STATE / NATIONAL EXIT POLL ANALYSIS BY REGION

[TruthIsAll]

										
SUMMARY REGIONAL STATE EXIT POLL STATISTICS

Note:
Vote = actual vote in thousands
Size = State exit poll sample size
Z    = Bush Z-score (standard deviations from the mean) 
MoE    = margin of Error
StDev  = standard deviation from the mean
Poll(p)= Kerry's exit poll percentage
Vote   = Kerry's vote percentage
Dev    = Poll - Vote spread
Prob   = Probability of deviation based on Z-score 

Formulas:
Stdev  = sqrt(p*(1-p)/n)
MoE    = 1.96* StDev 
Z      = Dev/ StDev

Prob = 1 - NORMSDIST(Z)
Regional Z-scores are approximated by averaging the states.

Kerry deviated from 59.60% in the East (using state exit poll
data) to 56.09% in the vote, the largest regional deviation.

Region	Z	Vote	Size	MoE	StDev	Poll	Vote	Dev	Prob  	1 in
EAST	2.22	26854	16261	2.75%	1.40%	59.60	56.09	-3.51%	9.49E-13	1,053,596,824,745
MIDWEST	1.87	30676	19377	2.61%	1.33%	50.28	48.68	-1.60%	3.60E-03	278
SOUTH	1.81	39034	20332	2.56%	1.31%	43.75	42.33	-1.42%	1.41E-06	709,051
WEST	1.26	24491	17637	2.91%	1.49%	51.81	50.87	-0.94%	2.86E-07	3,500,172

TOTAL	1.79	121055	73607	2.71%	1.38%	50.55	48.72	-1.82% 

......................................................

These are the probabilities that Bush would exceed the given
Z-score and the number of states it was exceeded in each
region, based on the deviation from the exit poll to the vote:

EAST    Z> 2.23 in 8 of 13 states = 1 in 1,053,596,824,745
MIDWEST Z> 1.07 in 6 of 12 states = 1 in 278
SOUTH   Z> 1.21 in 9 of 13 states = 1 in 709,051
WEST    Z> 1.32 in 9 of 13 states = 1 in 3,500,172

The Probability is calculated using the Excel Binomial
Distribution function:
Prob   = 1- BINOMDIST (n-1, N, p(Z), True)
Where n= number of states in the region in which Z>X,
  and N= total number of states in the region.

p(Z) = probability of the deviation between the vote and exit
poll as measured by the Z-score in X standard deviations.
p(Z=X) = 1 - NORMSDIST(X)
True = Cumulative distribution parameter

................................................	
											
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%
MIDWEST   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						

..................................................


CALCULATION OF REGIONAL WEIGHTS 
USING TOTAL VOTE & STATE EXIT POLL DATA

The South had a 27.62% weighting in the 13047 National Exit
Poll but was 32.24% of the vote, a very odd occurrence.

Region	  Vote	Pct	Exit	Pct	Bush  1 in
EAST	 26854	22.18%	16261	22.09%	1,053,596,824,745		
MIDWEST	 30676	25.34%	19377	26.32%	278			
SOUTH	 39034	32.24%	20332	27.62%	709,051
WEST	 24491	20.23%	17637	23.96%	3,500,172
	
TOTAL 121055  100.0%	73607	100.0%							
											
......................................................														

STATE EXIT POLL REGIONAL WEIGHTS 
APPLIED TO NATIONAL EXIT POLL 13047
Kerry wins by 50.83-47.66%

Region        Exit	Pct	Bush	Kerry	Nader
EAST	      16261	22.09%	6667	9431	163
MIDWEST	      19377	26.32%	9495	9689	194
SOUTH	      20332	27.62%	10979	8946	203
WEST	      17637	23.96%	7937	9348	176

TOTAL	    73607	100.0%	35078	37414	736
		             47.66%	50.83%	1.00%						
..................................................											

FINAL RECORDED VOTE REGIONAL WEIGHTS
APPLIED TO NATIONAL EXIT POLL 13047
Kerry wins by 50.45-48.03%
		
Region        Vote	Pct	Bush	Kerry	Nader
EAST	       26854	22.18%	11010	15575	269
MIDWEST	       30676	25.34%	15031	15338	307
SOUTH	       39034	32.24%	21078	17175	390
WEST	       24491	20.23%	11021	12980	245

TOTAL	      121055	100.0%	58141	61069	1211
      	                     48.03% 50.45% 1.00%						
...........................................................



STATE EXIT POLL AND VOTE STATISTICS BY REGION

Z      = Z-score
Votes  = votes in thousands
Size(n)= state exit poll sample size
MoE    = margin of Error
StDev  = standard deviation from the mean
Poll(P)= Kerry's exit poll percentage
Vote   = Kerry's vote percentage
Dev    = Poll - Vote spread
Prob   = Probability of deviation based on Z-score 

Formulas:
Stdev  = sqrt(p*(1-p)/n)
MoE    = 1.96* StDev 
Z      = Dev/ StDev

Prob = 1 - NORMSDIST(Z)

Region	Z	Votes	 Size  MoE	StDev	Poll	Vote	Dev	Prob 	 1 in	For
EAST											
CT	1.94	1551	872	3.27%	1.67%	58.47	55.24	-3.24%	2.62%	38	Bush
DC	1.02	372	795	1.92%	0.98%	91.63	90.63	-1.00%	15.40%	6	Bush
DE	2.60	224	770	3.48%	1.78%	58.44	53.82	-4.62%	0.46%	216	Bush
MA	2.38	2875	889	3.10%	1.58%	66.46	62.70	-3.76%	0.88%	114	Bush
MD	0.50	2359	1000	3.07%	1.57%	57.04	56.25	-0.79%	30.71%	3	Bush
ME	0.32	727	1968	2.20%	1.12%	54.83	54.48	-0.36%	37.58%	3	Bush
NH	4.16	672	1849	2.27%	1.16%	55.50	50.68	-4.81%	0.00%	63547	Bush
NJ	2.36	3581	1520	2.49%	1.27%	56.13	53.13	-3.00%	0.92%	108	Bush
NY	4.10	7277	1452	2.47%	1.26%	63.97	58.79	-5.17%	0.00%	49430	Bush
PA	2.89	5732	1930	2.22%	1.13%	54.41	51.13	-3.28%	0.19%	521	Bush
RI	2.23	429	809	3.30%	1.69%	64.24	60.48	-3.76%	1.29%	77	Bush
VT	2.95	305	685	3.56%	1.81%	65.69	60.34	-5.35%	0.16%	628	Bush
WV	1.43	750	1722	2.35%	1.20%	45.19	43.48	-1.72%	7.61%	13	Bush

AVG	2.22	26854	16261	2.75%	1.40%	59.60	56.09	-3.51%	9.49E-13	1,053,596,824,745	
									
MIDWEST											
IA	1.13	1494	2502	1.96%	1.00%	50.67	49.54	-1.13%	12.93%	8	Bush
IL	1.61	5239	1392	2.60%	1.33%	57.13	54.99	-2.14%	5.35%	19	Bush
IN	0.94	2448	926	3.17%	1.62%	40.97	39.46	-1.51%	17.44%	6	Bush
KS	1.27	1171	654	3.65%	1.86%	34.60	36.97	2.37%	10.15%	10	Kerry
MI	0.81	4793	2452	1.98%	1.01%	52.55	51.73	-0.82%	20.89%	5	Bush
MN	2.67	2792	2178	2.09%	1.07%	54.61	51.76	-2.85%	0.38%	263	Bush
MO	1.07	2715	2158	2.11%	1.07%	47.48	46.33	-1.15%	14.28%	7	Bush
ND	1.35	308	649	3.63%	1.85%	33.58	36.09	2.51%	8.80%	11	Kerry
NE	2.33	767	785	3.37%	1.72%	36.54	32.53	-4.01%	0.99%	102	Bush
OH	2.94	5599	1963	2.21%	1.13%	52.06	48.75	-3.31%	0.17%	602	Bush
SD	1.34	382	1495	2.45%	1.25%	37.42	39.09	1.67%	9.09%	11	Kerry
WI	0.01	2968	2223	2.08%	1.06%	50.21	50.20	-0.02%	49.44%	2	Bush

AVG	1.87	30676	19377	2.61%	1.33%	50.28	48.68	-1.60%	3.60E-03	278	
											
SOUTH											
AL	2.20	1870	730	3.57%	1.82%	41.08	37.08	-4.00%	1.40%	72	Bush
AR	1.64	1998	1402	2.61%	1.33%	46.93	44.74	-2.19%	5.01%	20	Bush
FL	2.63	7548	2846	1.84%	0.94%	49.93	47.47	-2.46%	0.43%	231	Bush
GA	1.21	3280	1536	2.48%	1.26%	43.11	41.58	-1.53%	11.33%	9	Bush
KY	0.50	1782	1034	3.00%	1.53%	40.76	39.99	-0.76%	30.92%	3	Bush
LA	1.54	1922	1669	2.38%	1.22%	44.50	42.63	-1.87%	6.22%	16	Bush
MS	1.88	1130	798	3.44%	1.75%	43.20	39.91	-3.30%	3.00%	33	Bush
NC	3.35	3487	2167	2.10%	1.07%	47.31	43.72	-3.60%	0.04%	2508	Bush
OK	0.24	1464	1539	2.38%	1.21%	34.73	34.44	-0.29%	40.59%	2	Bush
SC	3.75	1600	1735	2.34%	1.20%	45.79	41.31	-4.48%	0.01%	11104	Bush
TN	1.39	2421	1774	2.29%	1.17%	41.15	42.78	1.63%	8.19%	12	Kerry
TX	1.40	7360	1671	2.31%	1.18%	36.84	38.49	1.65%	8.08%	12	Kerry
VA	1.75	3172	1431	2.59%	1.32%	47.96	45.65	-2.31%	4.01%	25	Bush

AVG	1.81	39034	20332	2.56%	1.31%	43.75	42.33	-1.42%	1.41E-06	709,051	
											
WEST											
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	
											
											
											

[dzika]

1. Geographic z-Score Map (state exit polls)

Using the z-scores posted above by TIA...



Full map with Alaska

[TruthIsAll]

5. STATE EXIT POLL SAMPLE SIZE COMPARED TO VOTING WEIGHT

Percentage Comparison of State Exit Poll respondents to
Regional Voting Percentages and Zscores 

Purpose: To compare state exit poll sample size to actual vote
weight.

(1) StVote = state vote % of region total
(2) StEx = state exit respondents as % of regional total
(3) StVote % Exit = (1)/(2)
(4) ZS = Z-score (standard deviations from Mean)
(5) ZS * (3) = a measure of deviation magnitude

	StVote StEx	StVote	      Zs *
	%Tot   %Tot	%Exit	Zs	StV%StEx
EAST					
CT	5.78%	5.36%	108%	1.94	2.09
DC	1.39%	4.89%	28%	1.02	0.29
DE	0.83%	4.74%	18%	2.60	0.46
MA	10.71%	5.47%	196%	2.38	4.65
MD	8.78%	6.15%	143%	0.50	0.72
ME	2.71%	12.10%	22%	0.32	0.07
NH	2.50%	11.37%	22%	4.16	0.92
NJ	13.34%	9.35%	143%	2.36	3.36
NY	27.10%	8.93%	303%	4.10	12.46
PA	21.35%	11.87%	180%	2.89	5.20
RI	1.60%	4.98%	32%	2.23	0.72
VT	1.14%	4.21%	27%	2.95	0.80
WV	2.79%	10.59%	26%	1.43	0.38
Total	100%	100%
Avg	22.18%	22.09%	100%	2.22	2.23
					
MIDWEST
IA	4.87%	12.91%	38%	1.13	0.43
IL	17.08%	7.18%	238%	1.61	3.83
IN	7.98%	4.78%	167%	0.94	1.56
KS	3.82%	3.38%	113%	1.27	1.44
MI	15.62%	12.65%	123%	0.81	1.00
MN	9.10%	11.24%	81%	2.67	2.16
MO	8.85%	11.14%	79%	1.07	0.85
ND	1.00%	3.35%	30%	1.35	0.41
NE	2.50%	4.05%	62%	2.33	1.44
OH	18.25%	10.13%	180%	2.94	5.29
SD	1.25%	7.72%	16%	1.34	0.22
WI	9.68%	11.47%	84%	0.01	0.01
Total	100%	100%
%Total	25.34%	26.32%	96%	1.87	1.80
					
SOUTH					
AL	4.79%	3.59%	133%	2.20	2.93
AR	5.12%	6.90%	74%	1.64	1.22
FL	19.34%	14.00%	138%	2.63	3.63
GA	8.40%	7.55%	111%	1.21	1.35
KY	4.57%	5.09%	90%	0.50	0.45
LA	4.92%	8.21%	60%	1.54	0.92
MS	2.89%	3.92%	74%	1.88	1.39
NC	8.93%	10.66%	84%	3.35	2.81
OK	3.75%	7.57%	50%	0.24	0.12
SC	4.10%	8.53%	48%	3.75	1.80
TN	6.20%	8.73%	71%	1.39	0.99
TX	18.86%	8.22%	229%	1.40	3.21
VA	8.13%	7.04%	115%	1.75	2.02
Total	100%	100%
%Total	32.24%	27.62%	117%	1.81	2.11
					
WEST					
AK	1.23%	5.16%	24%	2.44	0.58
AZ	4.26%	10.54%	40%	1.36	0.55
CA	50.04%	10.88%	460%	0.46	2.13
CO	8.59%	14.26%	60%	1.73	1.04
HI	1.74%	2.83%	61%	0.47	0.29
ID	2.41%	3.17%	76%	1.32	1.00
MT	1.80%	3.63%	50%	0.12	0.06
NM	3.05%	11.06%	28%	1.70	0.47
NV	3.33%	12.00%	28%	1.83	0.51
OR	7.39%	6.03%	123%	0.49	0.60
UT	3.70%	4.52%	82%	1.77	1.45
WA	11.49%	12.04%	95%	1.36	1.30
WY	0.97%	3.88%	25%	1.33	0.33
Total	100%	100%
%Total	20.23%	23.96%	84%	1.26	1.06
					
					
					
Exit sample weight				

EAST	22.18%	22.09%	100%	2.22	2.23
MIDWT	25.34%	26.32%	96%	1.87	1.80
SOUTH	32.24%	27.62%	117%	1.81	2.11
WEST	20.23%	23.96%	84%	1.26	1.06

Total	100.0%	100.0%			

[dzika]

6. "STATE EXIT POLL SAMPLE SIZE COMPARED TO VOTING WEIGHT" MAP

Full map

TIA, could you give some more interpretation for your results on this one?

[TruthIsAll]

8. Exit Poll Weights and the Z-score

I feel that two major factors are jumping out of the exit poll analysis. The number polled in each state was not proportional to the voting population. We should not expect them to be, since the margin of error is a function of sample size, not population size. But why not illustrate which states were "under-polled", since they are important as far as the popular vote is concerned.

In addition, let's factor in the Zscore, a multiple of the number of standard deviations from the sample exit poll mean and the vote.

The product of the Z-score and the polling weight is one way to dramatize those states which are of most concern, by virtue of bot population size and discrepancy from the exit poll.

[Melissa G]

13. Cool Graphics Dzika! n/t

[ccarter84]

7. Thank you both for your hard work....

especially thanks to dzika for giving graphical represenation to the data & to TIA for the relentless perseverance of this issue.
-CC

[rumpel]

10. Ditto. Nice visual presentation, dzika.

[liam_laddie]

2. exit poll totals

TIA - thanks in excessive amounts for all your work...

Ques. - does the 73,307 total of exit-poll interviews reflect:
1) the "completed responses" only? Or were there more
attempted interviews? I think E/M expects about a 45-55%
not-completed rate.
2) the mystery 660 voters added overnight? How could these
be identified? Might it mean an analysis of the last raw data
total interviews by precinct, and see where, if identified in the
"final" poll results, these were "added" (so to speak)?

There is anecdotal evidence here, Hamilton County, Ohio, that
in precinct 4-M, the "bush reluctant responder hypothesis" had some legs; this according to three Dem observers who were outside, with or near the pollster, at various shifts during the day. There were many non-participants who were "Republican types" according to his comments to my observers. And Kerry seemed to have "about a 60-40 margin by mid day..." He apparently did not poll only 4-M voters, the target precinct, but likely and inadvertantly mixed in the other three precincts voting at same school. It was raining hard off-and-on that day, and the weather in other regions may have affected this phenomenon, too.

The sample was only 31 (21-10 Kerry) so it was about 2/3 desired total (40-50, I believe.) Bush carried 4-M, 312-211, with 101 of these shown as absentees (19%) a quite high rate, in my book. 4-M's absentees are 15.8% of registrations 101 / 640
Ham Cty as a whole had a 10.7% rate, of ballots cast, and 7.9% of registrations.
So 4-M had about double absentee rate.

How large a correction factor could be applied to change the last raw data results to account for any BRRH effect, and still give Kerry a win? I mention this only because the final raw data numbers seem to be a constant in your calcs, and may not, in fact, be a true constant. By what absolute number revision could
this factor be included and still have Kerry on top? I am most definitely not a statistician / analyst, but have been only collecting data here for others to crunch.

Please keep crunching yourself!
liam_laddie

[dzika]

3. Finding the final 660 responders

NEP released a dataset for the raw responses in the National Exit Poll. This dataset includes the final 660 responses. There is no timestamp on each record so it's not easy to determine which records were in the last 660 to be added to the sample.

NEP does provide a weight for each record that must be applied to the raw sample.

I've been wondering if there would be a way to take the 660 responses with the highest weights (towards Bush) and determine if that set of records would be enough to explain the discrepancy between the 12:20am and 1:40am exit polls.

This is the only idea I've had so far for trying determine which responses were in the earlier samples. I have been assuming that NEP probably re-weighted all of the samples so there was no point in trying this.

[passy]

17. As for the 660 added they don't exist!

http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=203&topic_id=366605&mesg_id=366730
In this post i showed how the numbers don't make sense.

"In the 13047 poll around 3804 women had voted for Kerry. (13047x54%x54%)
In the 13660 poll around 3761 women had voted for Kerry. (13660x54%x51%)
That means that 43 women who had voted for Kerry were removed from the final poll.
In the 13047 poll around 2820 men had voted for Kerry. (13047x46%x47%)
In the 13660 poll around 2764 men had voted for Kerry. (13660x46%x44%)
That means that 56 women who had voted for Kerry were removed from the final poll.
Altogether 99 persons voting for Kerry were removed from the final group used for the poll of 13660.
That also means that an extra 712 * voters were added to the final poll numbers.
It doesn't make sense if all of the people in the group of 13047 are also supposed to be in the group of 13660."

You could go deeper in the numbers and find that the re-weighting doesn't make any sense.
I think we can simply dismiss the 13660 exit poll as being irrelevant. It's a waste of time to try and analyze the re-weighting used in that poll.

This was some of the categories that were increased: Southern voters, 30-44 year old voters, conservatives ,returning * voters.

I haven't analyzed the numbers but the numbers should match somewhere. Like there must be a lot of conservative, southern, 30-44 year old men, returning * voter, that were included in the final poll and quite a few eastern, 18-30 year old women, returning Gore voters, that were taken out.
These are real people aren't they. So if you increase one category's weight for example conservatives, and these conservatives are men it would also increase the weight percentage for men.
Anyway this is getting way to confusing for me if someone else wants to explain it then go ahead, the only thing is that when I read all the numbers it's hard to remember that there is real people behind them.

[TruthIsAll]

4. Response

Ques. - does the 73,307 total of exit-poll interviews reflect:
1) the "completed responses" only? Or were there more
attempted interviews? I think E/M expects about a 45-55%
not-completed rate.

I believe 73,307 are completed. In fact I seem to recall reading that there were over 100,000 interviews (I could be wrong).

2) the mystery 660 voters added overnight? How could these
be identified? Might it mean an analysis of the last raw data
total interviews by precinct, and see where, if identified in the
"final" poll results, these were "added" (so to speak)?

Beats me. In any case they are irrelevant as far as I am concerned, because they could not have switched the 51-48 Kerry lead among the first 13047 to a Bush win of 51-48. That was impossible.

Draw your own conclusions.

[liam_laddie]

9. Agreed!

TIA - thanks for those answers. Clears up any minor concerns I began
to have after the 4-M pollster's lack of success in total interviews and probable underweighting of Bush. I'll sum all four precincts and
see what the variables are...

[berniew1]

12. The NEP only is indicative of election day voters, not AB voters

Isn't that the case? If early voting and AB voters were different than election day voters, this could have an effect. Can you separate out the ABs and early votes in the precinct in question?

[TruthIsAll]

11. State, Region, National. What more do you want?

.

[kster]

14. kick.nt

[TruthIsAll]

15. You will never hear this mentioned on CNN...

.

[TruthIsAll]

16. State + Regioanl + National = FRAUD n/t

/

[The Doctor.]

18. Thank you TIA.

and nice graphic Dzika.

Now what I'd LOVE to see is E-voting machine densities correlated with MOE 'red shifts'.

Now THAT would be compelling!

If I didn't have mortally pressing concerns right now I'd give it a whirl.