Pennsylvania Vote Model: Three Key Factors not Considered in the Polls

[tiptoe]

Pennsylvania: A True Vote (not Recorded) Projection Model
TruthIsAll     http://www.geocities.com/electionmodel/PennsylvaniaVoteModel.htm

April 20, 2008

The latest PA polls show the race tightening. The latest 5-poll moving average () has Clinton leading by 47-43. But there are factors which don’t show up in the polls.

This model analysis takes the following factors into account:

1. 10–12% of voters are undecided. Professional pollsters usually allocate 67-90% to the “challenger”. Obama has just four years as a Senator and can be considered as the challenger — while the “incumbent” Clinton has high unfavorable ratings.
   
2. Democratic new voter registration in Philadelphia (strongly Obama). New voters are not polled.
   
3. Cell phone users are mostly younger (Obama) voters, and can't be reached by the pollsters.

One would expect that Obama should do better than his final polling, but...
The true vote, as most of us know, is never equal to the official, recorded vote.
There is no reason to assume that this election will be any different.

These are other factors that the model does not account for:

1. Insufficient polling machines and long lines
   
2. Uncounted votes
   
3. Electronically switched votes

So be forewarned: The following analyses and projections assume a fraud-free election.

Assume the following Base Case scenario:

1. Latest poll: 46-43 for Clinton.
   
2. Undecided voters will break 60% for Obama.
   
3. New urban voters are 60% for Obama and comprise approximately 10% of the electorate.
   
4. Cell phone voters are 60% for Obama and comprise approximately 10% of the electorate.

Given these (conservative) assumptions, Obama wins the Base Case with 51.3% of the vote.

Sensitivity Analysis I (25 scenarios)
Clinton's poll share ranges from 44–48% and Obama's from 43–47%.
Obama wins all 25 scenarios.
Obama's worst case scenario is 50.3%
His best case is 53.7%

Sensitivity Analysis II (20 scenarios)
Obama's share of the undecided vote ranges from 45–80%.
Obama wins 19 of 20 scenarios.
Obama's worst case scenario is 49.1%
His best case is 54.4%.

Base Case Scenario

        Obama    Clinton    Total               
Poll      43       46        89.0                
Undec     6.6      4.4       11.0                
New       6.0      4.0       10.0                
Cell      6.0      4.0       10.0                
 
Total    61.6     58.4      120.0              
Share    51.3     48.7      100.0 

Sensitivity Analysis I

Clinton           Obama % poll share                              
%Shr       43       44       45       46       47
 44       52.3     52.7     53.0     53.3     53.7
 45       51.8     52.2     52.5     52.8     53.2
 46       51.3     51.7     52.0     52.3     52.7
 47       50.8     51.2     51.5     51.8     52.2
 48       50.3     50.7     51.0     51.3     51.7 

Sensitivity Analysis II

Obama             Obama % of Undecided                                  
%New       45       55       60       75       80
 75       51.2     52.1     52.6     54.0     54.4
 70       50.8     51.7     52.2     53.5     54.0
 60       50.0     50.9     51.3     52.7     53.2
 50       49.1     50.0     50.5     51.9     52.3 

[PM7nj]

1. How do you know cell phone users aren't polled...

I've heard this over and over but still nobody has told me how they know this. I think it is just an assumption.

[tekisui]

2. I heard a piece on NPR about it.

I think it's true.

[AllentownJake]

4. They are pulled from Voter Reg rolls

If you put your cell phone down as your primary phone for your voter registration. The data that is being used by the Pollsters is the 2006 voter registration rolls that have been provided to the campaigns as well.

They poll likely voters i.e. registered voters. They don't know new voter registration since 2008 in January.

[OnTheOtherHand]

19. as said below, it depends on the pollster

As far as I can tell, Survey USA does straight RDD; Zogby/Newsmax use registration records. I haven't tried to go down the list.

[tritsofme]

34. Most pollsters use random dialing to get a truly random and representative sample.

To my knowledge it is not related to voter registration rolls.

[OnTheOtherHand]

43. Zogby apparently does use reg lists

"Samples are randomly drawn from purchased voter registration lists of Pennsylvania Democrats."
http://www.zogby.com/methodology/readmeth.dbm?ID=1298

I think that's unusual, but I haven't burrowed through all the methodology paragraphs. (Of course it would be insane to do this if it weren't a closed primary. It's iffy as it is.)

[tekisui]

3. I've been watching that 10-12% Undecided. It hasn't changed.

That is where the upsets and surprises are.

[AllentownJake]

5. To be honest

I expect undecideds to break 6-4 for Hillary. She's the known name in PA. PA voters when they don't know someone well and are undecided tend to break for the incumbent. I hope I'm wrong.

[quantass]

23. That is exactly what i've been thinking..

I wish the OP would update to incorporate the 60-40 break in favor of Hillary for undecideds but keep the other variables the same.

[tiptoe]

29. Have a look at Sensitivity Analysis II (see OP) ...

Here's what it looks like with a 40% column added (which would include scenarios of HRC getting 60% of Undecideds):

Sensitivity Analysis II (24 scenarios)
Obama             Obama % of Undecided                                  
%New       40       45       55       60       75       80
 75       50.8     51.2     52.1     52.6     54.0     54.4
 70       50.4     50.8     51.7     52.2     53.5     54.0
 60       49.6     50.0     50.9     51.3     52.7     53.2
 50       48.7     49.1     50.0     50.5     51.9     52.3 

[OnTheOtherHand]

47. except that the other variables are wrong

Many polls don't systematically exclude new voters, and they are weighted to compensate for the demographic bias of excluding cell phone users. It doesn't make sense to assume that these groups constitute 1/6 of the electorate that is completely excluded from "the polls."

And the poll margins have been all over the place, so it also makes no sense for TIA to grab one result as "the last poll." As I write, the last five polls posted to pollster.com have Clinton +6, +10, +7, +13, and +5. Look a bit further back and you can find -3.

TIA typically makes lots of weird assumptions, then offers to vary one or two of them to show how robust his conclusions are.

(edit to fix typo)

[EmperorHasNoClothes]

6. Polling methodology depends on the pollster.

No doubt some of them don't poll cell phone users and/or newly registered voters, but I don't think that is the case for all polls.

[AllentownJake]

7. They don't have the info

For the new registrations. No one does. Its a long process for the state to update the rolls than make it available to the parties and pollsters. They won't have that info till after the Primary.

[Eurobabe]

54. Polls are all BULLSHIT

I know, I used to do them. :hi:

Thankfully I don't do them anymore.

[featherman]

8. Another X factor - voter enthusiasm/turnout

This may be an uncounted edge for Obama. Let's hope. An upset Tuesday and the Fat Lady sings

[kentuck]

9. This may be one of those elections where all the polls are wrong...

Much like the NH election.

[tiptoe]

15. yep ! nt

[IDemo]

10. The impact of the cell phone 'problem' has been exaggerated

http://www.pollster.com/blogs/cell_phones_and_political_surv.php

[meow mix]

11. K&R this concurs with my analysis

good info!

[NJSecularist]

12. The majority of the undecideds have went to Clinton since March.

They did in Ohio. They did in Texas. I expect the same in Pennsylvania.

The pollsters already account for cell phone users in their polls. It is such a small group anyways that it really doesn't matter either. Some of you are trying to artificially inflate Obama's numbers as wishful thinking, and it just doesn't work that way.

[dsc]

13. they account for such things

cell phone users have been studied and their attidutes are accounted for in these polls. Their views are similar to people of their age who do own landlines. So they simply poll more people that have landlines who are simliar to those with cell phones.

[NJSecularist]

17. Exactly.

Many of the Obama supporters here are trying to artificially inflate Obama's numbers with no basis in anything other than wishful thiking.

[OnTheOtherHand]

18. generally overweights, not oversamples

For pollsters who do Random Digit Dialing surveys, they'll look at the age composition of the respondents they get, and to the extent that their sample has too few young people, they will upweight the younger respondents. If they are using a likely voter model, they apply the likely voter filter (or weights) after getting the population composition they want. (ETA: Pollsters who are working from registration lists could either overweight or oversample, I suppose.)

[housewolf]

14. Impressive analysis

Thanks for posting.

K&R

[cliffordu]

16. Yep, Impressive analysis.

kandr

[ImpeechBush]

20. re: cell phone users

http://arstechnica.com/news.ars/post/20080204-cell-phone-users-and-political-polling-not-much-deviation.html

Not saying this article has to be the final word on the matter but it seems to be pretty much steeped in fact. Conclusions - some polls do poll cell phone users although it is more expensive. Pew says based on it's research that it doesn't make much difference in spite of a demographic that would seem to favor Obama.

[OnTheOtherHand]

40. it's useful to understand why it doesn't matter much

The Ars Technica article is arguably misleading when it says that "cell users have the same political opinions as landline-loving Americans." (And there is a similarly confusing sentence in the Pew report.) What Pew found was that when results are weighted to match national demographics -- as they routinely are -- including cell phone users doesn't make much difference. You get more interviews with younger respondents, but it doesn't much influence the bottom line, because the desired share of young respondents in the weighted sample is predetermined.

(This doesn't mean that Pew decides in advance what share of voters will be young. That would depend on how young respondents fare on their likely voter screen.)

[grantcart]

21. another factor are fence sitters and even Clinton supporters that have decided

that in the best interests of the party its better to stop the campaign now. 70% of all democrats believe that Obama is going to be the nominee

[kmsarvis]

26. GOOD POINT!! THATS WHAT IVE BEEN THINKING!!!

I would imagine there are many Clinton supporters who aren't going to participate tuesday for that reason.

[Sensitivity]

35. Please STOP PRETENDING there is NO RACISM factor. "Undecideds" will vote WHITE in privacy of booth

It is a sad but undeniable phenomenon in states with a significant black population and history
of racial animosity.

[appleannie1]

22. Demographics also play a large part in PA polls.

So a lot depends on where in PA the people called lived.

I have received at least 10 calls from the Clinton camp including one canned one from Rendell.

I have received only 2 calls from the Obama camp.

Hillary is working hard which to me means, she is not all that sure of a win.

[Cosmocat]

49. Robo calls are dirt cheap ...

I live in Central PA, that I know of, we have received two calls from the Hill campaign, and one call and a stop by from the Obama campaign ...

The calls from Hill were for my wife, the call and stopby from Obama were for me ...

Definitely targeting their favorable demos ...

[InfiniteNether]

24. Anybody know what % of counties in PA have electronic voting? My guess is that

whichever ones do have them, those counties will break for Clinton. Whichever counties have paper ballots, Obama will win those. Kinda like NH.

[stillcool]

28. I think they all do. No paper.

http://www.verifiedvoting.org/verifier/map.php?&topic_string=5std&year=2008&state=Pennsylvania

[InfiniteNether]

55. Hey, thanks. This is bad news.

[quantass]

25. Reviewing past Contests...

i notice that despite the idea that cell phones may not be included, it doesnt seem to make that much a difference, if any.
From my studies for patterns in past states it appears Clinton will win by by no more than 12 pts...
Another pattern i noticed is that SUSA's last poll has her at 14pts and when you put the range of +/- 4pts then SUSA is exacting to about 80% of the contests they poll. So from my analysis SUSA seems to be the most accurate...they got it severely wrong in Missouri (way off compared to anyone else).

One other pattern i noticed in the othe contests is that when you have high undecideds they ALWAYS break in large quantities to Clinton which jacks up her final # on the day of voting.....

so as i said above...my final prediction is Hillary by no more than 12pts and as low as 8pts.

[adoraz]

27. Well then, what about Ohio?

As much as I'd love to believe this, how come Obama was polling pretty well in Ohio (even better than PA I believe), but still lost by 10 points?

Are you positive about them not counting the new voters?

[mckeown1128]

30. Some of the pundits today have said that the undecideds

are heavily Hillary type voters. I personally think undecided eds are going to break hard for Hillary.

[Sensitivity]

31. Assumptions have been proven FALSE in this campaign. Undecideds go for CLINTON in this primary.

[tritsofme]

32. lol, it looks like TIA nutter crowd will claim voter fraud if you Obama loses PA.

Its always fun to see their wheels spin.

The "if my candidate doesn't win its fraud" crowd is always amusing to watch.

Don't we have a dungeon made special for these types of things anyways?

[goldcanyonaz]

33. Yep, Diebold!

[bhikkhu]

38. With Rendell in HRC's pocket, that's a given

Expected from the beginning.

[DemocratSinceBirth]

51. They Stand The Scientific Method On Its Head

If reality doesn't comply with my theory then reality must be wrong.

[Alii]

59. I always thought that to be true in my case.

I have two rules:

1. Ron is never wrong.
2. If Ron is perceived to be wrong refer to rule number one.

Sorry, getting stupid...no sleep yet.

[Blue_State_Elitist]

36. Seems to me like the undecideds have broken for Clinton in a number of big contests.

[slinkerwink]

42. You're right. Clinton will win by 12% because of the undecideds.

[Alii]

56. Even if Hillary was to win by 10 to 12 points?

So what? A minor dent in his total pledge count. Obama up by 20+ in North Carolina, 12+ in Oregon, Indiana close, etc., etc. In the end a near wash. In short, Obama IS the democratic party's candidate. We need to get Richardson to Puerto Rico. Why does Puerto Rico have more delegates than many, many of the fifty states? And yet, can't vote in the general election.

But have faith, it's going to be much closer tomorrow than you think. And then...

The fun will disappear when the obvious becomes obvious.

p.s. One very interesting 'poll'...InTrade.

[bhikkhu]

37. K&R - all very good things to know, thanks.

But what's up with the scary avatar?

[boppers]

39. Undecideds tend to break on name recognition.

How much of a name/brand the two parties have built/maintained in the state will make for some interesting numbers.

[slinkerwink]

41. Obama isn't going to get 60% of the undecided voters! Don't be silly!

[yourguide]

58. Agreed, I think Hillary gets around 70% - 80% of the undecideds

It will break like it did in ohio.

[Eurobabe]

44. I think PA will be a surprise for all

Obama will win. Despite the bogus fucking stupid polls.

[Cosmocat]

50. Wishful thinking, and a bizarre analysis ...

in the OP ...

I live in Pa, I am behind Obama and I am freakin tired as heck of Hill at this point, but this is all wishful thinking ...

As many noted, this whole election, undecideds have broken Hill's way, for whatever reasons ... I get the new registered voters, and I get the cell phones, but the OP has taken the most favorable Obama poll to work from, a 5 point loss ... Polls are ranging from that to a high teen Hill win ... I think you have to average them, and put it in the 8-9 range ... Then, if you factor in some of these other things, maybe BO gets it to that five margin ...

I will say this again ... The reason Hill is going to win ... Pa is an old state ... Bottom line, too many old voters ... It isn't racism per se, but they simply will not go with an unstart younger african american over someone with the Clinton brand ...

Obama's campaign has outworked her here, he has outspent her, and there is a substancial college vote that be culled, but all that is going to do is maybe make it closer than Ohio ..

[Eurobabe]

53. Yeah, well I am from Philly suburbs, and when those ALL folks

break for Obama, and they will -- I'll be laughing my ass off. He WILL carry the suburbs of Phila and Pitt and nullify the T-zone "Alabama PA."

edit: BTW, just noticed, no profile, got something to hide?

:hi:

[Cosmocat]

63. Yeah ...

Something to hide ???

Not sure what the point is ... Since you seem to have the time on your hands and at least a tick of paranoia, check my posts ... I am fed up with Hill and behind BO ...

Also, I live in the "T" and if you have actually been to Alabama and you actually live in Central Pa ... Well, a pretty small minded way of catarogizing it ...

I would be as happy as anyone else if BO won and put Hill down, and I will freely admit if I was wrong, heck I will be HAPPY to be wrong ... But, it ain't going to happen ... 5-6 point loss is probably best case scenario ...

[Alii]

57. Obama can afford a single digit loss...even greater.

Hillary cannot. A win for Obama and the bottom falls out of her campaign.

I think the goal posts have already exceeded their limits.

Calculate the pledged delegates...that is the most significant number. Pennsylvania is very interesting.

Go Obama!

[aaroh]

45. undecided is more likely 7:5 to HIllary, making final results 50:50

[Alii]

46. Well, tonight it will all be passé.

The final vote will be the most accurate poll. And now, the next poll...exit polling.

I shall go out on a limb...Obama wins by 1.5%. And then, the pleasing plump lady can begin to warm up her vocal chords.

I shall add one more possible factor...many people might not want their acquaintances to know that they're actually voting for Barack Obama.

Oh well, again, in a few hours it will be all over except for the shouting. Hopefully no confetti for Hillary and Co. Unless it's a win by 5 points or less. If it is this close the delegate count should be very interesting.


Luck!

[OnTheOtherHand]

48. alas, you are a bit ahead of yourself

But there is certainly room for plenty of shouting.

[GoesTo11]

52. Good to recognize things that polls miss

But maybe not such a good job of analysis - since numbers are all pulled out of a hat except for the poll, and that poll was pulled out of a hat full of polls.

Anyway, the factors you mention do make the race more unpredictable and in total, probably increase Obama's chance to make it close.

Another factor - is there going to be a Suppress-the-Vote drive in Philadelphia?

[Onlooker]

60. Thanks for a silly and arrogant post

Polling firms use proprietary methods for predicting votes. They don't just call people and say, this one's for Obama, that one's for Clinton. They use demographic information, historical information, and other pieces of information to make their assessment. In this pretentious article, the author seems to lack any knowledge of how pollsters work. My assumption is that if Clinton leads 46-43%, all the factors cited by the OP have been taken into account. It's silly to think that Zogby, Rasmussen, and others don't factor in new voters, voters who can't be reached, and other info. I would tend to rely on the major pollsters, who use well vetted and highly sophisticated tools developed by people who are more interested in statistics than in politics, then on some off-the-cuff analysis by someone who won't even post his/her name or credentials.

[tritsofme]

61. kick

Just because TIA cracks me up.

[slackmaster]

62. There is only one poll that actually matters

It's called "the election".

[Scurrilous]

64. Thanks for the info.

:thumbsup: