TRUTHISALL: The 2006 FINAL National Exit Poll does NOT compute - again!

[autorank]

Once again, the FINAL National Exit Poll does NOT compute.
http://www.cnn.com/ELECTION/2006/pages/results/states/US/H/00/epolls.0.html

THE DEMOCRATS DID MUCH BETTER THAN THE FINAL EXIT POLL
INDICATES!

They always do.

Remember Kerry in 2004? 

He won the 13047 respondent National Exit Poll timeline 
at 12:22am by 51-47%.

But the Final Exit Poll (13660 respondents at 2pm) said Bush
won by 51-48%.

Fast forward to 2006.

The 7pm Montana exit poll said Tester won by 53-46.
The Final Exit Poll: 50-47.5
The recorded vote: 49-48

The 7pm Virginia exit poll said Webb won by 53-46. 
The Final Exit Poll: 50.1-49.9	
The recorded vote: 50-49.

What do 2004 and 2006 also have in common?
The Final Exit Poll was matched to the recorded vote.
It always is. That's SOP.
The Democratic vote was 3% too low.

Bottom line:

If the recorded vote was bogus and the election was rigged 
through uncounted ballots and switched votes, you would 
never know it from the Final Exit Poll. 

But if you view the earlier exit poll timeline, you would 
be alerted to fraud. And if you analyze the demographics, 
you would confirm the theft.

Let's start our analysis with the 116 GENERIC PRE-ELECTION 
POLL TREND LINE.  The Democratic vote share has been a 
steadily increasing trend line. 

On Nov. 7, the Dems held a 14.6% lead over the GOP.

Here's graphic proof:
http://www.geocities.com/electionmodel/Election2006_16921_image001.png

The Generic trend line on Nov.7: 
Dem 51.8% - GOP 38.6%
Convert to 2-party shares: 
Dem 57.3% - GOP 42.7%
That's a 14.6% spread.
We will refer to the 14% spread in the following analysis.
_________________________________________

Lets look at the 2006 NATIONAL EXIT POLL, posted on CNN:

PARTY-ID			
.....	Mix	Dem	Rep
Dem	38%	90%	9%  C 38% too low, 90% too low, 9% too high
Rep	38%	7%	93% C 38% too high, 93% too high, 7% too low
Ind	25%	49%	46% C 49% WTF! Independents voted 60/40 for Dems

Total	101%	49.1%	50.3% C WTF! Are they serious?

IT'S PROOF THAT THE 2006 FINAL EXIT POLL IS BOGUS:
According to poll, the GOP won by 50.3-49.1%. 
Really?

1) 2006 Voters identified as 38% Democratic, 38% Republican,
25%
Independent.

THIS IS ABSOLUTELY IMPOSSIBLE !

Here's why: 

a) The weights don't sum to 100. 
OK, no big deal here.

b) Dems outnumbered Repubs. 
Who was more motivated to vote this time?

c) The weights were 38D-35R-27I at the 12:22am 2004 NEP
timeline. 
Look it up.

d) THE CLINCHER: 
The 2006 vote based on PARTY-ID weights/vote shares 
are IMPOSSIBLE! If the weights/shares are to be 
believed, then the GOP won the Generic vote! Why, 
then, would you believe them?

The NEP UNDERSTATES the Democratic Generic vote share by 7%.

It OVERSTATES the GOP Generic vote share by 7%.

How do we know this?

Simple. The Dems won the final Generic Polls by more than 14%!

Since 2004, the Final NEP has become laughable, a sick joke.
Don't they realize they can't fool us anymore?
Don't they realize that we can crunch the numbers?
Would someone please get this to Olbermann?

2006 NATIONAL EXIT POLL 
Sample 13208 MOE 0.87%
Weights/shares adjusted to derive a 12% Democratic Generic
spread 

PARTY-ID (adjusted)			
Dem	40%	93%	7%
Rep	35%	11%	89%
Ind	25%	60%	40%

Total	100%	56.1%	44.0%

That's more like it!
________________________________________________

HOW VOTED IN 2004
Using this demographic, the spread is 55.8 Dem-44.2 GOP. 
That's an 11.6% spread. But it's too low. Why? 
Because the Bush/Kerry/Other weights are bogus. 
Kerry won by 52-47%. The third party vote was 1%.

This is an analysis of how impossible Final Exit Poll weights 
were used to match a corrupt 2004 vote count: 
http://www.geocities.com/electionmodel/BogusWeights.htm

2006 NATIONAL EXIT POLL 
Sample 13208 MOE 0.87%

HOW VOTED in 2004
........Mix Dem GOP
Kerry   45% 94% 5%  
Bush    46% 13% 85% 
Other    5% 62% 21% 
No       4% 79% 18%

Total 97.6% 54.5% 43.1%
2-pty  100% 55.8% 44.2%

Now let's adjust the weights and vote shares to 
derive a 14% spread. 

We use 51 Kerry/46 Bush/1 Other/2 No weights.

The adjusted weights are based on the TRUE Kerry/Bush 
vote BEFORE it was stolen with uncounted spoiled/lost 
ballots and vote switching. 

2006 NATIONAL EXIT POLL 

VOTED IN 2004
(adjusted weights and vote shares)			
........Mix   Dem    GOP
Kerry	  51%	 94%	5%  C Kerry's true 2004 vote
Bush	  46%	 13%	85% C Bush's true vote
Other	   1%	 62%	21% C Third parties had 1% of the vote
DNV	   2%	 79%	18% C did not vote in 2004

Total	98.3%	 56.1% 42.2%
2-pty	 100%  57.1% 42.9%

The adjusted Democratic 2-party national vote 
share is now 57.1%.  That's within 0.2% of the 
Nov.7 trend line (see above).

_______________________________________________________________

GENDER 
Based on the 2006 National Exit Poll 2-party vote shares,
the national split was 54.4% Dem-45.6% GOP. 
That's an 8.8% spread. Much too low.

2006 NATIONAL EXIT POLL 
GENDER
(adjusted weights and vote shares)			
.......Mix Dem GOP
Male   48% 51% 47%
Female 52% 56% 43%

Total 98.5% 53.6% 44.9% 
2-pty  100% 54.4% 45.6%

Ask these questions, regarding national vote shares: 

WHY THE 2.8% DISCREPANCY BETWEEN "GENDER" AND
"HOW
VOTED"? 

WHY THE 5% DISCREPANCY BETWEEN "GENDER" AND
"PARTY-ID"?

Once again, let's adjust the weights and vote shares to get a
result which
approximates the Generic vote.

GENDER (Adjusted) 			
.......Mix Dem  GOP
Male 	46% 53%  47%
Female 54% 57%  43%
Total 100% 55.2% 44.8%

This is just further confirmation that the Final 
2006 NEP was matched to a corrupt vote count, 
just as it was in in 2004 and 2000.

Edison-Mitofsky never consider the possibility 
of a corrupt vote 
count in discussing their exit poll methodology. 

WHY DO THEY DO THIS?
WHY DO THEY ALWAYS ASSUME ZERO FRAUD?
WHY DO THEY ALWAYS ASSUME A PRISTINE VOTE COUNT?

THAT'S WHY THE FINAL NATIONAL EXIT POLLS ARE ALWAYS WRONG.

THAT'S WHY THE FINAL EXIT POLLS ALWAYS LOW-BALL THE DEMOCRATIC
VOTE.

THAT'S WHY THE FINAL EXIT POLLS NEVER MATCH FINAL PRE-ELECTION
POLLS.

THAT'S WHY THE EARLY, UNCONTAMINATED EXIT POLLS ARE CLOSE TO
THE TRUTH.

AND THAT'S WHY THEY'LL NEVER SHOW US RAW EXIT POLL DATA.
________________________________________________

[Name removed]

1. Deleted message

Message removed by moderator. Click here to review the message board rules.

[autorank]

25. Live links to graphs from the OP - (the OP is in plain text)

http://www.cnn.com/ELECTION/2006/pages/results/states/US/H/00/epolls.0.html

[skipos]

2. Yet, PRE-elections polls in 2006, as in 2004, were pretty consistent with the actual results

http://www.realclearpolitics.com/epolls/writeup/election_2006_senate_scorecard-61.html

[OnTheOtherHand]

10. or arguably Dems got the late surge

http://www.pollster.com/mystery_pollster/first_impressions_a_good_day_f.php

In the twelve most competitive Senate races, Democratic candidates improved upon the Pollster.com averages by about two points.

Oh well. It's fun arguing with TIA, but I'm trying to spend my limited time actually looking at suspect vote counts.

[autorank]

22. Take a look at this...

It has pre election polls and outcomes for 62 races, Democrat versus Republican incumbbent.

http://www.electionfraudnews.com/Elections/2006WatchPostEl.1.htm

[autorank]

24. Skipos...a response just for you. Kerry lead the pre-electoin polls.

TRUTHISALL:

http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=203&topic_id=458020&mesg_id=458024

Skipos,

That is a myth. Kerry was leading the pre-election polls. You
are looking
at RCP, which is just showing LV polls. 

Final 18 Poll Summary:									
Kerry won 11, Bush 6, 1 tie				
Kerry won 5 of 9 Registered Voter (RV) Polls
and 6 of 9 Likely Voter (LV) Polls																		
											
Polling Data Source:
http://www.pollingreport.com/wh04gen.htm
http://www.economist.com/media/pdf/YouGovS.pdf						

											
		Total	Poll	Total	Weighted Average	 Moving Average 
Projection	
	    Sample	Sample	MoE	KERRY	BUSH	      KERRY	BUSH	KERRY	BUSH
Date	      26961	Type	0.60%	47.80	47.14	       6-period	 50.71
48.29
											
2-Nov	Harris	1509	LV	2.52%	50	47		48.5	47.5	50.98	48.02
2-Nov	Zogby	1200	LV	2.83%	50	47		48.2	47.2	51.69	47.31
1-Nov	Marist	1166	LV	2.87%	50	49		47.8	47.2	50.32	48.68
1-Nov	Econo	2903	RV	1.82%	50	47		47.5	46.7	51.06	47.94
1-Nov	TIPP	1284	LV	2.73%	44	47		47.0	46.8	48.81	50.19
											
1-Nov	CBS	1125	RV	2.92%	47	48		47.5	47.0	49.97	49.03
31-Oct	FOX	1400	RV	2.62%	48	45		47.8	47.0	51.73	47.27
31-Oct	DemCorp	1018	LV	3.07%	48	47		47.5	47.0	48.82	50.18
31-Oct	Gallup	1866	RV	2.27%	48	46		46.8	47.2	51.44	47.56
31-Oct	NBC	1014	LV	3.08%	47	48		46.8	47.5	49.48	49.52
											
31-Oct	ABC	3511	RV	1.65%	47	48		47.0	47.3	50.11	48.89
30-Oct	ARG	1258	LV	2.76%	49	48		46.8	47.8	51.49	47.51
30-Oct	Pew	2408	RV	2.00%	46	45		46.8	47.5	52.98	46.02
29-Oct	Newsw	1005	RV	3.09%	44	48		47.0	48.0	49.08	49.92
26-Oct	ICR	817	RV	3.43%	48	48		47.8	48.0	48.35	50.65
											
24-Oct	LAT	1698	RV	2.38%	48	47		47.7	48.0	51.60	47.40
21-Oct	Time	803	LV	3.46%	46	51		47.5	48.5	46.49	52.51
20-Oct	AP	976	LV	3.14%	49	46		49.0	46.0	53.39	45.61
											
											
					
											
											
	BUSH	KERRY									
											
			Zogby Poll								
1 LV	47	50	10/4-31/04 REUTERS/ZOGBY TRACKING POLL: 3-day
rolling sample of
approx. 1,200 likely voters nationwide. MoE ± 2.9. 								
2-Nov					Bush	Kerry	Nader	Other			
				10/29-31/04	48	47	1	4			
											
2 LV	49	50	Marist College Poll. Nov. 1, 2004. N=1,166
registered voters
nationwide (MoE ± 3); 1,026 likely voters (MoE ± 3).								
					Bush	Kerry	Unsure				
				11/1/2004	49	50	1				
											
											
3 RV	47	50	Economist  YouGov 		2903	total; MoE +/-2%					
				10/30-11/01							
					Bush	Kerry					
					45	49					
											
4 LV	47	44	TIPP tracking poll conducted by TechnoMetrica
Market
Intelligence. Oct. 30-Nov. 1, 2004. N=1,284 likely voters
nationwide. MoE ±
2.8.   						
					Bush	Kerry			
				10/30 - 11/1/04	47	44			
5 RV	48	47	CBS News Poll. Oct. 29-Nov. 1, 2004. N=1,125 likely
voters
nationwide. MoE ± 3. 						
					Bush/	Kerry/			
					Cheney	Edwards			
				10/29 - 11/1/04	48	47			
									
6 LV 	47	50	The Harris Poll. Oct. 29-Nov. 1, 2004: N=1,509
likely voters
nationwide who express a preference. MoE ± 2.5.						
2-Nov					Bush	Kerry	Nader	Other (vol.)	
				10/29 - 11/1/04	49	48	2	1	
									
7 RV	45	48	FOX News/Opinion Dynamics Poll. Oct. 30-31, 2004.
N=1,400
registered voters nationwide (MoE ± 3); 1,200 likely voters
(MoE ±
3).						
					George	John	Other 	Wouldn't	
					W. Bush	Kerry	Not Sure	Vote (vol.)	
				10/30-31/04	45	48	7	-	
									
8 LV	47	48	Democracy Corps Poll conducted by Greenberg Quinlan
Rosner
Research (D). Oct. 29-31, 2004. N=1,018 likely voters
nationwide. MoE ±
3.1.						
					George	John	Ralph	Other	Unsure
					Bush	Kerry	Nader	(vol.)	
				10/29-31/04	47	48	1	1	3

9 RV	46	48	CNN/USA Today/Gallup Poll. Oct. 29-31, 2004.
N=1,866 registered
voters nationwide (MoE ± 3); 1,573 likely voters (MoE ±
3).						
					Bush/	Kerry/	Nader/	Other	None/
					Cheney	Edwards	Camejo	(vol.)	Unsure
				10/29-31/04	46	48	1	1	4

10 LV	48	47	NBC News/Wall Street Journal Poll conducted by the
polling
organizations of Peter Hart (D) and Bill McInturff (R). Oct.
29-31, 2004.
N=1,014 likely voters nationwide. MoE ± 3.1.						
					Bush/	Kerry/	Nader/	None/	Unsure
					Cheney	Edwards	Camejo	Other (vol.)	
				10/29-31/04	48	47	1	2	2
									
11 RV	48	47	ABC News Tracking Poll and Washington Post
Tracking Poll.
Rolling sample. Fieldwork by TNS. ABC News and The Washington
Post share
data collection for this tracking poll, but calculate and
report the
results independently. WASHINGTON POST: Oct. 28-31, 200						
					Bush/	Kerry/	Nader/	None/	No
					Cheney	Edwards	Camejo	Wouldn't	Opinion
			ABC News Tracking Poll 						
				10/28-31/04	48	47	1	2	2
									
									
12 LV	48	49	American Research Group Poll. Oct. 28-30, 2004.
N=1,500
registered voters nationwide (MoE ± 2.5); 1,258 likely voters
(MoE ±
2.8).						
					Bush/	Kerry/	Other/		
					Cheney	Edwards	Unsure		
					48	49	3		
13 RV	45	46	Pew Research Center for the People & the Press
survey
conducted by Princeton Survey Research Associates
International. Oct.
27-30, 2004. N=2,408 registered voters nationwide (MoE ± 2.5);
1,925 likely
voters (MoE ± 2.5).						
					Bush/	Kerry/	Nader/	Other/	
					Cheney	Edwards	Camejo	Unsure	
				10/27-30/04	45	46	1	8	
									
14 RV	48	44	Newsweek Poll conducted by Princeton Survey
Research Associates
International. Oct. 27-29, 2004. N=1,005 registered voters
nationwide (MoE ±
4); 882 likely voters (MoE ± 4).						
					Bush/	Kerry/	Nader/	Other (vol.)/	
					Cheney	Edwards	Camejo	Undecided	
				10/27-29/04	48	44	1	7	
15 RV	48	48	ICR/International Communications Research poll.
Oct. 22-26,
2004. N=817 registered voters nationwide (MoE ± 3.4); 741
likely voters
(MoE ± 3.6).						
					Bush/	Kerry/	Other	Neither	Unsure
					Cheney	Edwards	(vol.)	(vol.)	
				10/22-26/04	48	48	-	1	4
									
16 RV	47	48	Los Angeles Times Poll. Oct. 21-24, 2004. N=1,698
registered
voters nationwide (MoE ± 3); 881 likely voters (MoE ±
3).						
					Bush/	Kerry/	Unsure		
					Cheney	Edwards			
				10/21-24/04	47	48	5		
									
17 LV	51	46	Time Poll conducted by Schulman, Ronca &
Bucuvalas (SRBI)
Public Affairs. Oct. 19-21, 2004. N=1,059 registered voters
nationwide (MoE
± 3); 803 likely voters (MoE ± 4). 						
					Bush	Kerry	Nader	Unsure	
				10/19-21/04	51	46	2	1	
									.
18 LV	46	49	Associated Press-Ipsos poll conducted by
Ipsos-Public Affairs.
Oct. 18-20, 2004. N=1,330 registered voters nationwide (MoE ±
2.5); 976
likely voters (MoE ± 3).						
					Bush/	Kerry/	Nader/	Other/	
					Cheney	Edwards	Camejo	None (vol.)/	
				10/18-20/04	46	49	2	3	

[OnTheOtherHand]

26. and why limit ourselves to likely voters? n/t

[Cooley Hurd]

3. K&R - the numbers are there...

...does not compute.:grr:

[DemBones DemBones]

4. Anybody here surprised?

:shrug:

[No DUplicitous DUpe]

5. This Preliminary exit poll data from 2004 was all the proof I needed

to confirm Kerry won by at least 52%-48%. And thank TIA...He is my Hero!
http://www.wasingtonpost.com/wp-srv/politics/elections/2004/graphics/exitPolls.html

[understandinglife]

6. Recommended.

NOT ONE LINE OF SOFTWARE BETWEEN A VOTER AND A VALID ELECTION.

[mom cat]

31. Absolutely. Only paper and a million sets of eyes!

[tridim]

7. This also explains the "races tightening" pre-election

Races don't tighten from 10% down to 1% down in one week, for no apparent reason like they did this year.

[Joe Chi Minh]

57. Why do I have the suspicion that the MSM "prepared the way"

for this strange tightening from a vast disparity to a very close race? I remember the MSM, even here in the UK, where the same dark forces of corporatism have always been hand-in-glove with the neocons, being most reluctant to consider the possibility of the Democrats winning the Senate.

I felt that was their game the moment they started on about those putative "close races", particularly in Virginia and Montana. Strange they were the locations of the key Senate races, isn't it? They must have thought "At least we can salvage those two for the Senate by fraud".

[pa28]

8. We need real election standardization and reform.

Forget Diebold for a minute. Republican suppression techniques, Democratic voter purges and voter intimidation are a matter of record.

This is too important to put on the back burner.

[The_Warmth]

9. Great work!

K/R this one folks.

[Kurovski]

11. Done. And bookmarked.

:kick:

[Febble]

12. Well, I can answer some of these questions

And indeed I did in this DKos post:

http://www.dailykos.com/storyonly/2006/11/4/135126/905

The reason "they do this" is because they are trying to project the result, not act as an independent check on it. However, as in 2004, the process was quite transparent, and announced in advance. Contrary to assertions here on DU before the elections, the "unadjusted" cross-tabs were not suppressed; exit poll results were released after 5.00pm but before close of polls; CNN posted the unadjusted cross-tabs (which were not intended to be projections, and they remained unadjusted for some time, allowing any one who wanted to download them. Hidden in plain sight, you might say.

So whatever anyone thinks about the reasons for the discrepancies between the unadjusted cross-tabs and the final cross-tabs, can we please STOP the insinuations that anybody involved in the exit poll operations, including both the pollsters and the networks is deliberately trying to cover something up? Now, surely is time for a few around here to eat crow as regards their allegations that somehow in 2004 something untoward happened and Mitofsky, or whoever, was coerced into faking the results post hoc and lying about what probably caused the discrepancy.

The exit poll process is complex, but transparent, and nobody covered anything up this year, just as nobody covered anything up in 2004, and just as nobody covered anything up in 1992, when there was also a large adjustment, but nobody much noticed as Clinton won anything.

And now how about we get some decent numbers. I'm not sure exactly when TIA downloaded what, but OTOH was downloading everything at close-of-poll in each state (i.e. when all but the odd stray call should have been in, but before any adjustment in line with incoming returns). I posted his preliminary list of deviations between exit poll and results here, although I hope eventually E-M will release, as they did in 2004 their actual "Best Geo" estimates - estimates made solely from responses, and not adjusted to pre-election expectations. But for now, these give us a ball-park idea of the extent of the deviation, and, as in 2004 the count appears to be significantly "redshifted" from the poll. However, Montana and Virginia do not particularly stand out, and the distribution is normal (no outliers).

http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=203&topic_id=457571&mesg_id=457723

I'd also point out that a poll just before the election found that Republican voters were significantly less likely say they would be willing to participate in an exit poll than Democrats

http://www.pollster.com/mystery_pollster/the_overlooked_exit_poll_quest.php

so it would be idiotic to assume that there would be no bias in an exit poll conducted shortly afterwards.

In other words, there is absolutely nothing you can conclude about the location of fraud from these exit poll deviations. There might well have been fraud in the election. Sarasota County, Florida, looks horrible.

So how everyone stops staring at exit polls like tea-leaves in the hope of seeing what they want to see, and actually starts looking at some pretty egregious evidence of races lost due to, at best malfunction, and at worst, outright fraud?

http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=203&topic_id=457470&mesg_id=457470

http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x457990

[Sancho]

40. Which is something we've discussed before...where is the precinct data?

Sarasota in 2006, just like the Castor/Martinez race in 2002, etc. would benefit from poll evidence that elections were fixed. Without the data in a timely form and by precinct, it's all speculation.

I assume that smart pollsters would learn problem areas over the years and put the really experienced interviewers at the hot spots, record all the demographics of anyone entering the voting area, and other things so that there would be as accurate a result as possible in those expected races. Why would E-M not do that?

Regardless of poll variability, TIA makes a point with the demographics and trends being unlikely in some cases. It may not "prove" fraud, but the pollsters could really help provide evidence if they wanted to...

[autorank]

44. Sancho, you make a very good point..."if they wanted to."

The pollsters could help find fraud IF THEY WANTED TO. But who knows what they want and someone soon will say who cares. John Conyers asked for the raw data from the exit poll and there was a gnashing of teeth here about confidentiality. What a joke. Precinct data can't be released. Guess what, it was in the 2000 NEP. The data has to be blurred blablabla. It's all a cover to keep the data locked down.

Well Conyers has subpoena power now. That means that the networks will produce the data or they'll be in contempt of Congress. Reid was ready to go after ABCs license on the 911 documentary. Imagine how the Committee on the Judiciary will react when these very unpopular institutions, the networks and papers in the Media Consortium that paid for the exits, try to drag their feet.

CONTEMPT OF CONGRESS. Go ahead Network Consortium or EM...try to resist a subpoena from the House Committee on the Judiciary.

[OnTheOtherHand]

51. a question of scale

Look, people, the national exit poll is conducted in 1000-1500 precincts -- an average of 20-30 per state. (I think it was about 1000 precincts this year.) In Florida they might have had 50 precincts. Florida has 67 counties.

Expecting a national exit poll to document problems in Sarasota County is like trying to catch a pollen grain with a tennis racket. Wrong tool for the job.

[DemReadingDU]

13. Did other states have thier final exit poll fixed?

What about Ohio? Democrat Sherrod Brown won, not incumbant Republican Mike DeWine. But did Brown win by a larger margin than than the poll indicated?

[applegrove]

14. Don't you think the early exit polls are day people and the ending ones

include all those who work. They are separate pictures of the population. The days probably have more poor, soccer moms & elderly voting. The evenings...working people. We would expect a different picture of the exit polls.

[Febble]

18. Not exactly

the cross-tabs at close-of-poll should include all responses apart from the odd straggler. It's the difference between these close-of-poll cross-tabs and the final ones (or the final results) that tell you roughly how "off" the poll (or the count) was.

[OnTheOtherHand]

28. just one more thing about that

The first national House tabulations went up at, I think, 7 PM Eastern, so those surely wouldn't include late respondents from much of the country. But the state tabs all were posted at close of poll, so they would.

EDIT TO ADD: And I certainly don't think that early vs. late accounts for the red shift in that first House tab, especially since it obviously doesn't account for the red shift in the Senate races. Nor do I think it is a fraud signal.

[Febble]

34. Thanks

Yes, of course, you are right. I was thinking of the Senate polls.

[rumpel]

15. k&r

:kick:

[zeemike]

16. Perhaps that is why so many close races

And yet no Republican looser like Burns or Allen asked for a recount.

[progressoid]

17. Um......What?

I am very suspicious of election integrity and exit polls, but the jumble posted above does nothing to confirm or deny those suspicions. What a cluster-fuck of assumptions, and supposition.

[elehhhhna]

19. Nice talk. Read the op again, for comprehension.

Opinions are your own, but FACTS belong to everyone.

[OnTheOtherHand]

27. maybe TIA could use a fact-checker

You can check the (weighted) 2006 national exit poll tab here:

http://www.cnn.com/ELECTION/2006/pages/results/states/US/H/00/epolls.0.html

Where on earth did he get 49.1% D, 50.3% R?

Oh-kay.

[progressoid]

30. I read it. It's INcomprehensible.

But if you want to discuss opinions vs facts, fine.

Here are just a couple of the many problems in the OP:

GENDER
Based on the 2006 National Exit Poll 2-party vote shares, the national split was 54.4% Dem-45.6% GOP. That's an 8.8% spread. Much too low.

The poster says "That's an 8.8% spread. Much too low." Really, How does he/she know? Where are the facts to back up that statement. It's an OPINION.

Then there is this:
Once again, let's adjust the weights and vote shares to get a result which approximates the Generic vote.

WHAT? The poster wants to adjust the numbers? He/she is fudging the numbers to fit the argument - ironically, the same thing he/she accuses the NEP of doing! Earlier, he/she says it is to "Convert to 2-party shares". Apparently it's OK to completely disregard independent voters responses??.

This is so convoluted and fractured, that it becomes meaningless.

Yes, the voting system is broken. But this use of circular reasoning, innuendo, and straw men arguments are not going to help the cause.

[Febble]

33. It is also

idiotic to assume that there can be no bias in the poll, particulary in the face of actual evidence that there would be (not to mention plenty of evidence that there has been in the past)

http://www.pollster.com/mystery_pollster/the_overlooked_exit_poll_quest.php

[ProgressiveEconomist]

20. What do you say about the 'increasing nonresponse bias' alternative

.explanation for higher Democratic percentages in exit polls than in actual votes or pre-election polls? mhatrw posted an intriguing article from US News and World Report in another DU discussion.

Exit polls have been carried out in the US since 1968. Do you have any of the dozens of exit-poll nonresponse rates by year? Have exit-poll nonresponse rates since 2000 been much higher than 20th century exit-poll nonresponse rates?

Do your estimates take possible nonresponse biases into account?

From http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=364x2654934 :

"mhatrw (1000+ posts) Thu Nov-09-06 09:07 AM Surprise, Surprise!!! Exits polls once again "biased" for Democrats!

The fault may be with the interviewers (do they approach only the voters they think simpatico), but it may also be with the respondents. Mitofsky has told me that almost everyone approached to fill out an exit poll questionnaire in countries like Mexico and Russia does so, while about half the people approached in the United States refuse.

Perhaps Republicans are more likely than Democrats to refuse ..."

[Febble]

21. Well there is plenty of evidence

for differential participation rates, but they do have to be inferred - they cannot be directly measured - and if you can't trust the count, inference is tricky. Not impossible however, and one of the strongest indicators were a couple of actual experiments with randomly allocated manipulated variables - and bias was greater in one condition than in the other.

However, a poll conducted just before the Tuesday's election provided what I think may be the first attempt to address directly the question as to whether Democrats are more willing to participate in exit polls than Republicans:

http://www.pollster.com/2006/10/22-week/

Of course, it's only a poll - and all polls are subject to bias! But it's a strong finding:

More Democrats (72%) than Republicans (66%) said they were likely to fill out an exit poll questionnaire. The gap was far bigger -- and highly statistically significant -- among those who felt strongly (typically a better predictor of actual behavior): 44% of Democrats said they would be "very likely" to participate in an exit poll compared to only 35% of Republicans.

[ProgressiveEconomist]

29. Sure, DIFFERENTIAL nonresponse for Rs and Ds would have to be

inferred, but not raw nonresponse rates over time. Without a time trend in the latter, a time trend in the former is less plausible, IMO. Do you agree?

Thanks for such a quick response to a technical question--it was almost too much to hiope for.

[Febble]

32. I don't know

But I am fairly sure there was a decline between 2000 and 2004. The target sample size was 100 in 2004, but the mean was 80, even though turnout was high. Interviewing rate would have been designed to net 100 samples on the basis of past response rates. That's why I infer a decline. I THINK I remember Mitofsky saying something about declining response rates - I meant to ask him a while back, but never got round to it, and now he's dead.

I will try and find out, unless someone else answers sooner.

Oh, and yes, I agree - except that the evidence from 2004 (and apparently, according to Mitofsky, in previous elections) is that the discrepancy tends to correlate with factors likely to be associated with selection bias (eg. long interviewing intervals) rather than response rate per se. If there is selection bias, ironically, completion rate (and therefore apparent response rate) may tend to go up (if interviewers are subconsciously selecting more willing-looking voters).

[ProgressiveEconomist]

35. Early voting and absentee ballots are other sources of possible

exit poll biases towards Democrats, if Republicans are more likely to vote absentee and are more likely to vote early. So time trends in percentages of the vote accounted for by absentees and by early voting are important to look at too IMO.

Both have been rising, IIRC, and Republicans tend to take more advantage of them IIRC.

[Febble]

36. Although pollsters

also do telephone polls of these groups. They may be more accurate, as they will be less prone to selection bias, so it is possible that the bias may reduce as a greater proportion of the "exit" polls are actually telephone polls of voters who have already voted.

[OnTheOtherHand]

37. two things

I don't have real data on response rates over time -- Mark Blumenthal mentions that the completion rate was 60% in 1992 and fell gradually to 51% in 2000 (but was 53% in 2004). But there is no inexorable connection between completion rates and non-response bias. For instance, if many Democratic voters became convinced that it was important to participate in exit polls, then both the completion rate and the bias would increase. The exit poll discrepancy was larger in 1992 than in '96 or '00.

One of Mitofsky's papers describes an experiment with three lengths of exit poll questionnaires. The longest one -- which was much like the one actually used in most states (two sides 8.5x11) -- had the lowest completion rate and considerable bias. The medium-sized one (I think it was one side 8.5x11) had higher completion rate and lower bias. The shortest one (half-sheet, just a few questions) had the highest completion rate, but higher bias than the medium-sized one. (He didn't provide the numbers.) Also, in the first San Diego "parallel election," higher completion rates appeared to be correlated with greater bias, although that hasn't been true of other parallel elections I've looked at.

All that was the first thing. ;) (I would still be curious about earlier rates myself -- it just has never mattered that much to me. The true believers rightly point out that pre-election telephone polls generally have considerably lower completion rates.)

OK, second, absentee/early voters do create a problem for the exit polls. As Febble mentioned, the pollsters conduct telephone polls in an increasing number of states to try to measure them, but it's not ideal. On the other hand, Democrats seem to be doing better in early voting than they have in the past, so it's really a wild card.

[ProgressiveEconomist]

43. 'true believers rightly point out that pre-election polls generally have

.considerably LOWER completion rates'

Now THAT's the "seminar" answer for autorank and TruthIsAll to my original question about the influence of exit-poll nonresponse on comparisons!

Most people won't follow up with the failure of that answer to cover the really important comparison, between exit polls and actual precinct results!

------------------------------------------------------------------------------------------------------------------------------

Many thanks for the quick reply, with data and even a reference on nonresponse rates over time!

IMO, if Congress really wanted to audit election results with exit polls, then it would mandate that a mandatory standard for national elections would include more co-ordination between election administrators and polling firms than we've seen.

This thread has convinced me that current datagathering for auditing elections is quite anadequate and needs drastic overhauling.

IMO signing a voter registration card should indicate intent for voters to comply with exit/audit polls, which would involve a combination of at-precinct interviews, phone interviews, and in-person followups. The randomization of voters for inclusion in an exit/audit sample could be carried out well before election day and include registered voters who did not in fact vote. When voters in the sample show up at the polls, poll workers could verify their contact information and give them a card to hand to pollsters outside, and with a toll-free number to call in if an immediate interview is not convenient. There could be some kind of "lottery" associated with participating in the poll, to maximize voluntary compliance. Survey personnel could stay around for a couple of days to follow up hard cases with in-person interview attempts.

Of course there still will be complete nonresponse and item nonresponse to deal with, but if polling firms had registration data on the nonrespondents, they could do a much better job trying to correct for them statistically than they can now.

[OnTheOtherHand]

52. auditing

It remains to be seen whether exit polls can be used effectively for audits, but one would want more precincts and shorter questionnaires. Of course, if we institutionalize exit polls as an audit mechanism along the lines you describe, that's a whole new ball game. Personally, I spend much of my professional life thinking about things that can go wrong with polls, and I would rather focus on improving vote counts (and protecting the right to vote). It seems to me that at best the margin of error of an exit poll will be substantially larger than we should tolerate as a margin of error for a vote count. If we used some congressionally mandated national exit poll as a device for keeping unauditable paperless DREs, that would be pretty damn Rube-Goldberg. All that said, I'm interested in the experiments with exit polls and parallel elections as auditing supplements, as long as they don't suck the oxygen from more urgent reforms.

Election jurisdictions vary widely in how they provide access to election returns and other precinct-level data. For instance, Sarasota has posted the vote broken out by early/absentee/election day; Manatee hasn't. But Sarasota's "voter statistics" (may be the wrong name) page just displays a pie chart of total registration by party. Some jurisdictions have counts quickly available at the precinct and others don't, but good luck figuring out which is which -- never mind whether the quick counts match the final returns. This decentralized 'system' isn't all bad, but it sure does complicate statistical forensics. Of course, again, even the best statistical forensics are no substitute for inherently trustworthy counts. Auditing survey responses and vote count data is sick fun, but auditing real paper would be better.

[bleever]

23. This would suggest that turnout was sufficient to trump fraud.

If it had been closer, it could easily have fallen within the "margin of fraud".

K&R.

[melody]

38. excellent post. thank you n/t

[autorank]

39. For the invisible people...from TruthIsAll

..btw, if you think this is all so useless, why do you bother
commenting.  I don't respond to people I think have notihng to
say, I just pass on by.  Yet there's all this attention and
sarcasm upstream.  

"I will not be as those who spend the day in complaining
of headache, and the night in drinking the wine that gives
it."
      - Johann Wolfgang von Goethe 

TRUTHISALL;

      TIA responds:

      The numbers I quoted were the from the PRELIMINARY EXIT
POLL (13028 respondents) taken from the CNN site. The FINAL
EXIT POLL (12251 respondents) was updated at 1PM.

      If anyone can find the 13028 RESPONDENT exit poll
results, let us know.
      Or has it just vanished into the ether?

      Virtually all of the demographic weights and vote shares
were changed after midnight. Here are the three demographics
from each timeline:

      PRELIMINARY NATIONAL EXIT POLL (CNN)
      Sample-size 13208
      MOE 0.87%

      GENDER
      ......Mix Dem GOP
      Male 48% 51% 47%
      Female 52% 56% 43%
      Total 98.5% 53.6% 44.9%
      2-pty 100 54.4% 45.6%

      PARTY ID
      ......Mix Dem Rep
      Dem 38% 90% 9%
      Rep 38% 7% 93%
      Ind 25% 49% 46%
      Total 100% 49.1% 50.3%

      VOTED IN 2004
      ......Mix Dem Rep
      Kerry 43% 92% 7%
      Bush 49% 15% 83%
      Other 4% 66% 23%
      No 4% 66% 32%
      Total 98.1% 52.2% 45.9%
      2-pty 100% 53.2% 46.8%
      ___________________________

      FINAL NATIONAL EXIT POLL (CNN)
      13,251 Respondents
      Updated: 1:00 p.m.

      GENDER
      ......Mix Dem Rep
      Male 49% 50% 47%
      Female51% 55% 43%
      Total 97.5% 52.6% 45.0%
      2-pty 100% 53.9% 46.1%

      PARTY ID
      .....Mix Dem Rep
      Dem 38% 93% 7%
      Rep 36% 8% 91%
      Ind 26% 57% 39%
      Total 98.6% 53.0% 45.6%
      2-pty 100% 53.8% 46.2%

      VOTED IN 2004
      ......Mix Dem Rep
      Kerry 43% 92% 7%
      Bush 49% 15% 83%
      Other 4% 66% 23%
      No 4% 66% 32%
      Total 98.1% 52.2% 45.9%
      2-pty 100% 53.2% 46.8%

[Febble]

47. Well, since you ask

and since you seem to relish pointing out my invisibility, even though, I gather, you can't quite resist taking the odd peek, here's my answer:

I "bother commenting" because I see nothing to be gained, and much to be lost, by leaving uncorrected what appears to me to be misleading information and faulty inference.

There is real evidence that Florida Congressional District 13 was either hacked or the ballot definition was screwed up, and that Jennings would have won if it had not been. And yet so entrenched is the completely unsupported belief that exit polls are infallible, that some people would rather infer fraud from those than from real data.

A pre-election poll provided strong evidence that Republicans would be less likely to respond to an exit poll than Democrats. The exit poll showed a discrepancy that is consistent with an over-sampling of Democrats. There is other strong evidence that Democrats tend to be over-sampled, year after year in exit polls. The exit poll discrepancy in 2004 was strongly correlated with factors likely to be associated with departures from non-random sampling i.e. with selection bias. The exit poll discrepancies at precinct level in 2004 were completely uncorrelated with change in Bush's vote share. Actual experimental evidence has shown a tendency for Democrats to be over-sampled in polls, relative to Republicans.

And yet, like those who cling to a belief that the universe was created 6,000 years ago, despite overwhelming evidence to support the Theory of Evolution, there are those who ignore overwhelming evidence that exit polls can be biased, and that in the US that bias tends to be in the direction of oversampling of Democrats, and insist that any discrepancy in the exit polls must be due to fraud. And "ignore" those who point out that the Emperor has No Clothes.

Well, it is possible that discrepancies in the exit polls are due to fraud. Indeed, fraud will tend to produce discrepancies in exit polls. But to infer fraud from exit poll evidence, and, moreoever, to infer it in specific places, is not only unwarranted but risks, as it is doing right now, and as it did in 2004, directing attention from the abuses for which there is excellent evidence and which warrant immediate investigation - from blatant voter suppression in Virginia (which would not show up in an exit poll) to wildly anomalous undervotes that could have cost Jennings her seat, in Florida (which are unlikely to show up in a generic House exit poll either.

I care about looking at data for what it can tell us, not about looking at it for what it can do to support my preconceptions. That's why.

[Sancho]

50. Again, your explanation seems to avoid the issue for pollsters...

if there is has been a "bias" in sampling, why would pollsters know that for years in election after election without identifying and dealing with it. Why would the female/gender interviewer bias not result in a change in hiring of interviewers? Some of us in the social sciences discovered this kind of bias years ago when the sex of IQ test givers interacted with the age/sex of test takers. The same thing with length of surveys, etc.

If there is a bias in interview completers, why was there not a heroic effort to obtain the "nonignorable nonrespondent" as researchers have called it for years?

Demographics, as pointed out by TIA, often describe a stratified sample. IF you have enough responses to sample to the desired MOE, then all you need after that is an observable population description at the interviewed poll: something that seems to be missing. Where's the sample to population descriptions (precinct or poll station level)?

Finally, where is the "raw" data. We deal with confidentiality all the time in social science data. There's no reason that the data can't be inspected in multiple ways that hide the respondents' identity.

What efforts did the pollsters make to target important and disputed races in Florida or Ohio with more polling, more experienced interviewers, different length forms, observed descriptions of demographics, and exit polls that offered a chance to describe "problems" with voting machines or voting?

Somehow, saying, "All we want to do is confirm the official results." is simply avoiding the obvious. If the explanations that EM offers are accurate and they are not covering up, then they are making a pretty lame effort to fix and correct the polling process for such well-paid and expert scientists from my perspective. Pollsters need to take responsibility for their weird or warped process and stop explaining why they are not accurate.

Challenge: In 2008, let's see the polls be on-target; and also see the polls used as evidence of fraud if it exists! No more excuses.

[OnTheOtherHand]

54. great, so offer yourself to conduct the exits

You want to see heroic efforts to eliminate bias. So far, the sponsors are satisfied with reasonable efforts to obtain usable data, cautiously interpreted so they don't blow calls. But maybe you can win them over.

[Febble]

55. There are good answers to your good questions

The first being simply that being aware that sampling tends to have a Democratic bias does not allow you to quantify the magnitude of it until afterwards (if you believe the results) or until after you have done some fairly sophisticated correlational analyses as to what methodological factors were associated with the apparent bias (although even then, you won't get an absolute value for the magnitude).

But what you seem to forget is that the polls are not designed as an independent check on the count. The pollsters do not make projections based on exit poll responses alone, precisely because they assume a risk of bias - however, they also know that it is rarely more than a few percentage points on the margin which will make a negligible difference to the utility of the cross-tabulations, and in any case, they reweight them to what they assume is an accurate count.

Yes, the pollsters select a stratified sample. Yes, the initial weightings relate to the stratification of the sample; post-stratification weighting is also done, based on observed visible characteristics of non-responders and the assumption the vote-count is correct. You may dislike the process, you may dislike the assumption, but it is perfectly standard practice. As for your other questions, I suggest you search the E-M site, or even contact the pollsters. There is, however, a lot of information about the sampling process on the site.

The 2004 raw data was, as always, archived publicly, and was available for free dowload for over a year. I expect the 2006 data will be too. The respondents' identity is hidden because precinct identifiers are not given. If they were, some respondents probably could be identified. That is why they are not given. "Blurred" data was however made available for Ohio. Both the raw data (together with weights) and the Ohio blurred set are available for anyone to inspect "in multiple ways".

I know of no efforts the NEP pollsters made to target disputed races, although they tend to aim for larger samples for close races, as they will need a smaller standard error. Remember they are not interested in checking the count. They are interested in characterising the electorate and projecting the official results.

Your last sentence seems yet again reveal a misunderstanding of the design of the NEP exit polls. If you want an exit poll to check on fraud, you need a different design of exit poll (although I would caution against any poll as an effective check on election integrity - I think there are better ways).

See my DKos diary here for further information about the NEP polls:

http://www.dailykos.com/storyonly/2006/11/4/135126/905

[Sancho]

88. I think I understand..

First, I would have done the correlational investigations (automatically) as part of sampling analysis. At least in the testing & measurement world, a preschool readiness test for 150 five year olds would automatically generate PCA, fit statistics, etc....much less a national effort.

Regardless, I understand the intent is not to detect fraud. But, after all the controversy ( and expectations since Cronkite predicted elections in 1952) folks expect the polls to least predict a winner (dichotomous decision analysis is not exactly the same thing as measurement precision/accuracy)! I would think the pollsters would do whatever they could to get an accurate poll so that TV predictions were correct and pundits can describe why people voted the way they did...

Over the last few polls, it has become pretty obvious that close elections can be manipulated or fouled-up and polls are ONE way to trigger an investigation when unexpected undervotes, strange distributions, and other patterns occur.

I just think the pollsters have too many explanations and excuses for a familiar process that they are paid to do, and there is some reason to think that they might not want to embarrass the boss with the paycheck.

I have downloaded the 2004 data, and one day when there's time, I'll play with it, but it would really, really be better to break it down by precinct. When I open the SPSS files, the precinct is missing. Even if the precincts were coded and not identifiable, it would help analysis to have those units of analysis.

I know that some "predictors" may seem to have abnormalities. For example, I'm one of those in Florida who has registered as a Republican for a particular term so that I can participate in a primary, even though it's pretty unusual for me to vote anything except straight Democratic. That would look weird to a pollster but would account for registrations to misfit voting. That's exactly the kind of question that needs to be asked!

I've certainly conducted surveys on paper and online. Sometimes on sensitive topics. I would almost always anticipate self-selected bias and plan a special effort to profile the non-responders to an interview (like invite them to a focus group or call them later on the phone, etc.). I would almost always plan on extra effort in the more difficult sample areas or where accuracy is most important. My coding would typically include the interviewer ID so that I can detect "interviewer effects" (gender of the pollster, etc.).

I still can't tell if the pollsters really can't solve the problems, as we keep hearing, or don't want to do a better job in order to avoid evidence of hanky panky. I also don't see anyone describing the "population" they are sampling from...if I interviewed 100 volunteers, I have that demographic profile, but I also have to do a description of the demographics going in the door (by observation, voter roll signatures, whatever) of the non-respondents.

Why wouldn't the pollsters have noticed the "undervote" in Florida the day after the election? Why is it so hard to see that in 2004, a bunch of people in some counties consistently voted for increasing taxes, Democratic local representatives, and against Republican amendments, EXCEPT for ONE selected congressional race (Castor/Martinez)? Yes, poll data could be used as appropriate evidence in such cases to trigger investigations - if it was released quickly, by precinct, and with accurate sampling.

[Febble]

89. Well, again there are some answers here

Yes, of course the pollsters try to get an accurate poll. And they do, pretty well, given the constraints of a vast survey conducted on a single day with over a thousand interviewers. And they do a post mortem each year to try to find out what the key factor were that tended to contribute to apparent bias. One key factor is sampling rate, and adherence to it. They may have increased it this year, and they were going to make some changes to the training manual. Indeed, the discrepancy looks lower.

But they are not, nor have they ever, had the intention of monitoring the integrity of the count. Their clients pay them to do as accurate a job as they can at projecting the results and providing a good characterisation of the electorate (who voted for whom and why"). I have no reason to think they don't do as good a job as they can, but no survey design, however good the methodology, be immune from bias in the sample, and this is particularly true of surveys in which repondents are selected face-to-face, and where response rate is low.

I'm not sure what you have downloaded - precinct numbers are given, as are the weights. But the files are no longer available for free download AFAIK, so perhaps you haven't downloaded the right thing. Or perhaps you are a subscriber. But if you have them, you do some of the cross-tabulations you are interested in regarding party ID etc.

Yes, the pollsters subsequently investigated factors associated with bias. Their report, which surely you must know, is here:

http://www.exit-poll.net/election-night/EvaluationJan192005.pdf

I was subsequently hired to investigate further, but my findings corroborated these, largely. In addition I found this:

http://inside.bard.edu/~lindeman/slides.html
http://journals.democraticunderground.com/Febble/3

You might like to check my other journal entries for more info.

As for your last question - I don't know if you have read the relevant post, but you seem to be unaware that the number of precincts sampled in each state is in the order of tens - i.e. often less then one per county. There could well have been no sampled precincts in Sarasota, or one or two. Not nearly enough for the kind of inference you want.

NEP exit poll data is useless for analysis by county.

[Time for change]

41. Can you please clarify?

You say that the CNN 2006 exit poll had:
Democrats voted 90% for Dems
Republicans voted 7% for Dems
Independents voted 49% for Dems

But from what I can tell lookin at the CNN poll, the numbers are:
Democrats voted 93% for Dems
Republicans voted 8% for Dems
Independents voted 57% for Dems

This is the site:
http://www.cnn.com/ELECTION/2006/pages/results/states/US/H/00/epolls.0.html

Are you using a different poll than I'm looking at?

[Awsi Dooger]

49. Apparently an earlier version of the same exit poll

TIA says the party ID category from the pre-weighted House exit poll was out of whack favoring the GOP. I was sampling senate exit polls at the time and posted a couple of them here for screen shots, but not until late did I check the House exit poll.

The weird number is independents only 49-46 toward Democrats in the House poll. That's what TIA is claiming. I have no idea if CNN listed that number but if so it obviousy makes no sense. Even in the South, Democrats fared better than 49-46 among independents.

[OnTheOtherHand]

53. the 7:07 PM national tab was 58-38

for independents in House races. Now it is 57-39. No clue where TIA got his number there (nor where he got his number for Virginia).

[Awsi Dooger]

82. Some sites had a typo of the Virginia exit poll number

MyDD initially reported it at 52-46 for Webb, then changed it to 52-47 quickly, saying the initial number was a mistake, copied incorrectly off ThinkProgress. For a brief period I saw the 52-46 on other sites, apparently toted from MyDD.

I don't remember 53-46 listed for Virginia. One other possibility is the Rhode Island number, which was immediately below Virginia when the exit poll leaks came out, was 53-46. Maryland, Montana and Rhode Island were all 53-46 in the initially leaked numbers, so perhaps TIA mistakenly copied down or remembered 53-46 for Virginia.

I have no idea. This year I assumed screen shots would be priority one for the fraud crew, but just in case I posted a couple of threads with an early exit poll link. I bored of it quickly when the responses were almost nil, and there were no similar threads.

http://thinkprogress.org/2006/11/07/senate-exit-polls-early/

[OnTheOtherHand]

83. I have a similar thought about that weird indy number

All the national #s I saw showed the Dems well up among independents.

[philb]

42. Do the exit polls have info on the undecided Congressional races?

FLorida
Georgia
NC
New Mex
Wyoming
Connecticut
Ohio

[autorank]

45. Here you go...this should ave some of what you want...

Check this out http://electionfraudnews.com/Elections/2006WatchPostEl.1.htm

It's a follow up on this http://www.scoop.co.nz/stories/HL0611/S00119.htm

[Febble]

46. autorank's links

are, I believe, to pre-election polls, not exit polls. House races were not separately polled, but there was a generic House exit poll, subdivided by region (East, MidWest, South and West)

http://www.exit-poll.net/election-night/stateraces.html

So the short answer is, no.


edited for typo

[In Truth We Trust]

48. K&R Hand Counted Paper Ballots NOW!!! or fascism forever. Thank you TIA and thank you Michael too!

[katinmn]

56. Thanks TIA and Auto. Thankfully, turnout trumped fraud.

[Joe Chi Minh]

58. You, TIA and all the others working on on the electoral fraud,

Auto, seem to have made an epic contribution to the success of the Democrats in this election. I think it quite likely your efforts will have made it immeasurably more difficult for them to have subverted the various poll results on the scale they needed to. Motivating Americans everywhere to get out and vote.

[civildisoBDence]

59. A simple explanation, and a simple question...

People lie to pollsters. We often tell them what we think they want to hear. What these numbers show is that Democrats are more likely to be proud of who the voted for, Republicans more likely to be ashamed.

The question: if Republicans have the ability to manipulate votes, why would they engineer a narrow defeat instead of a narrow victory? Why take that risk and then leave yourself short?

Paper ballots would minimize fraud AND marginalize conspiracy theories.

Newsprism

[The Count]

61. People do not lie about actions they too,. They tried but were watched.

All the "narrow" defeats - last 3 senate races for senate - were in fact vast victories. And they had already removed 4 million voters BEFORE the elections.
read this:

« Steal Back Your Vote
HOW THEY STOLE THE MID-TERM ELECTION
Published by Greg Palast November 6th, 2006 in Articles
http://www.gregpalast.com/how-they-stole-the-mid-term-election#more-1530
by Greg Palast

EXIT POLLS ARE ALWAYS RIGHT!!!!!

PREVENT/PUNISH ELECTION FRAUD - FIRST OF ALL!!!!!!

[pberq]

60. Kick & Nominated - let's keep this front and center

[slackmaster]

62. Exit polls were designed to create media content, not to be accurate

WHY DO THEY DO THIS?
WHY DO THEY ALWAYS ASSUME ZERO FRAUD?
WHY DO THEY ALWAYS ASSUME A PRISTINE VOTE COUNT?

Because fraud detection is NOT what the polls were designed for.

[Up2Late]

63. Yup, good thing the MSM agreed, this time, to keep the "exit polls" secret...

...for most of election day and didn't really cover what the polls said until after they were "adjusted."

Does that sound much like Soviet-era, eastern block "news" reporting to anyone else besides me?

And isn't that nearly the exact definition of the word "Conspiracy?" Ah, but that's just my theory.:banghead:

[Febble]

64. Except that they didn't....

what are you talking about? CNN posted the cross-tabs at close of poll in each state like they always do.

[Up2Late]

66. No it wasn't "...like they always do..." Usually the exit poll either leak...

...out during the day, or they are made available as they are done during the day, this time they locked up everyone who usually leak the exit polls in rooms without telephones and access to the outside world, so that none of the "early exit Polls" would leak before the actual results were know, which completely defeats one of the main purposes of Exit Polling!

Was this the first election you paid attention to or did you fall into the trap of only watching CNN this year? Exit polling was crap this year.

But Hey, if you don't believe me, maybe you'd believe Andrew Kohut, director of the Pew Research Center for the People and the Press! Listen to this interview he did on Election day (November 7, 2006) with NPR's All Things Considered:

<http://www.npr.org/templates/story/story.php?storyId=6449790>

Poll: African-Americans Concerned with Voting

Listen to this story...(at link above)

All Things Considered, November 7, 2006 · Recent polling at the Pew Center paints a mixed picture of American voters' confidence in the way their votes are counted, with Democrats and African Americans expressing heightened concern. Melissa Block talks with Andrew Kohut, director of the Pew Research Center for the People and the Press.

<http://www.npr.org/templates/story/story.php?storyId=6449790>

No matter how much people here complain about how much "....NPR has changed...", The news they report is still are 100 times more reliable than what you get while watching the 24/7 Cable "news" channels!

PLUS, all of their News reports are kept on-line, even the ones with LIVE Breaking News, so you can go back to any past election or major news event (back to 1996) and listen to how it was covered at the time, mistakes and all.

Just go to these links and click on the "Past Shows" links and pick a date! It's all there, back to mid-1996, that is:

<http://www.npr.org/templates/rundowns/rundown.php?prgId=2>

<http://www.npr.org/templates/rundowns/rundown.php?prgId=3>

[Febble]

68. Well, you might like to read

my DKos post here:

http://www.dailykos.com/user/Febble

What the pollsters did this time was prevent any leaks before 5.00pm, well before close-of-poll, the reason being that very little can be inferred from incomplete samples. CNN posted close-of-poll crosstabs for each state, which would not have been adjusted to the vote count, because at that stage there would have been no results. These were downloadable by anyone (and were dowloaded by OnTheOtherHand), and would have remained unchanged for a couple of hours, at which point they would have been gradually adjusted in line with the vote-returns, and in lines with any projections being made for each state.

By recording the close-of-poll crosstabs, and comparing them with the final crosstabs (or final results) you can get an estimate of the magnitude of the adjustment required, and thus of the magnitude of the discrepancy between the estimates made at close of poll i.e. from exit responses and pre-election polls only, and the official result.

So, no, there was no cover-up, just as there wasn't a cover-up in 2004. It was all posted in plain sight.

[Up2Late]

70. So I guess since YOU posted that info over at DailyKOS it MUST be true!?!?

Come on, nothing you wrote here make any logical sense. What you posted here is exactly why this years "exit polls" are completely BOGUS!

You wrote: "...These were downloadable(sic) by anyone (and were downloaded by OnTheOtherHand), and would have remained unchanged for a couple of hours, at which point they would have been gradually adjusted in line with the vote-returns, and in lines with any projections being made for each state...."

You think exit polls that are

"...gradually adjusted in line with the vote-returns..." is O.K.!?!?:banghead:

That's B*llSh*t!!! Or to use a word you would more easily understand, That's Bollocks!!!

If you allow the exit polls to be "...gradually adjusted in line with the vote-returns..." they become completely bogus! It's nearly the perfect definition of Fascist Propaganda!

As for your post over at DailyKOS, you should listen to this person and the advice they give you in this post:

It could have been the COUNT that was wrong (1+ / 0-)

Recommended by:
bdevil89

Your discussion presumes that the official count was accurate.

No good scientist would presume something that major.

Your analysis of 2004 must take into account 2 possibiliies(sic):
either the exit polls were wrong OR the official count was corrupted.

You must examine both possibilities in order to have a solid analysis

Thus this sentence from your diary presumes that the "incoming vote" was accurate:
Again, if the incoming vote returns indicate a systematic divergence from the exit poll response, the pollsters have reason to suspect bias in their sample.

What if it was not?

The fatal flaw in the (Evaluation of Edison-Mitofsky www.exit-poll.net/election-night/ ) is that they refused to even acknowledge the possiblity(sic) that the official count was corrupted - anywhere.

by AskQuestions on Sat Nov 04, 2006 at 03:13:01 PM PST

Parent

Sorry, I'm not going to waste anymore of my day arguing with you and your illogical, circular logic, because as I can see from you post at DailyKOS, that it would be nothing but a waste of both of our time, or as an old colloquialism here says:

"Never try to teach a pig to sing. It wastes your time, and it annoys the pig."

[OnTheOtherHand]

71. WTF?

Febble stated that CNN posted the cross-tabs at the time that the polls closed in each state. You retorted that the NEP locked everyone in a room to prevent leaks. Febble agreed that this was true, until 5 PM Eastern, which is before the polls close in any state.

At which point it would be appropriate to apologize for misreading her original post, and concede that of course she is right on the facts.

But instead, you launch into a rant about the adjustment process. Your rant doesn't actually challenge any of the facts in Febble's DKos post, although it makes clear that you don't like how the process works. Bully for you.

Are you going to concede the fact that CNN posted the cross-tabs at the time that the polls closed in each state? or does that require too much concentration?

[Febble]

73. No, it certainly doesn't make it true

but nor did I make it up. It's up to you to evaluate.

But you yourself are not making sense. Did you actually read the post? The exit polls are not designed as an independent check on the count. That is why they are gradually adjusted to it. But, nonetheless, as I say in my post, if you want to know what the estimates were at close-of-poll, i.e. without any adjustments, all you had to do was download the cross-tabs at close of poll. Which many people did. If you want to regard those as the "true" estimate that's up to you. But it certainly isn't "secret" which is what your upthread post alleges. The unadjusted cross-tabs were posted, at close of poll, on CNN, as in 2004, and were available for anyone to download.

As for the post you cite, it is simply wrong: there is nothing wrong with scientists making assumptions as long as they state those assumptions clearly, which the pollsters do. They also provided the data before the adjustments. The fact is that your poll post was in error. The adjustments were not hidden.

But yes, I rather agree with you about the pig.

[msedano]

65. thanks for the update and (relative) simplicity

i shared this with some people, including at least one carping simpleton. many dug it with good comrehension, but others said "Huh?" Here's a Spanish word for you. The "huh" folks who refuse to understand the numbers and the words are what we call menso. A woman would be mensa -- just like the name of the society of bright people. Menso / Mensa means a stupid person, a dud, a block, a stone, a less than senseless thing.

[rumpel]

67. btw

LA County Edison Mitofsky request of exit poll data request of selected precincts.

Contract

September 19,2006
Conny McCormack
Los Angeles County Registrar/Recorder/County Clerk
Dear Conny,
Edisonhlitofsky has been commissioned to provide election coverage on behalf of the National Election
Pool whose members include: ABC, the Associated Press, CBS, CNN, Fox, and NBC.
As part of our research, we wish to obtain results for the Governor and US Senator races on
November 7th, 2006 via snap tallies for the following 23 precincts:
0090004A 0950022B 1450062A
2500029A 3400023A 3850562A
3550098A 3850101A 9001648A
9000318A 9000518A 7150146B
9001013A 9001094A 9000429A
9002054A 9005428A
3990015A 4150048A
5200042A 6230051A
6700010A 6700054A
It is the policy of Edison/Mitofsky and the National Election Pool members to not report its voter
analysis until all polls have closed in your state.
If you have any questions, please call me toll free at 800-725- 1645.
Greg Taylor
Election Research
Edison/Mitofsky

http://lacounty.info/omd/q4_2006/cms1_051017.pdf#xml=http://search2.co.la.ca.us/omd/xmlread.asp?K2DocKey=http://lacounty.info/omd/q4_2006/cms1_051017.pdf@OMD&QueryText=

[OnTheOtherHand]

69. snap tallies

aka "quick counts." These are indeed "exit poll data," but not in the sense usually understood on DU. Vote counts, not exit interviews.

[rumpel]

72. In other words, are they tweaking the Exit polls, with the

snap tallies throughout the day?

Or is it done at the end of the day. Are they interviewing people in the same precincts?

[OnTheOtherHand]

74. not throughout the day

p. 2 of the PDF you linked to states, "Snap tallies take place after the polls have closed...."

Typically the exit poll precinct sample is a subset of the quick count precinct sample, so, presumably they interviewed people in some of those precincts. As they get a quick count for each exit poll precinct, it replaces the interview data in the estimation model. They also incorporate county vote counts. (I've never seen the model, so my knowledge gets fuzzier as we go, but so far, I think I'm fine.)

[rumpel]

75. but we do not tally the races at the precinct in LA county

it is being taken by the election judge and one more pollworker to the collection center -it is quite immediate - when we close- we pack and leave.
From the collection centers the "packages" are driven or flown by helicopter by the sherriffs to the registrar HQ in Norwalk.

So - I am puzzled - the earliest the HQ can even tally these are past 10:30p

[Febble]

77. Ah.

yes, there was a large minority of precincts in the 2004 where there were no precinct level counts available to compare with the precinct level exit poll tallies. In these cases, the pollsters get the precinct counts from the county, which may or may not incorporate absentee and early vote counts - it amazes me how vague these tabulations seem to be in places, regarding the sources of the numbers - it's just asking for errors and worse.

[rumpel]

79. To clarify, the only tally we do is the total of ballots

spoils, absentee, provisonal - compare it with the total of signatures, as well as the other books ( ex: provisional) We all witness and sign an oath and off the judge goes.

btw - by law each precinct is supposed to post this - however we did not in, I believe it was, '98 when I worked as poll worker- nor was it done over the entire time since, including 11/7/06

[Febble]

81. It boggles my mind

frankly! In the UK it is so straightforward (I know, I know....)

But turnout is also of importance to models, so the data you say is available may be what they wanted. But they'd want the vote-shares too.

[Febble]

76. No, I think you are misunderstanding.

Very few precincts in any state actually have interviewers. The number per state is in the tens - not hundreds - which means in many states less than one per county (so exit polls are no good for analysis at county level, or, indeed for analysis at CD level, which is why they don't make estimates for individual House races.)

But the pollsters do request vote returns from a larger number of precincts, and they also collect county tabulations. The correspondence you posted appears to refer to a request for precinct counts, because, as you probably know, once the precinct results start coming in, they gradually replace the exit poll responses for the purpose of making the estimates, and then these in turn are further "tweaked" by county tabulations. All this time the confidence limits of the estimates are tightening, and when they reach a critical value (99.5% confidence) state races can be "called". In the mean time, the cross-tabulations posted at close-of-poll and based only on data availabe pre-result (exit poll responses; pre-election expectations) are periodically adjusted in line with the current estimates of the final result. The first adjustment is made, I understand, 2 to 3 hours after close-of-poll.

[rumpel]

78. So are they basically requesting data from the county database

which includes each voters' previous voting history instead of actual interview?

[Febble]

80. Well, I don't know for sure

but from the look of it, I think they are simply asking to be informed of the results for those precincts on election night. But from what you say, in your county, those would not be available from the precincts, but would have to be got from the county tabulations.

The "exit polls" include very much more than interviews of voters as they "exit" the polling place. They include pre-election polls; telephone polls of absentee and early voters; precinct vote returns (either counted at the precinct or from county tabulations) and county results. All these things go into the "projections" although obviously at "close of poll" in each state, no vote-returns are included in the posted cross-tabulations. So those are "uncontaminated" as TIA would say.

But the actual estimates for each race are dynamically adjusted as more and more vote-return data becomes available, and this data includes data from precincts selected in advance, as per the correspondence you posted. Periodically the cross-tabulations are updated to match the estimates. When the "t value" of the estimates reach a criterion value (ie. when confidence in the estimates reaches 99.5%) the race is "called".

So yes, if you like, the vote returns are used "instead" of the exit poll responses, as they become available, but more precincts are selected to provide vote return data than are selected to provide exit poll data.

The exit polls, it may bear repeating, are not designed as a check on the integrity of the election. They have to be reverse engineered to do so, and, IMO, do so badly as a result. But the close-of-poll cross-tabulations give you a peek at the values of the estimates before vote-returns are incorporated.

[autorank]

84. TIA: National and State Exit Polls on CNN: the numbers were changing.

TRUTHISALL:

The sample-size of 13,208 in the preliminary exit poll was increased to 13,251 in the final.

Here's the current CNN version:
http://www.cnn.com/ELECTION/2006/pages/results/sta...

For comparison, here is the NYT demographic summary:









__________________________________________________________________
These sites tracked the exit polls.

THINKPROGRESS.ORG
A link to early exit polls: Webb 52-47
http://thinkprogress.org/2006/11/07/senate-exit-po.../

POLITICALWIRE.COM
Another link to early numbers, confirming the Webb 53-46 lead:
http://politicalwire.com/archives/2006/11/07/exit_...

Some comments from POLLSTER.COM which was tracking the exit polls:
http://www.pollster.com/exit_polls/live_blogging_e...

How come no one at POLLSTER.COM mentioned the 7% Webb lead at 7pm?
__________________________________________________

Allen's approximate number from this tabulation is
(.49*.53) + (.51*.43) = 47.9
Webb's number (.49*.46) + (.51*.56) = 51.1

Posted by: bebop | November 7, 2006 7:49 PM
___________________________________________

The exit poll percentages on CNN are now different than your original post. Recalculating the marginals now gives Allen the lead in the exit poll.
Why might the exit poll percentages have changed?
Posted by: Pat Ross | November 7, 2006 9:10 PM
___________________________________________

I've tried to estimate exit poll margins from a few of the tabulations for the 22 Senate polls I have so far (all of which I think I saved before they were updated). Those tabulations presumably are based on composite estimates incorporating pre-election returns.

When I compare the margins to the Pollster.com pre-election average margins, the exit polls appear to be running about 3.8 points more Dem than the pre-election polls -- which suggests that the actual gap could be wider. Several caveats on that: (1) I can already tell that my eyeballed margin in Missouri is about a point too large, so the gap could narrow. (2) Exit poll discrepancies have generally run high in the Northeast, which is overrepresented. The biggest discrepancies so far appear to be in non-competitive races, with the possible exception of CT. (3) Some part of this may be attributable to Democratic surge, and I don't have enough info yet to estimate that possible effect.

Posted by: Mark Lindeman | November 7, 2006 10:00 PM
__________________________________________________

CNN is now reporting Webb with a 2500 vote lead. But it doesn't jive with the link I posted earlier. CNN has Webb leading at roughly 1,141,000 to 1,138,000. However, the link makes those numbers strange, perhaps counted or reported in a different order, since the link currently has Allen ahead roughly 1,139,000 to 1,137,000.

Posted by: Gary Kilbride | November 8, 2006 12:05 AM
___________________________________________

In the past 10 minutes, all the voting data from Yellowstone County has disappeared! The last I saw Tester was slightly ahead with over 40% of precincts reporting. Now no votes are showing up.

http://www.cnn.com/ELECTION/2006//pages/results/st...
http://election.cbsnews.com/campaign2006/county.sh...
What is going on?!

Posted by: Marc Gans | November 8, 2006 3:27 AM


SEE WWW.TRUTHISALL.NET

[OnTheOtherHand]

85. because the tab didn't show Webb +7 at 7 PM

I could send him the save, but how would he know that I didn't rig it? How can he know that I'm not also bebop, faking history in real time?

Perhaps someone else could send him a screen shot. There must have been several thousand of us at least, right?

[Awsi Dooger]

87. Maybe someone could upload the screen shot

Here's one site among many: http://rapidshare.de/

The file is uploaded and the link pasted. That way anyone has access to the link and you don't have to deal with PMs or emails. With a file of screen shot size, the uploading and downloading would be a flash.

[Awsi Dooger]

86. OK, I see where TIA got the +7 for Webb

It was reported on Political Wire.com:

http://politicalwire.com/index.html?page=3

"Early Senate numbers (uncomfirmed and with caveats):

Democrats are leading in Pennsylvania (+15), Ohio (+14), New Jersey (+8), Rhode Island (+7), Virginia (+7), Montana (+9), Missouri (+2), Maryland (+7).

Republicans are leading in Tennessee (+4) and Arizona (+4).

Lieberman leading by 5 in Connecticut."

Those margins are the same as TIA is using on the 7 PM chart, other than Tennessee is -4 instead of -3, and Montana is +9 and not +7.

Regarding Pollster.com, Mark Blumenthal specifically wrote he was not relying on "leaked" results, but an approximate tabulation using gender percentages from CNN: 49% men, 51% women; Allen leading 53-46% among men, Webb ahead 56-43% among women. The first mention of the CNN availability is 7:05 and the estimate using gender percentages is posted at 7:20. I assume the numbers did not change in that time frame. If not, CNN's exit poll page never provided a 7 point margin for Webb, or anything approximating that. I had not seen a theoretical +7 for Webb until TIA provided the link to Political Wire.

During election night on Pollster.com I was commenting on the strange discrepancy between CNN and MSNBC's numbers as opposed to what the Virginia secretary of state's website was reporting. My reference to CNN meant TV, not website. Exit polls were irrelevant at that point. I was trying to root in the damn race. Believe it or not they're going to decide the result on vote counts, not early exit poll leaks. Same thing with Montana and the Yellowstone County situation in the early morning hours Wednesday. Everyone could tell the exit polls had redshifted again and there would be plenty of debate here and elsewhere for months/years. That wasn't exactly a priority on election night.

Late addition: I see OTOH posted that he has the screen shot and the 7 PM tab doesn't show Webb +7.

OK, now the Giants/Bears game. The pre-game poll has Giants by 1.

[mom cat]

90. Surprise! Nobody would ever have imagined that such a thing

could ever happen!