EXIT POLL QUESTION: Party IDs vs. MOEs

[Bill Bored]

OK, someone please explain this, preferably in 25 words or
less and without the use of the terms "rBr",
"Bias Index", "ln(alpha)",
"alpha" or "WPE." (Well, OK, use them if
you must!)

From the national poll, Party IDs and 2-party vote totals in
the West shifted from:

11/2/2004 3:59:05 PM
No matter how you voted today, do you
usually think of yourself as a: (n=1,698)
           Tot Kery Bush
Democrat    38   91   7
Republican  33    8  91
Independent 29   53  40

Vote Estimate: K 53 B 45
Margin: 8
===========================================
11/2/2004 7:33:46 PM
No matter how you voted today, do you
usually think of yourself as a: (n=2,130)
           Tot Kery Bush
Democrat    37   91   7
Republican  34    8  91
Independent 30   55  40

Vote Estimate: K 53 B 45
Margin: 8
===========================================
11/3/2004 1:24:53 PM
No matter how you voted today, do you
usually think of yourself as a: (n=2,544)
           Tot Kery Bush
Democrat    34  91   8
Republican  36   7  92
Independent 30  52  44

Vote Estimate: K 50 B 49
Margin: 1
===========================================

So between 7:33 PM and the final result, you had:
a 7% swing in the margin between Kerry and Bush,
a 5% change in Party ID Weighting, toward Republican,
a 7% shift in Independents toward Bush,
and even greater shifts compared to earlier in the day (3:59
PM).

Yet the polls in these states were all within the MOE!

How could this region be so far out of whack and the states
still be within the MOE? CA, the biggest one, was only off by
.5%, according to TIA.

Let's hear some theories about this and not just a blizzard of
numbers that might somehow explain it.

[TruthIsAll]

1. Re: Accurate state exit polls; Bogus National(Regional) Exit (13660)

From the national poll, Party IDs and 2-party vote totals in
the West shifted from:

11/2/2004 15:59
MOE	2.37%

No matter how you voted today, do you			
usually think of yourself as a: (n=1,698)			

Tot		       Kerry	Bush
Democrat   	38	91	7
Republican 	33	8	91
Independent	29	53	40
Calc		       52.59	44.29
===========================================			

11/2/2004 19:33
MOE	2.12%

No matter how you voted today, do you			
usually think of yourself as a: (n=2,130)			

Tot		       Kerry	Bush
Democrat   	37	91	7
Republican 	34	8	91
Independent	30	55	40
Calc		      52.89	45.53
===========================================			
11/3/2004 13:24
MOE	1.94%

No matter how you voted today, do you			
usually think of yourself as a: (n=2,544)			

Tot		       Kerry   Bush
Democrat       34	   91	   8
Republican     36	   7	   92
Independent    39	   52	   44
Calc		        49.06	49.04
===========================================

So between 7:33 PM and the final result, you had:
a 7% swing in the margin between Kerry and Bush,
a 5% change in Party ID Weighting, toward Republican,
a 7% shift in Independents toward Bush,
and even greater shifts compared to earlier in the day (3:59
PM).

Yet the polls in these states were all within the MOE!

How could this region be so far out of whack and the states
still be within the MOE? CA, the biggest one, was only off by
.5%, according to TIA.


*************************************************
Bill, I have calculated the voting percentages based on the
given weights in addition to the regional MoE.

First of all, the 7:33pm Exit Poll MoE = 2.12%,
There was a 95% probability that Kerry's final Western
regional vote would exceed 50.77%.

Kerry won all exit poll timelines up to and including the
13047 respondents at 12:22am (not shown).

We have already shown (using 2000 vote stats) that the
"How Voted in 2000" demographic weightings (43%
Bush/37% Gore) are absolutely IMPOSSIBLE in the FINAL national
exit poll of 13660. They are least 3% TOO HIGH for Bush and
TOO LOW by 3% for Gore. That equates to 4 million votes
switched from Kerry to Bush. We also KNOW that the Final 13660
Exit Poll was matched to the vote count.

The final set of Western regional numbers (at 13:24) are bogus
as well. As a component of the final 13660, these had to be
adjusted also.

THE STATE EXIT POLLS WERE ACCURATE. KERRY WAS A SOLID WINNER
IN THE WEST (PRIMARILY DUE TO CA). THE STATE POLLS AGREE WITH
THE 7:33 NATIONAL EXIT POLL. THERE IS NO CONFLICT.

AS WE SHOULD ALL KNOW BY NOW, THE FINAL NATIONAL EXIT POLL OF
13660 WAS MANIPULATED TO MATCH THE VOTE. EARLIER STATE AND
NATIONAL POLLING TIMELINES WERE CORRECT.

THE NATIONAL EXIT PARTY ID WAS CHANGED: 
FROM 38-35-27 TO 37-37-26

SO, TOO, THE WESTERN PARTY ID HAD TO BE CHANGED: 
FROM 37-34-30 (OVER 100% DUE TO ROUNDOFF?) TO 34-36-30.

THAT'S ALL THERE IS TO IT.

[TruthIsAll]

2. A FEW RELEVANT LINKS TO THE DATA

ALL YOU NEED TO KNOW ABOUT THE NEP TIMELINE
Edited on Thu Apr-28-05 07:25 PM
http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x363938


13047 NATIONAL EXIT POLL CONFIRMS THE STATE EXIT POLLS WITHIN .06%
Edited on Sun Apr-24-05 11:44 PM
http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x362208

STATE / NATIONAL EXIT POLL ANALYSIS BY REGION
Edited on Sun Apr-24-05 03:10 PM
http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x362106

A TIA CHALLENGE: REFUTE THESE ODDS!
Edited on Sat Apr-23-05 06:16 PM
http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x361819

Z-score Probability (Bush Z>1 in 35 states): 1 in 4500 trillion
Edited on Sat Apr-23-05 01:28 PM
http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x361736

Red Shift: Z-scores and Probabilities. Take a deep breath.
Edited on Sat Apr-23-05 03:26 AM
http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=203&topic_id=361617&mesg_id=361617

[Time for change]

20. TIA, can you give us a source for for this thread that you posted?

http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x362106

Thanks.

[TruthIsAll]

52. 12:22am State exit polls (Simon download) and National 13047

/

[Bill Bored]

3. I think there is a conflict.

You said,

"THE STATE POLLS AGREE WITH
THE 7:33 NATIONAL EXIT POLL. THERE IS NO CONFLICT."

Fine, but that's not the conflict I had in mind.

If the state polls were within the MOE, so too should the national poll have been that represented the same states. But in the West, the national was way off while the state polls were close to the vote count. How could this be? There was only one reported vote count wasn't there?

The discrepancy between the state polls and the vote count should NOT be so much less than that of the regional (which is based on the national) and the same vote count.

[TruthIsAll]

4. Compare apples and apples: The 12:22am (13047) national vs. state exits

First of all, realize that you are basing your analysis solely
on the party-Id demographic. We are comparing the weighted
Western states exit polls to this demographic. Theoretically,
they should be equal, but it's apples and oranges; we cannot
expect them to match. But they are close: 

53% National Exit -West (Regional) 
52.89% National Exit- West (Party ID) 
51.81% State exit polls - West (weighted)

Look at the National Exit Poll timeline. 
THE RESULTS WERE NOT ADJUSTED TO THE FINAL VOTE COUNT: 
4pm - 8349 respondents
7:33pm - 11027 respondents  
12:22am - 13047 respondents
 
HERE ARE LEGIT KERRY PERCENTAGES FOR THE WESTERN REGION:
7:33pm  National Exit (Party ID): Kerry 52.89%-45.53%
12:22am National Exit (WEST): Kerry 53%-45% (rounded)
12:22am WEST State Exits (wtd): Kerry 51.81%

THESE BOGUS FINAL 13660 RESULTS WERE MATCHED TO THE VOTE
COUNT:
1:25pm National Exit (WEST region): - 50% (rounded)
1:25pm WEST State Exits-wtd (50.87%)

Party ID: Western region 
11/2/2004 7:33:46 PM
No matter how you voted today, do you
usually think of yourself as a: 
(n=2,130)
           Tot Kerry Bush
Democrat    37   91   7
Republican  34    8  91
Independent 30   55  40
Total Est   101 52.89	45.53

(Note: The 101% Total is due to roundoff. There is no way of
knowing the effect on Kerry/Bush percentages).


NATIONAL EXIT POLL - 13047 RESPONDENTS
BASED ON NEP REGIONAL WEIGHTING:	
Kerry wins by 50.53-47.95%

Region  Pct Bush	Kerry	Nader
EAST	   22%	41%	58%	1%
MIDW     26%	49%	50%	1%
SOUTH	   31%	54%	44%	1%
WEST	   21%	45%	53%	1% 

TOTAL  100% 47.95% 50.53% 1.0% 
Votes         58046 61169 1211						

Here is the corresponding 12:22am Western region total, based
on STATE exit polls, weighted by vote: 
Kerry = 51.81% 

Region	Z	Votes	 Size  MoE	StDev	Poll	Vote	Dev	Prob 	 1 in	For
											
AK	2.44	302	910	3.18%	1.62%	40.14	36.17	-3.97%	0.73%	138	Bush
AZ	1.36	1043	1859	2.27%	1.16%	46.60	45.03	-1.57%	8.75%	11	Bush
CA	0.46	12255	1919	2.22%	1.13%	55.73	55.21	-0.53%	32.14%	3	Bush
CO	1.73	2103	2515	1.95%	1.00%	49.07	47.35	-1.72%	4.20%		Bush
HI	0.47	426	499	4.38%	2.23%	53.32	54.37	1.05%	31.86%	3	Kerry
ID	1.32	590	559	3.91%	1.99%	33.33	30.71	-2.63%	9.39%	11	Bush
MT	0.12	440	640	3.78%	1.93%	39.28	39.51	0.22%	45.40%	2	Kerry
NM	1.70	748	1951	2.22%	1.13%	51.34	49.42	-1.93%	*4.43%	23	Bush
NV	1.83	816	2116	2.13%	1.09%	50.66	48.67	-1.99%	3.33%	30	Bush
OR	0.49	1810	1064	3.00%	1.53%	51.22	51.97	0.75%	31.32%	3	Kerry
UT	1.77	905	798	3.18%	1.62%	29.93	27.06	-2.87%	3.84%	26	Bush
WA	1.36	2815	2123	2.12%	1.08%	55.07	53.60	-1.47%	8.69%	12	Bush
WY	1.33	238	684	3.50%	1.78%	32.07	29.70	-2.37%	9.23%	11	Bush

AVG	1.26	24491	17637	2.91%	1.49%	51.81	50.87	0.94%	2.86E-07	3,500,172	
	

[TruthIsAll]

5. CORRECTION

I had this:
THESE BOGUS FINAL 13660 RESULTS WERE MATCHED TO THE VOTE
COUNT:
1:25pm National Exit (WEST region): - 50% (rounded)
1:25pm WEST State Exits-wtd (50.87%) <<< REPLACE "Exits" with "weighted average vote".

It should read:
FINAL 13660 RESULTS WERE MATCHED TO THE BOGUS VOTE COUNT:
1:25pm National Exit (WEST region): 50% (rounded)
Kerry's Western region state wtd average vote: 50.87%

[Bill Bored]

6. I'm comparing MoE to MoE.

Basically what you're saying is that in the West, the weighted state polls and the regional subset of the national poll agreed with each other within 1.2%. I don't have a problem with that.

My problem is simply that at least in the West, the national poll had to be adjusted beyond its MoE to match the vote count but the state polls for the same geographic region did not.

I'm not just basing this on Party IDs, I'm also basing it on the vote counts (polled and reported).

[TruthIsAll]

7. The south was padded at the expense of the west

http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=203&topic_id=362208&mesg_id=376144

[Time for change]

8. Here's the way I see it

When the polls were "adjusted" some time around 1 a.m. on November 3rd, in order to allow the TOTAL poll results to be consistent with the offical election results, EVERY sub-category, including each geographical region, had to be adjusted by a proportional amount -- whether or not the states in that region were within the margin of error.

[LightningFlash]

9. Think you hit the nail on the head with this one finally....n/t

[TruthIsAll]

10. TFC, you are absolutely correct. Its what I have been saying all along.

In fact, Mitofsky himself said it:
He re-weighted the final exit poll to match the votes.
That is a self-incriminating statement if there ever was one.

I guess when he said it, people had no idea what he really meant.
That is why the 43 Bush/37% Gore weighting was conceived in the How Voted in 2000 category. You can rest assure it was done in all the other categories. Either the weights or the vote percentages or both had to be adjusted.

Look at Party_ID:
13047: 38Dem /35 Repub
reweighted to
13660: 37 / 37

OTOH, the gender weights were not changed from 54 female/46 male, but...
Kerry's percentage of female voters dropped from
13047: 54%
to
13660: 51%

[OnTheOtherHand]

11. is that true?

Let's start with the official results. (After all, as we all know, I am an enemy of dem-- no, no, we're just trying to figure out what they are weighting to.)

I just navigated CNN.com and rounded on the fly to thousands, subject to QA/QC.

K B
AK 111 191
AZ 894 1104
CA 6745 5510
CO 1001 1101
HI 194 232
ID 181 409
MT 174 266
NV 397 419
NM 371 377
OR 943 867
UT 241 664
WA 1510 1305
WY 71 168
total 12833 12613

That's a 1-point Kerry margin. So, apparently a geographical weighting, not an arbitrary imposition of a national weighting.

My other thoughts get a bit technical, and start with the fact that the best geo estimates on E/M pp. 21-22 seem a bit bigger (albeit within MoEs), but the within-precinct errors (look, mom, no acronyms!) on 32-33 are much larger. The national poll is what, 250 precincts? so, 60-odd in the 13 Western states? (Maybe more like 50 -- looks like the West was about 20% of the national poll.) It's not so surprising if the initial regional estimates were noisier than the state estimates. (The n in California alone was 2399; the n for the west in the national poll was 2802.)

[LightningFlash]

12. The national state polls are 70,000 people. That's going state by state.

:thumbsdown:

[OnTheOtherHand]

13. yes, that's right, but what is your point?

That may help people to understand the apparent discrepancy that there were almost as many people interviewed in the CA state poll as in the entire Western region of the national poll. The national poll is a subset of precincts from the state polls, not the combination of all of them.

[LightningFlash]

35. No, it explains this point. Which you consistently ignore.

You can't cite faulty statistics, as they are facts. And numbers don't agree with any ad-hominem word arguments, since they're providing those statistics.

http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=203&topic_id=362208&mesg_id=376144

Your tacts have already refuted themselves far too many times. The fiscal result was padded from the "South" to the "West". If this was really due to a misreporting in the results, it would not be computational. You could in fact demonstrate a consistent pattern that would not falter, and in fact replicate itself in previous elections.

You can not, and the evidence falls on its own face. The result you issued was from South to West, and it was from no other place in variance that it happened. Not misreporting.

[OnTheOtherHand]

14. and speaking of QA/QC...

it's been pointed out that I awarded Hawaii to Bush (yecch). Strike that, reverse it. Still a 1-point margin (1.2 to two significant digits). Dunno if there are any other bloopers.

[Time for change]

15. Damn OTOH, you're such a naysayer!

Just kidding :) I hope this isn't deleted.

Actually, the final western poll could have been adjusted either on the basis of the Western region alone, or based on the national total, as I originally speculated -- I think. The actual adjustment was 7% -- from a 8% Kerry lead to a 1% Kerry lead. A national based western region adjustment would have required a 5.5% downward shift for Kerry (i.e., the national exit poll discrepancy) to 2.5% instead of 1%, but conceivably rounding errors could explain the slight difference?

Anyhow, I hadn't realized that a regional adjustment would come even closer to the actual adjustment, as you pointed out. So Bill's question still needs to be answered.

After looking at the exit poll data, along with the comparisons with the official results, I think it's safe to say the the apparent dilemma that Bill points out is the result of a combination of two things:

1) First, if you look at the actual state exit polls, rather than the western national sample of 2,544 that Bill depicts, Kerry did much worse -- in fact there's hardly any discrepancy at all with the official results. Given that the overall western region discrepancy between the exit polls and the official results is so small, it's not surprising that none of the western states were outside of the margin of error.

2) Secondly, even given the above, there were a couple of states exhibited fairly decent red shifts, comparable to the rest of the country. Rounding off, I come up with decent red shifts for AK (8), AZ (5), UT (5), ID (5), NM (4), CO (4), and NV (4). Yet, because of the relatively small numbers on a state by state basis, none of these quite reaches statistical significance, though some of them are very close.

So, that's the explanation.

But the really interesting thing that this demonstrates is the fact that Kerry did so much better in the Western region national polls than in the combination of western state polls, taken together. It's difficult to ascertain the statistical significance of this finding because I am not quite certain how best to calculate it. The main difficulty is that the Western region national poll probably consists of individual state contributions which are proportional to their electorate, whereas for the individual state polls the smaller states are relatively highly over-represented -- compared to a population based representation. But any way you calculate it, the difference is substantial.

I think that this could possibly be a very important finding -- if it turns out that the difference is highly statistically significant (and I think it is). The reason I suggest that is that the difference could be due to a difference in polling methodology, which (if we could pin down the reason) could give us a better understanding of the exit poll discrepancies in general.

[OnTheOtherHand]

16. relative red shift

I'm not sure which state-poll red shifts you are using. To elaborate on my earlier post, here is an anomaly to chew on (or maybe to spit out, I don't know) -- the unweighted CA results are very close to the official returns, and the "best Geo estimate" on p. 21 is only -3.0, but the average WPEs on p. 32 range from -10.9 to -13.8 depending on method.

(It may well be true that the region could have been weighted based on the national total, but E/M says they weight by "the best estimate of the candidate vote percentages from each geographic region." This language may only apply within states, but I see no reason why they wouldn't apply it to national regions, since they could. But then, E/M do and don't do lots of things that I don't see a reason for. Shrug. So, we don't know.)

[Time for change]

17. This is so bizzare

I started going through my papers to find an answer to your question about which state polls I used.

On November 11th Jonathon Simon posted results that he obtained from the internet, which became available because of a computer crash. I used his numbers (except for VA, WI, NY, NJ, and NC, for which I used other assorted sources, since Simon seemed unsure as to whether he had accurate data for those states) to compare with the official results. I then wrote a report based on that analysis and had my son, EOTE, post it on the DU, since I was not a member at the time: http://www.democraticunderground.com/discuss/duboard.php?az=show_topic&forum=104&topic_id=2682466. (Note that the title of the thread and the introductory paragraph, written by my son, are somewhat more adamant about the significance of the analysis than the report itself :) ) Unfortunately, the formatting for my table didn't work out very well. To find the magnitude of the red shift, look at the middle of the five numbers in the table. The number after that is the sample size, and then the p value.

Today I tried to find the original source posted by Jonathon Simon, and I did in fact find it: http://www.scoop.co.nz/stories/HL0411/S00142.htm . But the numbers are different than what is in my report!! I could understand a typo or two, but it looks like almost all the numbers are at least somewhat different. I specifically note the source in my report, and I can assure you that I didn't make those numbers up. I suppose that there is another explanation for this, but the two most likely that I can think of are 1) The numbers in Simon's report, dated November 11th, have been changed, or 2) I'm starting to lose my mind. Please help me out if you can, because I'd hate for the latter explanation to be true.

Furthermore, I did come across another, updated, and apparently more official report of the E-M exit poll results by state (on the internet) some days or weeks later (and I'm sure that many others did as well). These numbers were quite similar to, but not identical to the ones that Simon posted on November 11th (before they were changed). But now, when I try to find those reports, I keep getting "page not found".

With regard to the anomaly you found in California -- a similar anomaly exists for Nevada, but not for a lot of other states. Maybe Mitofsky got sloppy, but who knows. I think he did a lousy job of explaining what all those numbers were about in his report.

[OnTheOtherHand]

18. can you give me a specific anomaly?

It looks to me like the exit poll results are the same, some of the official results may differ, and the Simon "red shifts" are half your differences because he is doing a one-party shift. Or something like that.

[Time for change]

19. Thank you thank you

Yes, you're right. So the numbers weren't changed after all. And I'm only half losing my mind.

Anyhow, that was my source.

And incidentally, I notice that TIA posted a link on this thread with red shifts very similar to mine (when I looked at this post I remembered to multiply by 2), but with more complete data (i.e., for WI, NY, NC, NJ, VA). There are only four states that are off of Simon's results by more than 1%, most especially California). I'll have to ask where he got that info.

[Time for change]

21. The state vs. regional exit polls are way different for the West

As you noted, the last Western regional exit poll before the "adjustment" gave Kerry an 8 point lead, 53% to 45% in the West, with 2,130 sampled.

Looking at state by state polls, Kerry received only 48.6% of the two party vote, out of 17,078 sampled.

Even if you weight this by the size of the electorate (which you need to do because Kerry did much better in the larger states, like California), Kerry received only 52.4% of the vote, to 47.6%, a 4.8% difference.

The odds against that big of a difference between the state and the regional exit polls is one in 500.

I think that this strongly suggests that the regional and state exit polls may have been conducted differently, and this is important to know in trying to make sense of the exit polls.

[Bill Bored]

22. So this thread is still kicking around eh?

Well, to tell you the truth, I was going to imply that Mitofsky's precinct selection was bad, because the sample size of the national should have been sufficient to come up with a discrepancy close to that of the larger aggregate state sample, which in the West was small. Instead, we get 7 or 8 percent adjustments to match the vote count and it's beyond the MOE.

Mitofsky and USCV have both used the assumption that precinct selection was fine to justify both the rBr stuff, which I don't think anyone has actually proven yet, and the fraud theory, which both O'Dell and Baiman have said can't be proven by exit polls alone.

So now, we may even have another variable to consider, at least in the national poll: precinct selection.

That said, the larger sample with 1250 precincts, not including OR, is still usable, and we could compare this to the other regions to see if there are discrepancies similar to the West.

One question could be Oregon. Is it included in both the Western Region and the State totals? They vote my mail there and so have no actual "exit" polls and perhaps no precincts either.

[Time for change]

23. I don't think that precinct selection was the problem

The E-M report said that this was tested, and they found that it was slightly favorable to Bush (0.3%). This would make the overall red shift slightly more important than it already is, since on the basis of precinct selection alone you would expect a slight Blue shift. But the difference was small.

rBr is not justified in Mitofsky's current report -- it is simply stated as a fact, given Mitofsky's refusal to consider fraud as a possibility. Whether or not any evidence for rBr can be found in the mass of exit poll data is another question, for which we can only wait until information becomes available.

I believe that Oregon is not considered in either the national/regional or state exit polls. I base this on a DU post from Febble, who I believe is in a better to know:
http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=203&topic_id=371726&mesg_id=373379

Anyhow, my main point is this. Even though overall, the state and national exit polls seem to be comparable (Kerry won the national exit poll by 3.0% -- if you add 3% to his total from every state he would have won OH, IA, NM, and NV. Those are exactly the additional state exit polls that he won.), in the West there is a great disparity between state and national exit polls. To me, this strongly suggests that a different methodology is used. I think that this could be VERY important because Mitofsky's report concentrates on the national exit poll, so I would think that any future report that he puts out would do the same. Therefore, I am concerned that an analysis of the national exit poll data which ignores the state exit polls (especially Ohio, or the swing states in general) may be misleading.

[Bill Bored]

26. The 1250 precincts include the state polls.

While the 13,660 respondents are the national, all the WPE/Alpha/Bias Index stuff is based on the states: 1250 precincts.

There were actually 1469 precincts altogether though and there were 250 used for the national.

As usual, the numbers don't add up.

Why were > 200 precincts left out of the Jan. report (WPEs, etc.)? Another question for Febble and OTOH.

[Time for change]

27. Yes, this is very confusing

I just went back to look at the E-M report, and it says on page 31 that the national WPE that they calculated of -6.5 is based on 1460 precincts, sample size 114,559. Where do you get 1250?

[LightningFlash]

28. From official NEP data. 1,250 precincts reporting. n/t

[Time for change]

29. Thank you, do you have a link to that?

That might help me understand the discrepancy between your official NEP data statement and what is cited on page 31 of the E-M report. As Bill Bored suggests, that sounds somewhat suspicious.

[OnTheOtherHand]

30. sorry, I missed my name being called

The discrepancy between 1460 and 1250 is "explained" (sort of) on p. 34 of E-M:

"Note: The WPE values do not measure just the error of the exit poll in precincts that contain significant absentee vote. When absentees were greater than 15 % statewide, we removed precincts from this study that had the absentees merged with the precinct vote. In these precincts we cannot obtain counts of the election day vote separate from the absentee vote. Also, not included in this study are any precincts with fewer than 20 interviews as well as three additional precincts with large absolute WPE (112, -111, -80) indicating that the precincts or candidate vote were recorded incorrectly. Out of the 1,480 exit poll precincts, 1,250 were included in the analysis that follows."

(Why 1480? just before this it says that for some reason they couldn't get precinct returns for 20 precincts.)

In my aggregation, there are 71 precincts with fewer than 20 interviews. So it appears that most of the 'attrition' had to do with absentee voting.

[Time for change]

31. Thank you for the clarification OTOH

[mgr]

45. If the attrition is due to absentee voting,

It may be preponderate in the West.

Mike

[Bill Bored]

32. Page 36 of the same report

and other pages, and USCV and Febble all use 1250.

If the 1460 WPE was -6.5, we should see if this agrees with the 1250.
This could be done by taking a weighted average of the WPEs in the table at the bottom of page 36 which uses 1250. When I do this, I get -6.7744.
This means that the 210 missing precincts had a lower mean WPE. Since it seems that most of these had a lot of absentee votes, does this mean anything?

[Time for change]

33. I think OTOH has the explanation for the discrepancy between 1460 and 1250

See post # 30. According to his explanation, which sounds reasonable, 1250 was used, rather than 1460, for all of the WPE calculations.

I don't understand what you mean by a weighted average of the WPEs on page 36. What do you use to weight this?

[OnTheOtherHand]

34. Bill has a point

It's not quite right that 1250 was used for all of the WPE calculations.

The -6.5 is calculated against all 1460 precincts for which returns were available, see p. 31. Probably also used for comparisons with prior exit polls, but I haven't squinted at that.

They pared down to 1250 precincts for the various tables.

So Bill is probably right that the mean WPE is different in the 210 precincts that were dropped -- well, almost certainly it was different. I have no idea whether the diff is substantively or statistically significant.

[Bill Bored]

36. They didn't give an average for all 1250

so you can calculate it by weighting the average WPE in each band of partisanship in the table by the number of precincts in each. That's -6.7744% or something. Higher WPE than the -6.5.

That said, if the 210 precincts were mostly absentee and early voting, how do we also explain THIS statement, from the FAQ of some E-M rag or another. (I'm not sure which rag because someone PMed me with it, but it looks authentic. It's either from the Jan report or the methodology statements.)

"How do absentee votes affect projections?
In a number of states significant numbers of voters cast their ballots before Election Day. They either cast an absentee ballot or they vote at precincts set-up for voters in the weeks leading up to the election. These absentee/early voters cannot be included in the Election Day. Instead, they are interviewed by telephone shortly before the election. They are asked the same questions as voters interviewed in the exit polls. For the 13 states where absentee voters will be sampled by telephone their results will be combined with those of Election Day voters"

So they can't have it both ways. Either they threw these precincts away, or they combined them with the election day results, but didn't count them as precincts!

Have I mentioned how much I hate this exit poll stuff?

There are now 210 phantom precincts that were either included or excluded from election day results because they had a lot of absentee or early voting activity.

Again, see OTOH's post 30.

BTW, this is not an insignificant percentage of precincts: 210/1460=14.4%, but it's still only a fraction. And yet they had a sufficiently lower WPE than the 1250 used in the calculations to bring the total average from -6.7744 to -6.5. Since this change was caused by only 14.4% of precincts, their mean WPE must have been much less than 6.25%! Another weighted average calculation would show this. (Can we do this with medians?)

What this implies is that telephone polls of those who have already voted MAY BE a lot more accurate than exit polls.

[Time for change]

37. I think it would be good to know how the absentees were handled

My belief was that Kerry did much better with absentees than on election day. That was definitely the case in Snohomish and Yakima County, WA, for which Paul Lehto did an extensive analysis. Anyhow, I think that this would be important to know.

Unlike you, I am very interested in this exit poll stuff, because I feel that it may provide some clues to what happened on Nov. 2nd. But if you hate it, why do you keep on discussing it and starting threads on it?

[Bill Bored]

40. Because I'm a glutten for punishment?

Actually because I have a hope that when the real fraud is eventually uncovered, it will be shown to be consistent with something found in the exit polls.

Story County, IA is another prime example.

[OnTheOtherHand]

38. on absentees

Bill, I am way too distracted to get this one exactly right, but it seems to me you are making it even more confusing than it has to be.

When E/M are doing election day projections, they incorporate all the information they have from both the in-person exit polls and the telephone polls of early and absentee voters.

When E/M is analyzing WPE after the fact -- or, more specifically, when they are presenting tables of that analysis -- they throw out precincts where a lot of early or absentee votes have been incorporated, because those could obviously throw off the analysis of precinct error.

So, they can have it both ways. If you can imagine what it's like to do election day projections (I did this 20 years ago on a programmable calculator), lots of information is getting thrown into the stew. There is no time to analyze precinct-level WPE data for signs of fraud, or whatever -- they are just trying to incorporate all their data sources (including official returns, assumed reliable apart from egregious innocent error) as quickly as possible.

Now, about the calculation: We can do a back-of-the-envelope, crossing our fingers that the precincts thrown out aren't radically bigger or smaller than the rest, just like this: one-seventh of the precincts were thrown out. The average among the other six-sevenths was about 1/4 point more negative (we don't know how close -6.5 is to -6.50). So It is likely that the remaining one-seventh was about 1 1/2 points more positive. That puts them around -5 WPE. Did I do that right?

Probably not statistically robust. But if subtle selection bias by the interviewers was one source of error, it seems reasonable that a telephone poll will pretty much wipe it out. One voter may seem more friendly than another, but phone numbers are pretty neutral. (But people may be more reluctant to report their votes to a stranger over the phone than to check a box on a piece of paper.)

[Bill Bored]

39. Yes about -5%

But if these 210 precincts were a MIX of absentee and polling place voting, the WPE of the absentee votes might be much lower, and this includes early voting.

I wonder how this compares to previous years when there was little or no e-voting?

As you may know, I think Early Voting in particular provides an unprecedented opportunity for election fraud. If anyone knew the likely outcome of a race BEFORE Election Day, by having some actual pre-Election Day results, it would be a simple matter to alter the Election Day programming or results to reverse the early trend. The whole thing can be explained away by "aggressive Early Voting Campaigns" by the losing party, as has actually been done in the press in 2004.

One county in IA, Story County, had Kerry winning the early vote by a large margin, only to see it nearly evaporate on Election Day. He lost the state by only 10,000 votes, thousands of which were from this one county. And they had paper Op Scan ballots which could have been hand counted, but since one lousy state won't swing the Electoral College, I guess no one thought it was worth it.

I'm sure this isn't the only county or the only state where such discrepancies occurred although there may be demographic confounders which could be used to explain it all away. Nevertheless, Early Voting is a scam and should be abolished in favor of other ways to make it easier for voters to reach the polls, IMO.

Also, see this thread about early votin' in GA:
<http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x380457>

[OnTheOtherHand]

41. yup, good point

I coulda sworn I saw some figures somewhere about proportions of absentee/early votes in various states -- you could sort some of this out perhaps. See pp. 72-73 of E/M for a start.

My advice about Story County, IA, about which I know nothing, is to start indeed by comparing to previous years; that should give you some clue whether there is an outrageous discrepancy or not. Here's a link that indicates that Kerry doubled Gore's margin there
http://www.citizenwhip.com/NewsDetail.asp?Key=25
so it doesn't seem very likely that Bush stole huge numbers of votes there, but I dunno.

(edit: specifically, the link says: "Story County – Kerry wins by 2,366 votes (1,116 more than Gore), Nader lost 1,099 votes." Kerry won it about 52%-47%. So the "doubling" is not hugely impressive -- my point is only that Kerry's margin was wider, whatever that means.)

I have mixed feelings about Early Voting, but if we can replace it with other means, I'm all for that.

[Bill Bored]

42. Before Election Day, Kerry was ahead by 4,000 votes If mem. serves me.

Maybe more than that. There was also some controversy about some of the Early Voting places being closed early, or eliminated. Then the election day narrowing of the margin and losing the whole state by only 10,000 votes. I don't like the smell of it.

I also know for a fact that there were Dems for Bush in parts of IA. They registered Dem to vote for their friends in the caucuses for some local races, but they were unflappable Bushies, totally brainwashed by the so-called war on terror. Like someone had flown a 767 into a cornfield or something!

I bet Radar O'Reilly (also from Iowa) wouldn't have fallen for this stuff, and he'd have known those planes coming ahead of time too! But then so did the Bushies, around Aug 6 or so.

[OnTheOtherHand]

43. hey, we have a friend in Ames

(it took a while for me to register that Story County is Ames, and she moved there from more western Iowa not long ago) -- she seems pretty well plugged in -- so if there are any questions you would like her to ask around about, send me a message.

Iowa probably "should have been" pretty close
http://www.electoral-vote.com/2004/states/iowa.html
but heck if I know how close.

I could explain why TIA is wrong about the conflict between 43/37 and rBr, too. rBr (or all that participation bias stuff) actually tends to explain why the recalled Bush/Gore gap is much smaller than it has been in other surveys. It sounds weird, but hey, I didn't make up the other surveys. Anyway....

[Bill Bored]

44. It's the home of ISU too.

And the Aimes strain of Anthrax no doubt.
But I digress.

I think only a full hand recount would answer this question.

As far as Y2K voters, Bush, Gore, Kerry, etc. I really haven't gotten into the latest one if it's any different than the other iterations of this. That stuff just gives me a headache. Too many contradictions. The polls are wrong (Gore won the popular vote!) but we're saying they're right for one thing. And besides, I don't really care who won the popular vote anyway! It's really a moot point since there's no right for President in that context.

[Bill Bored]

46. No right to vote for President that is. nt

[Time for change]

24. Correction

My percentages were basically correct, but I miscalculated the test of statistical significance. The odds against that big of a difference between the state and regional exit polls (occurring by chance, that is) is one in 12, rather than one in 500. So, it isn't as big a deal as I thought.

[LightningFlash]

25. They're pretty much congruent with one another. NEP or state. n/t

[kiwi_expat]

47. Ohio's Cuy and Ham. counties: NATIONAL precincts have the highest WPEs

I know this is of absolutely no statistical importance, but it could be indicative, nevertheless:

On minvis' "Ohio Exit Poll Raw Data" thread, we have been able to identify the actual precinct names for the Hamilton (Cincinnati) and Cuyahoga (Cleveland) NEP data - and thus calculate the WPEs for those precincts.

Four of the 5 National exit poll precincts in those two counties have higher WPEs (28%, 12%, 12%, 9%) than those counties' 8 State-only exit poll precincts (0% to 8% WPE).

Maybe it is more difficult to administer the 4 (intermixed) National questionnaires than the single State questionnaire.


Has Mitofsky reported any WPE information on the individual states' data, by any chance?

[kiwi_expat]

48. I should have shown the WPE signs.

I should have said:
"Four of the 5 National exit poll precincts in those two counties have larger WPEs (-28%, -12%, -12%, -9%) than those counties' 8 State-only exit poll precincts (-8% to 8% WPE)".


I also see that Mitofsky has indeed reported the WPE information by individual states. (Page 32 of the E-M Report, as mentioned in post #16.) As others have pointed out, what is missing is the WPE information for the National exit poll.

[TruthIsAll]

51. Kiwi, take a look at the Exit Poll Response Optimizer

If there is a way to apply it to states, I will try to do so.

EXIT POLL RESPONSE OPTIMIZATION MODEL						
6/29/05 2:21 AM						
						
OBJECTIVE:						
Determine values of constrained variables required to derive
the Bush recorded 51.24% (2-party) vote using aggregate exit
poll response data						
						
Precinct Variable Input Range (Min, Max) Constraints:         
                                         	
1-Response rates =.53                                        
		
2-Bush 51.24%
3-Alpha (K/B) = 1.155  (abs. minimum required alpha;
Mitofsky's 1.12 is NOT FEASIBLE)				

4-WPE:  Min= Max= E-M (match to Mitofsk's -6.77%)						
						
VOTE	Mil.	Pct (Input)				
Kerry 	59.027	48.76%				
Bush	62.029	51.24%				
Total	121.056		

POLL	Mil.	Pct	
Kerry 	63.127	52.15%	
Bush	57.929	47.85%	

Bush-recorded	51.24%		

EXIT POLL RESPONSE (input)			
Aggregate	53.0%	R	
Kerry/Bush	1.155	alpha	
This is minimum alpha required to satisfy 			
the WPE/partisanship constraints			

PROBABILITY			
Bush vote gain	1.59E-11		
 1 in	62,953,509,332					
Kerry > 50% vote	100.0%					
 1 in	1					
						
PARTISAN ALPHA						
Kerry	1.062					
Bush	1.185					
						
RESPONSE INPUT CONSTRAINTS						
1250	Strong Bush		Strong Kerry			
Prcts	40	415	540	165	90	
						
KERRY WIN%						
Min	0%	20%	40%	60%	80%	
Max	20%	40%	60%	80%	100%	
						
RESPONSE						
Min	50%	50%	50%	50%	50%	
Max	56%	56%	56%	56%	56%	
						
ALPHA (K/B)						
Min	0.500	0.500	0.500	0.500	0.500	
Max	10.000	10.000	10.000	10.000	10.000	
						
WPE						
E-M	-10.00%	-6.1%	-8.5%	-5.9%	0.3%	
Min	-10.00%	-6.1%	-8.5%	-5.9%	0.3%	
Max	-10.00%	-6.1%	-8.5%	-5.9%	0.3%	
						
						
		OPTIMIZER OUTPUT SUMMARY				
						
	Vote Percentages		Votes (millions)	
	Poll	Vote	Diff	Poll	Vote	Diff 
Kerry	52.15%	48.76%	-3.39%	63.127	59.027	-4.100
Bush	47.85%	51.24%	3.39%	57.929	62.029	4.100
Diff	4.29%	-2.48%	-6.77%	5.199	-3.002	-8.201
						
	 HighB	Bush	Even	Kerry	HighK	Total/Avg
Prct	40	415	540	165	90	1250
Vote 3.874	40.191	52.296	15.979	8.716	121.056
Pct	3.2%	33.2%	43.2%	13.2%	7.2%	100.0%
						
RESP.	50.0%	56.0%	52.3%	50.0%	50.0%	53.0%
Dev	-3.0%	3.0%	-0.7%	-3.0%	-3.0%	0.0%
						
ALPHA						
K/B	1.520	1.153	1.174	1.098	0.996	1.155
Dev	31.6%	-0.2%	1.6%	-4.9%	-13.7%	0.0%
ln a	0.418	0.142	0.160	0.093	-0.004	0.144

VOTE						
Kerry	0.75	16.04	25.62	9.63	7.00	59.027
Pct	19.2%	39.9%	49.0%	60.3%	80.3%	48.76%
						
Bush	3.13	24.15	26.68	6.35	1.72	62.029
Pct	80.8%	60.1%	51.0%	39.7%	19.7%	51.24%
						
RESPONDERS						
Kerry	0.94	17.26	27.84	10.10	6.98	63.127
Pct	24.2%	43.0%	53.2%	63.2%	80.1%	52.15%
						
Bush	2.93	22.93	24.46	5.88	1.73	57.929
Pct	75.8%	57.0%	46.8%	36.8%	19.9%	47.85%
						
REFUSERS						
Kerry	14.2%	36.0%	44.3%	57.3%	80.4%	44.92%
Bush	85.8%	64.0%	55.7%	42.7%	19.6%	55.08%

VOTE DEVIATION						
Kerry	-0.19	-1.23	-2.22	-0.47	0.01	-4.10
						
WPE						
Calc	-10.0%	-6.1%	-8.5%	-5.9%	0.3%	-6.774%
E-M	-10.0%	-6.1%	-8.5%	-5.9%	0.3%	-6.774%
Diff	0.00%	0.0%	0.0%	0.0%	0.0%	0.0%

[Bill Bored]

49. Yes and no.

Hi Kiwi! You owe me an email or PM don't you?

The points on the scatterplots use different color/shape combinations to represent the different states. If you look at them closely, you might be able to pick out OH.

Here's a page that links to three of them! Nice and large:

http://www.dailykos.com/story/2005/5/24/213011/565

Say hi to Minvis, et al. (Where is that thread anyway?)

Have you guys seen the ESI OH Exit Poll analysis? (They promised to explain it someday because there a few of us who didn't actually ATTEND the AAPOR conference, although I can't imagine why.)

[kiwi_expat]

53. Bill, thanks for the link to dailykos .....

unfortunately I'm not able to use it, for some reason. Could you possibly please give us the scatterplots links from the dailykos page? Thanks!!

I've just kicked minvis' OEPRD thread back to the front page for you. :-)

Do you have a link for the ESI OH Exit Poll analysis??


I'll PM you tonight.

Regards
kiwi

[TruthIsAll]

50. Your probability calc is off

And I am using a very conservative MoE based on 2130 respondents.
MoE = 1/sqrt(2130)
MoE =0.02166757

The probability of the difference is
=NORMDIST(0.4846,0.524,0.0216/1.96,TRUE)
=0.000175029

= 1 in 5713