideological ID in each election with it yet it doesn't.  More or less, the same kind of people are voting in each election.  My thesis is that more conservatives didn't vote in 2004 compared to 2008.  The same did.  The shift came in moderate vote.

Answer me this: why did the ideology ID remain pretty constant in 2004 and 2008 (21-45-34 to 22-44-34) yet party ID shift so much (37-37-26 to 39-32-29)?  Wouldn't less Republicans mean less conservatives?  Wouldn't more Democrats mean more liberals?  It seems pretty unlikely that two election years of exit polls could produce the same two numbers.  Not to mention 1996, 1992, and 1988 also came up with similar ideology ID to 2004 and 2008.

If Ideology ID is supposedly more violatile, why does appear to be more constant compared to party ID election-to-election?

Lets separate out the pieces shall we?
1) You think that ideological ID somehow pts to D+7 or more being reasonable for a poll. It is most definitely not. You can take my bet if you feel confident about it.
2) You also are stating that they are both have their problems in polling. That I agree with.
3) You then imply that those problems make the head to heads as accurate as they're going to get. I disagree.
4) You also end up making the case that the act of just smoothing some of the edges(not even bringing it to what is largely expected come voting day, but just so its back within reason(D+5 or D+4) is less reliable. I disagree.
5) You implicitly state that party ID weights are useless when analysis has shown that once you even standardize any poll to another poll (D + 8 poll to D + 5 poll or vise versa or any combination) and the question is formatted in a similar way(unlike Fox's opinion research where people have to volunteer that they are an indie) the polls get closer together not farther away from each other(which is small proof that there is some validity to the notion).
6) If you're implying in this post that there wasn't a lot of conservatives that stayed home in 2008 you're incorrect. Also don't forget that in 2004 the exit polls were a little off in favor of the Dems(not that this is could be an entire explanation).

[Wonkish1]

Malia, lets first establish the fact that I'm willing to put my money where my mouth is. You are not. That clearly means that I have a lot more confidence in this than you do. I wonder why?


Data: http://www.gallup.com/poll/152021/conservatives-remain-largest-ideological-group.aspx Clearly showed a drop in the amount of self identified conservatives in 2008 from 2004 and has recovered since then. Other pollsters found the same thing.
--> This means one of 3 things
A) Gallup and these other pollsters were off.
B) The 2008 exit polls were off.
C) The 2004 exit polls were off.
---> Most likely answer: Probably C for the following reasons:
1) They already had issues that year
2) According to all of the other phone pollsters 2004 should have blown the pants off of the races in the 90s and 2008. It didn't which is very odd. This would have the affect of setting a lower bar for the 2008 numbers to be able to tie.
---> This is of course just a guess, but I think most would agree that this is the most reasonable guess. If you have a different theory I'm all ears.

The 4th poll from the top of RCP: http://www.politico.com/news/stories/1012/82389.html

Margin of error works differently when numbers percentages are divided into 3's instead of 2 equal parts.

In the polls you cited the C+ was more volatile than the party ID and margin of error doesn't have much of a roll in pointing that in this particular type of analysis. Now for most polls if I take a D + 5 poll and a D + 7 poll most likely the head to head numbers will converge closer when I either adjust the D + 7 to D + 5 or from D + 5 to D + 7. In the evidence you provided the results go farther apart. That doesn't bode well for the reliability of your new metric.

When you 'research turnout' you should look at the enthusiasm gaps in different years and unless you believe that people's self identified desires to actually show up and vote are meaningless(and they're all going to do it anyway) than that should show you pretty clearly.


Also, you use party ID history from 2 races in history and and ideological ID going back over 20 years. They are both very consistent. The only year where Party ID got out of its normal pattern of D + 3 or less was 2008(which makes complete sense). You seem to think that even though C ideological ID surged in 2010(both in exit polls and in phone polls) that it's going to be the same as 2008. It's not going to be and again I'm willing to put my money where my mouth is.

[Wonkish1]

On turnout,

Exit polls show 34% Conservative in both 2004 and 2008.  I know that's not good enough for you.  There's only one other piece of data to be looked at on this issue: state data on % of registered voters who voted.  I don't believe this tells the story of turnout at all better, but it's the only other thing we have other than "it felt like less conservatives voted."  48 states published this data (Wisconsin and North Dakota exceptions). 

Of the 31 states which voted for George W. Bush, 17 saw an increase in turnout, 13 saw a decrease in turnout and 1 no data reported in 2008.  Of the 22 states which voted for both George W. Bush and John McCain, 12 saw an increase in turnout, 9 saw a decrease, and 1 no data reported in turnout from 2004.   That's an inconclusive result to me. 

It sounds sensible to say "conservatives stayed home in 2008" or even "liberals stayed home in 2004," but is it true?  I have no data to say it is at my disposal and, frankly, my common sense disagrees as well.

Do you really know any politically active people who don't vote?  Even in bad times for their ideology?  I mean, serious politically-active-enough-to-carry-an-ideology-they-call-conservative-or-liberal people who skip elections?  All the conservatives in my life voted in 2008.  All the liberals I know voted in 2010.  All the conservatives and liberals in my life that I know vote in every election because they really do give a sh**t.

Disillusionment happens, but there is nothing to suggest conservatives were disillusioned in 2008.  Sarah Palin drew big crowds.  Fear of Obama was rampant.  34% of the exit polls still said conservatives were voting.  They didn't like McCain but, as far as I can tell, they showed up.  Ideologues always show up.

The only available metric that professionals have used for this is enthusiasm gap. Now you could argue that enthusiasm gap is useless, but you would be arguing with a lot of professionals that claim that it does have good predictive power in determining turnout of each side.

To answer your question, I for one didn't vote in 08(but not for reasons others mentioned in not voting) and I know of many others that didn't vote. I'm also aware of a decent amount of 'somewhat conservatives' that reluctantly cast a vote for Obama in 08. But this is just anecdotal evidence(but you did ask for it).

[Wonkish1]

I don't understand "the money where your mouth is" line.  Did you propose a wager that I missed?

(1)  First off, I asked you to show me some data that I haven't looked at which shows I'm wrong.  You linked a graph that's already on page one.  A las, let's look at it anyway:

That's one poll which shows 2004 to have three percentage points more conservatives than 2008 on a poll with a 2 percent margin of error.  That's not a significant move.  It's inline with the exit polls.  And, if I recall, Gallup had Obama winning by a larger margin than he did, too.  

(2) I've already established that the NATIONAL 2004 Exit Poll was completely accurate in its prediction of a 51-48 Bush win.  A good amount of STATE exit polls showed Kerry outperforming the real result, but as we know, subsamples are more inaccurate.  I believe one pollster in the tracking thread released subsamples of its swing states from it's national poll that had Ohio as Romney +19.  That is no doubt wrong but it in no way discredits the national result that the pollster found, because subsamples are not accurate.  The National Exit Poll has a margin of error of half a point and nailed the overall result.  The worst state exit poll was not off by more than two points in the overall result.  

Exit polls are far more accurate than Gallup telephone polls, even when comparing the former's worst days to the latter's best.

(3) Enthusiasm gap is no more or less real than ideology.  It's asking people what party they support or whom they support (both fluid) and then asking them how "enthused" they are about election (so fluid it's almost a gas).------I'm having a great day today.  I'm very enthused.  I might cry myself to sleep on Thursday night.  Who knows.

(3b) By the way, can I get a link to the expert studies on enthusiasm data?  Or do all you have are opinions of experts with no data?  If the latter, then no, I do not have any reason to be believe professionals are more correct than any joe on the street on this if they have no evidence to backup their case.

Margin of error works differently when numbers percentages are divided into 3's instead of 2 equal parts.

You're absolutely correct!  But absolutely wrong.

Asking all of the sample what their Party or Ideology is does not divide the sample into anything.  It gives you three responses, yes, but so does Romney, Obama, Other and Approve, Disapprove, Neither.  It carries the overall survey's margin of error.

Enthusiasm gap does, however, divide the sample into three subsamples: Are you enthused about the election (Obama supporters only)?  Are you enthused about the election (Romney supporters only)? Are you enthused about the election (Undecided and Third Party supporters only)?  Those all carry higher margins of error.

Ah, funny... Is that you're way of sidestepping a bet?

It's not one poll, damn!!! It's the average for entire year! That tracking polls every day for months plus dozens more for the year. That isn't one poll. READ the footnotes(the annual numbers are over 20,000 polled for off years with a margin of error of less than 1% and on election years the numbers are even higher).

It's not 3 points, it's 5 points. As in the spread between liberals and conservatives condensed by 5 pts (conservatives fell and liberals rose). That is seismic. One of the most seismic moves in politics confirmed by tons of polls averaged over those couple years. Battleground showed the same thing. You had every single pollster without exception show a drop in both party ID and ideological ID in 2008 for GOP and conservatives. And we're supposed to believe that ideology ID didn't change at all and that conservatives showed up at the same level as in 2004. Now you and me both know that without pulling a hundred pages from a book there is no way to actually prove this one way or another(and you can't either), but the Michael Barone's of the world that write political Almanac's breaking down practically every single county in the country would laugh at these notions. They run contrary to everything that has been reported about those elections.

I would say that if every ideology ID phone pollster in aggregate over 2 years shows a dip with a combined sample size potentially over 100k than it's probably pretty accurate. Conservative ID likely dropped in 2008.

Gallup has historically been the strongest pollster in mid term elections for a very long time. Their model takes account enthusiasm gap into its estimates. That is some pretty strong evidence that while it's far from perfect it may be one of the best ways to determine which side will get stronger turnout.


It's calculus as you approach 3 equal divisions(33, 33, 33) the percentage of margin of error falls for each individual category. When you divide it 4 ways(25, 25, 25, 25) the margin of error is even smaller for each. When you divide it 10 ways(10, 10, 10, ...) the margin of error falls into the decimals. The problem with your retort is that very tiny results doesn't have much of an impact on 2 otherwise larger results. So when you see 47, 45 it's mostly a binary division with a higher margin of error for each result.

[Wonkish1]

My only criticism here is in the use of "D+x" terminology, and in a way inconsistent with how the Cook PVI calculates it. That is: If Obama wins a district 60-40 but the nationwide total is 50-50, the district is D+10, not D+20. It just feels off to have it a raw subtraction calculation.

Well it just matters on what you're referencing back against. Cook is calculating based on variation off the mean. We're trying to guess what a reasonable range is for the mean.

My only contention is that the 32 and 39 of 2008 should represent the extreme out bound of what is likely for this election because historically in every single presidential and midterm election for over years the spread has always been D+3 or less minus 2 years. 1 of which was 08 and the other was one year where it was D+4. To expect a duplication of 2008 or worse I think is a bit unreasonable. What is it really? Who knows, but there is likely a reasonable range where it would fall in D + 5 to tie.

[Wonkish1]

Malia, I repeated numerous times, but okay.

Since I personally don't want to give out my personal information let's do it this way: Loser donates to the other parties national convention and then posts the confirmation email on the thread with the personal information removed.

What do you think about $50? The bet is that if the D + spread is 7 or above you win and if the D + spread is 6 or less I win. Deal?

Also anything we can do to keep each other honest(mainly you because I'm not worried about myself going through with it if I lose)? Like maybe agreeing that the Mods are supposed to delete our account and block our IP if we fail to honor?

No it's not 1 poll. Notice how it states(20 polls for 2011) and at the bottom it states that 2011 had more than 20,000 respondents. The margin of error listed there is 1 pt not 2. You're confused. They conduct the poll numerous times over the course of the year and then they publish the average of all of them at the end of the year(Jan 2012 for all of 2011).

Battleground found the similar results. And that is **not one poll** that is an average of all of the polls over each of those years.

Yes you are confused. All I said was that the margin of error for each answer in a poll divided into 2 answers is different than the margin of error for each answer in a poll that is divided into 3 answers. To be specific and not give the wrong impression it's more so true as the percentages approach the 1/n where n is the number of questions asked. A poll question with 3 possible answers, but 1 of which barely registers and 2 of which are much higher has a margin of error for each answer more similar to a poll with 2 possible answers than one with 3. Since, the Party ID and ideology ID feature answers that are much closer to 33, 33, 33 means that those numbers should be closer to their real % than a head to head poll with 2 possible answers. I.e. the margin of error is for each answer is less.

Think about it this way. Let's say I had a poll with 100 different answers in it that each polled at 1%. Do you really think that a margin of error of 3 or 4 pts applies to each answer? No it doesn't. So you think the odds of a single result being off by a multiple of 300 or 400% is likely? No it's not. On the flip side if a result that has 45% has a margin of error of 3 or 4% than if it was off it wouldn't be off by 300 or 400% it would be off by 3/45 = 6% of that singular result.

[Wonkish1]

D+6 or above and I'm in, since D+6 is what I am officially predicting, though there's probably something better we can donate to than a political party, but winner will get to pick.

I like winner picks somewhere to donate to(as long as it doesn't involve giving out personal info to each other).

I'm in for $50. D + 5 or less and I win. D + 6 and you win.

Deal?