O'Dell says constant mean rBr has not been refuted, but refutes it himself

[eomer]

In the Conclusions section of his paper, Bruce O'Dell says that the USCV working paper does not refute constant mean rBr.

Here is the gist of his argument for that conclusion:

In fact, using either alpha or ln(alpha), the USCV O’Dell Simulator cited in the Working Paper
not only accurately reproduces the aggregate E/M Mean WPE, Median WPE and participation
rates, at the same time it shows no statistically-significant correlation between exit poll bias and
partisanship when precinct-level alpha is analyzed.

In other words, he is basing this conclusion on the fact that the linear regression line for alpha by precinct partisanship and the one for ln(alpha) by precinct partisanship both have a slope of roughly zero.

But this argument ignores the possibility of a non-linear relationship. And in fact, other parts of O'Dell's paper reach the conclusion that a non-linear relationship exists and that response rates do vary significantly by precinct partisanship.

Here are the response rates by quintile (starting with high-Bush) from O'Dell's simulation that successfully reproduced the overall response rate and the mean and median WPEs:

Kerry
Response
Rates
70.4%
59.6%
56.3%
57.8%
53.6%

Bush
Response
Rates
54.0%
53.3%
48.0%
49.7%
51.5%

In another part of the paper O'Dell seems to recognize that there is a significant correlation between precinct partisanship and response rate in those numbers:

However, the simulation’s computed response rates are different for the extremes on the right
and left of the chart. On the left of the chart, in the 90 High Kerry precincts (or about 7% of the
overall sample), computed response rates are roughly equal. The Working Paper does not
analyze the anomaly in High-Kerry precincts.

On the right of the chart, in the 40 High Bush precincts (about 3% of the total sample), the
computed response rate differential is about twice as big as in the center of the chart, or about a
16% difference.

That is a large aggregate difference in response rate. It is an inevitable consequence of
simulating a set of 40 precincts with Mean WPE = -10, Median WPE = -5.8 and an overall
response rate of 56%. A small number of precincts with very large negative WPE are required.

(emphasis added)

So why does O'Dell say on the one hand that there is a large aggregate difference in response rate and on the other hand that constant mean rBr has not been refuted? The following paragraph is apparently his reasoning for saying that:

The authors of the “Working Paper” conclude “the Kerry and Bush exit poll response rates must
be non-uniform in order to match E/M's published numbers”, but recall that the response rate
differential looks rather uniform in the middle 89% of the sample, and paradoxically, some of the
apparent excess WPE in High Bush precincts can be attributed to the inherent “skew effect” of
uniform mean response bias that we discussed earlier – along with a few extreme outliers.

It seems from this that O'Dell is looking to the flattening effect you get by applying Febble's function to WPE in order to say the apparent spike is not real. The problem with this logic is that it is not high WPE we are looking at. If it were then Febble's function might flatten out an apparent spike. But Febble's function does not apply in the case of a response rate. A response rate is not a skewed representation of bias like WPE is. It is a direct and honest representation of bias. If you have a spike in response rate then you do have a spike in bias and you don't have constant mean rBr, by definition. Febble's function has nothing to do with it.

O'Dell's paper refutes constant mean rBr. He found that you have to have response rates that vary significantly by precinct partisanship in order to explain the data if you assume no other source of bias. That finding is inconsistent with a constant mean rBr theory by definition.

Link to O'Dell's paper:
http://www.digitalagility.com/data/ODell_Response_to_USCV_Working_Paper.pdf

[TruthIsAll]

1. The Optimizer illustrates your point...

EXIT POLL RESPONSE OPTIMIZATION MODEL							
6/16/05 9:08 AM							
							
OBJECTIVE:							
Determine values of constrained variables required to derive a
target Kerry/Bush percentage split using aggregate exit poll
response data.							
						
Precinct Variable Input Range (Min, Max) Constraints:         
                                         							
1-Response: equate to weighted average 					
2-Kerry 2-party percentage vote				
3-Alpha (K/B): equate to weighted average					
4-WPE: E-M							
							
Calc	VOTE 	POLL	DEV				
Kerry 	48.77%	52.16%	-3.39%				
Bush	51.23%	47.84%	3.39%				

Actual
Bush	51.23%						
							
Probability	1.59E-11						
 1 in	62,950,429,501						
							
WEIGHTED AVERAGE							
Response 	53.0%						
Alpha (K/B)	1.155						
							
RESPONSE INPUT CONSTRAINTS							
1250	Strong Bush		Strong Kerry				
Prcts	40	415	540	165	90		
							
KERRY WIN%							
Min	0%	25%	41%	59%	75%		
Max	25%	41%	59%	75%	100%		
							
RESPONSE							
Min	50%	50%	50%	50%	50%		
Max	57%	57%	57%	57%	57%		
							
ALPHA (K/B)							
Min	0.90	0.90	0.90	0.90	0.90		
Max	10.00	10.00	10.00	10.00	10.00		
							
WPE							
E-M	-10.0%	-6.1%	-8.5%	-5.9%	0.3%		
Min	-10.0%	-6.1%	-8.5%	-5.9%	0.3%		
Max	-10.0%	-6.1%	-8.5%	-5.9%	0.3%		
							
OPTIMIZER OUTPUT SUMMARY							
							
		EXIT POLL / VOTE DEVIATION					
							
	Poll	Vote	Diff		Poll	Vote	Diff (mm)
Kerry	52.16%	48.77%	-3.39%		63.11	59.01	-4.10
Bush	47.84%	51.23%	3.39%		57.89	61.99	4.10
							
							
CATEG.	HighB	Bush	Even	Kerry	HighK		Total/Avg
Number	40	415	540	165	90		1250
Pct	3.2%	33.2%	43.2%	13.2%	7.2%		100.0%
							
RESP.	56.6%	51.8%	53.3%	55.8%	50.0%		53.0%
Dev	3.6%	-1.2%	0.3%	2.8%	-3.0%		
							
ALPHA							
K/B	1.52	1.15	1.18	1.09	1.00		1.155
ln(a)	0.42	0.14	0.16	0.09	0.00		0.14
Dev	31.5%	-0.4%	1.9%	-5.2%	-13.8%		
							
VOTES							
Kerry	19.3%	40.6%	48.1%	62.4%	78.5%		48.77%
Bush	80.7%	59.4%	51.9%	37.6%	21.5%		51.23%
							
POLL 							
Kerry	24.3%	43.7%	52.4%	65.3%	78.4%		52.16%
Bush	75.7%	56.3%	47.6%	34.7%	21.6%		47.84%
							
							
WPE	-10.0%	-6.1%	-8.5%	-5.9%	0.3%		-6.77%
E-M	-10.0%	-6.1%	-8.5%	-5.9%	0.3%		-6.77%
Diff	0.0%	0.0%	0.0%	0.0%	0.0%		0.0%
							
							
							
OPTIMIZATION MODEL							
							
Categ.	HighB	Bush	Even	Kerry	HighK		Total/Avg
Prcts	40	415	540	165	90		1250
Kerry	19.3%	40.6%	48.1%	62.4%	78.5%		48.77%
							
ALPHA	1.52	1.15	1.18	1.09	1.00		1.16
AvgDev	31%	0%	2%	-5%	-14%		
							
RESPONDERS							
Total	23	215	288	92	45		663
Pct	56.6%	51.8%	53.3%	55.8%	50.0%		53.00%
							
Kerry	5	94	151	60	35		345
Pct	24.3%	43.7%	52.4%	65.3%	78.4%		52.16%
Bush	17	121	137	32	10		317
Pct	75.7%	56.3%	47.6%	34.7%	21.6%		47.84%
							
REFUSERS							
Total	17	200	252	73	45		587
Pct	43.4%	48.2%	46.7%	44.2%	50.0%		47.00%
							
Kerry	2	75	109	43	35		264
Pct	12.8%	37.3%	43.3%	58.6%	78.7%		44.90%
Bush	15	125	143	30	10		323
Pct	87.2%	62.7%	56.7%	41.4%	21.3%		55.10%
							
							
TOTAL VOTES							
Kerry	8	169	260	103	71		610
Pct	19.3%	40.6%	48.1%	62.4%	78.5%		48.77%
Bush	32	246	280	62	19		640
Pct	80.7%	59.4%	51.9%	37.6%	21.5%		51.23%
							
							
2-PARTY VOTE (mm)							
TOTAL	3.87	40.17	52.27	15.97	8.71		121.00
							
Kerry 	0.75	16.32	25.15	9.96	6.84		59.01
Bush	3.13	23.86	27.12	6.01	1.87		61.99
Diff	-2.38	-7.54	-1.97	3.95	4.97		-2.98
							
EXIT POLL VOTE (mm)							
Kerry 	0.94	17.54	27.37	10.43	6.83		63.11
Bush	2.93	22.63	24.90	5.54	1.88		57.89
Diff	-1.99	-5.09	2.47	4.89	4.94		5.22
							
VOTE DEVIATION							
Kerry 	-0.19	-1.23	-2.22	-0.47	0.01		-4.10
							
WPE							
Kv-Bv	-61.4%	-18.8%	-3.8%	24.7%	57.1%		-2.46%
Kp-Bp	-51.4%	-12.7%	4.7%	30.6%	56.8%		4.31%
Diff	-10.0%	-6.1%	-8.5%	-5.9%	0.3%		-6.77%
							
E-M	-10.0%	-6.1%	-8.5%	-5.9%	0.3%		-6.77%
							
							
K/B	Kerry	WPE	Prob.	1 in			
1.00	48.77%	0.01%	5.03E-01	2			
1.02	49.19%	-0.85%	2.04E-01	5			
1.04	49.62%	-1.70%	4.81E-02	21			
1.06	50.04%	-2.55%	6.23E-03	160			
1.08	50.47%	-3.40%	4.28E-04	2,334			
1.10	50.90%	-4.25%	1.53E-05	65,300			
1.12	51.32%	-5.11%	2.81E-07	3,555,747			
1.14	51.78%	-6.02%	1.79E-09	559,644,344			
1.15	51.99%	-6.45%	1.33E-10	7,521,468,533			
1.16	52.21%	-6.87%	8.36E-12	119,593,696,538			
1.17	52.42%	-7.29%	4.44E-13	2,250,674,476,447			
							

[TruthIsAll]

2. ALPHA - NONLINEAR AND LINEAR (NON-ZERO SLOPE) CURVE FIT

[papau]

3. You must write a book- The stat proof discussion should be airport reading

I wish that we could expose this - and fix it - before 2006.

I suspect we may never get the basic data - but your chart needs to show "precinct" data. Very frustrating that selling the precinct data has priority over making sense of the data.

[autorank]

6. papau, I second the idea of a TIA book! Election fraud is all about ...

...finding the sell out. Democracy bought and sold to the highest bidder. I feel fortunate to watch the truth win out in these debates. The analytic talent here, TIA, eomer, etc., is incredible. And it's all about one thing, making our vote count.

[petgoat]

128. I third the idea of a TIA book!

TIA's argument about the 2000 voters is very commonsensible and easy to
explain to lay people (like me).

I'd like to hear the statisticians' take on the DNC study. How can
modeling on the Hogan campaign make sense when it appears from the
outcome (Hogan 38%, Taft 58%) that the Dems just lacked enthusiasm?

[TruthIsAll]

4. ABSOLUTE MAXIMUM BUSH: 52.57%

EXIT POLL RESPONSE OPTIMIZATION MODEL							
6/16/05 10:55 AM							
							
OBJECTIVE:							
Determine maximum Bush percentage using aggregate exit poll
response data assumptions.							
						
Precinct Variable Input Range (Min, Max) Constraints:         
                                         							
1-Response rates: equate to weighted average                  
                                 							
2-Kerry 2-party percentage vote                               
                 							
3-Alpha (K/B): equate to weighted average                     
                                 							
4-WPE: input  (optional: Min= Max= E-M)							
							
	VOTE 	POLL	DEV				
Kerry 	47.43%	50.82%	-3.39%				
Bush	52.57%	49.18%	3.39%				
Actual Bush	51.23%						
							
Probability	1.59E-11						
 1 in	62,950,429,501						
							
WEIGHTED AVERAGE							
Response 	53.0%						
Alpha (K/B)	1.155						
							
RESPONSE INPUT CONSTRAINTS							
1250	Strong Bush		Strong Kerry				
Prcts	40	415	540	165	90		
							
KERRY WIN%							
Min	0%	25%	41%	59%	75%		
Max	25%	41%	59%	75%	100%		
							
RESPONSE							
Min	50%	50%	50%	50%	50%		
Max	57%	57%	57%	57%	57%		
							
ALPHA (K/B)							
Min	0.90	0.90	0.90	0.90	0.90		
Max	10.00	10.00	10.00	10.00	10.00		
							
WPE							
E-M	-10.0%	-6.1%	-8.5%	-5.9%	0.3%		
Min	-10.0%	-6.1%	-8.5%	-5.9%	0.3%		
Max	-10.0%	-6.1%	-8.5%	-5.9%	0.3%		
							
OPTIMIZER OUTPUT SUMMARY							
							
		EXIT POLL / VOTE DEVIATION					
							
	Poll	Vote	Diff		Poll	Vote	Diff (mm)
Kerry	50.82%	47.43%	-3.39%		61.49	57.40	-4.10
Bush	49.18%	52.57%	3.39%		59.51	63.60	4.10
							
							
CATEG.	HighB	Bush	Even	Kerry	HighK		Total/Avg
Number	40	415	540	165	90		1250
Pct	3.2%	33.2%	43.2%	13.2%	7.2%		100.0%
							
RESP.	56.6%	51.8%	53.3%	55.8%	50.0%		53.0%
Dev	3.6%	-1.2%	0.3%	2.8%	-3.0%		0.0%
							
ALPHA							
K/B	1.40	1.15	1.18	1.10	1.00		1.155
Dev	21.2%	-0.1%	2.3%	-4.8%	-13.8%		0.0%
ln(A)	0.34	0.14	0.17	0.10	0.00		0.14

VOTES							
Kerry	25.0%	39.7%	46.9%	59.0%	75.0%		47.43%
Bush	75.0%	60.3%	53.1%	41.0%	25.0%		52.57%
							
POLL 							
Kerry	30.0%	42.8%	51.1%	62.0%	74.9%		50.82%
Bush	70.0%	57.2%	48.9%	38.1%	25.2%		49.18%
							
							
WPE	-10.0%	-6.1%	-8.5%	-5.9%	0.3%		-6.77%
E-M	-10.0%	-6.1%	-8.5%	-5.9%	0.3%		-6.77%
Diff	0.0%	0.0%	0.0%	0.0%	0.0%		0.0%
							
							
							
OPTIMIZATION MODEL							
							
Categ.	HighB	Bush	Even	Kerry	HighK		Total/Avg
Prcts	40	415	540	165	90		1250
Kerry	25.0%	39.7%	46.9%	59.0%	75.0%		47.43%
							
ALPHA	1.40	1.15	1.18	1.10	1.00		1.15
ln (alpha)	0.34	0.14	0.17	0.10	0.00		0.14
AvgDev	21%	0%	2%	-5%	-14%		0%
							
RESPONDERS							
Total	23	215	288	92	45		663
Pct	56.6%	51.8%	53.3%	55.8%	50.0%		53.00%
							
Kerry	7	92	147	57	34		337
Pct	30.0%	42.8%	51.1%	62.0%	74.9%		50.82%
Bush	16	123	141	35	11		326
Pct	70.0%	57.2%	48.9%	38.1%	25.2%		49.18%
							
REFUSERS							
Total	17	200	252	73	45		587
Pct	43.4%	48.2%	46.7%	44.2%	50.0%		47.00%
							
Kerry	3	73	106	40	34		256
Pct	18.5%	36.5%	42.0%	55.3%	75.2%		43.56%
Bush	14	127	146	33	11		331
Pct	81.5%	63.5%	58.0%	44.7%	24.8%		56.44%
							
							
TOTAL VOTES							
Kerry	10	165	253	97	68		593
Pct	25.0%	39.7%	46.9%	59.0%	75.0%		47.43%
Bush	30	250	287	68	23		657
Pct	75.0%	60.3%	53.1%	41.0%	25.0%		52.57%
							
							
2-PARTY VOTE (mm)							
TOTAL	3.87	40.17	52.27	15.97	8.71		121.00
							
Kerry 	0.97	15.96	24.51	9.42	6.53		57.40
Bush	2.90	24.21	27.76	6.55	2.18		63.60
Diff	-1.94	-8.26	-3.25	2.87	4.36		-6.21
							
EXIT POLL VOTE (mm)							
Kerry 	1.16	17.18	26.73	9.89	6.52		61.49
Bush	2.71	22.99	25.54	6.08	2.19		59.51
Diff	-1.55	-5.81	1.19	3.82	4.33		1.99
							
VOTE DEVIATION							
Kerry 	-0.19	-1.23	-2.22	-0.47	0.01		-4.10
							
WPE							
Kv-Bv	-50.0%	-20.6%	-6.2%	18.0%	50.0%		-5.13%
Kp-Bp	-40.0%	-14.5%	2.3%	23.9%	49.7%		1.64%
Diff	-10.0%	-6.1%	-8.5%	-5.9%	0.3%		-6.77%
							
E-M	-10.0%	-6.1%	-8.5%	-5.9%	0.3%		-6.77%
							
							
K/B	Kerry	WPE	Prob.	1 in			
1.00	48.77%	0.01%	5.03E-01	2			
1.02	49.19%	-0.85%	2.04E-01	5			
1.04	49.62%	-1.70%	4.81E-02	21			
1.06	50.04%	-2.55%	6.23E-03	160			
1.08	50.47%	-3.40%	4.28E-04	2,334			
1.10	50.90%	-4.25%	1.53E-05	65,300			
1.12	51.32%	-5.11%	2.81E-07	3,555,747			
1.14	51.78%	-6.02%	1.79E-09	559,644,344			
1.15	51.99%	-6.45%	1.33E-10	7,521,468,533			
1.16	52.21%	-6.87%	8.36E-12	119,593,696,538			
1.17	52.42%	-7.29%	4.44E-13	2,250,674,476,447			

[TruthIsAll]

5. ABSOLUTE MAXIMUM KERRY: 56.50%

EXIT POLL RESPONSE OPTIMIZATION MODEL							
6/16/05 11:05 AM							
							
OBJECTIVE:							
Determine maximum Kerry vote using aggregate exit poll
response data subject to the following constraints.
 						
Precinct Variable Input Range (Min, Max) Constraints:         
                                         							
1-Response rates: equate to weighted average                  
                                 							
2-Kerry 2-party percentage vote                               
                 							
3-Alpha (K/B): equate to weighted average                     
                                 							
4-WPE: input  (optional: Min= Max= E-M)							
							
	VOTE 	POLL	DEV				
Kerry 	56.50%	59.88%	-3.39%				
Bush	43.50%	40.12%	3.39%				
Actual Bush	51.23%						
							
Probability	1.59E-11						
 1 in	62,949,989,550						
							
WEIGHTED AVERAGE							
Response 	53.0%						
Alpha (K/B)	1.155						
							
RESPONSE INPUT CONSTRAINTS							
1250	Strong Bush		Strong Kerry				
Prcts	40	415	540	165	90		
							
KERRY WIN%							
Min	0%	25%	41%	59%	75%		
Max	25%	41%	59%	75%	100%		
							
RESPONSE							
Min	50%	50%	50%	50%	50%		
Max	57%	57%	57%	57%	57%		
							
ALPHA (K/B)							
Min	0.90	0.90	0.90	0.90	0.90		
Max	10.00	10.00	10.00	10.00	10.00		
							
WPE							
E-M	-10.0%	-6.1%	-8.5%	-5.9%	0.3%		
Min	-10.0%	-6.1%	-8.5%	-5.9%	0.3%		
Max	-10.0%	-6.1%	-8.5%	-5.9%	0.3%		
							
OPTIMIZER OUTPUT SUMMARY							
							
		EXIT POLL / VOTE DEVIATION					
							
	Poll	Vote	Diff		Poll	Vote	Diff (mm)
Kerry	59.88%	56.50%	-3.39%		72.46	68.36	-4.10
Bush	40.12%	43.50%	3.39%		48.54	52.64	4.10
							
							
CATEG.	HighB	Bush	Even	Kerry	HighK		Total/Avg
Number	40	415	540	165	90		1250
Pct	3.2%	33.2%	43.2%	13.2%	7.2%		100.0%
							
RESP.	56.6%	51.8%	53.3%	55.8%	50.0%		53.0%
Dev	3.6%	-1.2%	0.3%	2.8%	-3.0%		0.0%
							
ALPHA							
K/B	2.04	1.15	1.14	1.08	1.00		1.155
Dev	76.4%	-0.5%	-0.9%	-6.6%	-13.7%		0.0%
ln (A)	0.71	0.14	0.13	0.08	0.00		0.14

VOTES							
Kerry	9.6%	41.0%	59.0%	75.0%	99.9%		56.50%
Bush	90.4%	59.0%	41.0%	25.0%	0.1%		43.50%
							
POLL 							
Kerry	14.6%	44.0%	63.3%	78.0%	99.7%		59.88%
Bush	85.4%	56.0%	36.8%	22.1%	0.3%		40.12%
							
							
WPE	-10.0%	-6.1%	-8.5%	-5.9%	0.3%		-6.77%
E-M	-10.0%	-6.1%	-8.5%	-5.9%	0.3%		-6.77%
Diff	0.0%	0.0%	0.0%	0.0%	0.0%		0.0%
							
							
							
OPTIMIZATION MODEL							
							
Categ.	HighB	Bush	Even	Kerry	HighK		Total/Avg
Prcts	40	415	540	165	90		1250
Kerry	9.6%	41.0%	59.0%	75.0%	99.9%		56.50%
							
ALPHA	2.04	1.15	1.14	1.08	1.00		1.16
ln(a)	0.71	0.14	0.13	0.08	0.00		0.14
AvgDev	76%	-1%	-1%	-7%	-14%		0%
							
RESPONDERS							
Total	23	215	288	92	45		663
Pct	56.6%	51.8%	53.3%	55.8%	50.0%		53.00%
							
Kerry	3	95	182	72	45		397
Pct	14.6%	44.0%	63.3%	78.0%	99.7%		59.88%
Bush	19	120	106	20	0		266
Pct	85.4%	56.0%	36.8%	22.1%	0.3%		40.12%
							
REFUSERS							
Total	17	200	252	73	45		587
Pct	43.4%	48.2%	46.7%	44.2%	50.0%		47.00%
							
Kerry	1	76	136	52	45		310
Pct	3.1%	37.7%	54.1%	71.3%	100.0%		52.62%
Bush	17	125	116	21	0		278
Pct	96.9%	62.3%	45.9%	28.7%	0.0%		47.38%
							
							
TOTAL VOTES							
Kerry	4	170	319	124	90		706
Pct	9.6%	41.0%	59.0%	75.0%	99.9%		56.50%
Bush	36	245	221	41	0		544
Pct	90.4%	59.0%	41.0%	25.0%	0.1%		43.50%
							
							
2-PARTY VOTE (mm)							
TOTAL	3.87	40.17	52.27	15.97	8.71		121.00
							
Kerry 	0.37	16.47	30.84	11.98	8.70		68.36
Bush	3.50	23.70	21.43	3.99	0.01		52.64
Diff	-3.13	-7.23	9.41	7.99	8.69		15.72
							
EXIT POLL VOTE (mm)							
Kerry 	0.57	17.70	33.06	12.45	8.69		72.46
Bush	3.31	22.48	19.21	3.52	0.03		48.54
Diff	-2.74	-4.78	13.85	8.93	8.66		23.92
							
VOTE DEVIATION							
Kerry 	-0.19	-1.23	-2.22	-0.47	0.01		-4.10
							
WPE							
Kv-Bv	-80.7%	-18.0%	18.0%	50.0%	99.7%		13.00%
Kp-Bp	-70.7%	-11.9%	26.5%	55.9%	99.4%		19.77%
Diff	-10.0%	-6.1%	-8.5%	-5.9%	0.3%		-6.77%
							
E-M	-10.0%	-6.1%	-8.5%	-5.9%	0.3%		-6.77%
							
							
K/B	Kerry	WPE	Prob.	1 in			
1.00	48.77%	0.01%	5.03E-01	2			
1.02	49.19%	-0.85%	2.04E-01	5			
1.04	49.62%	-1.70%	4.81E-02	21			
1.06	50.04%	-2.55%	6.23E-03	160			
1.08	50.47%	-3.40%	4.28E-04	2,334			
1.10	50.90%	-4.25%	1.53E-05	65,300			
1.12	51.32%	-5.11%	2.81E-07	3,555,747			
1.14	51.78%	-6.02%	1.79E-09	559,644,344			
1.15	51.99%	-6.45%	1.33E-10	7,521,468,533			
1.16	52.21%	-6.87%	8.36E-12	119,593,696,538			
1.17	52.42%	-7.29%	4.44E-13	2,250,674,476,447			

[Bill Bored]

7. OK Question:

In 25 words or less, is response rate inversely correlated with red shift?

Forget partisanship. Forget Bias Index. Are there data that show that in precincts with low response rates, there is more of a Red shift?

[OnTheOtherHand]

8. I was sort of hoping someone else would do this...

There is a fundamental difference between descriptive and inferential statistics.

Descriptively, the mean and median alpha are higher in the high-Bush precincts than in the other groups of precincts. (I am more comfortable talking about the alpha than the K and B "response rates," because we can't actually observe the latter.) And O'Dell notes that this is true.

Does this refute his conclusion? Is the man delusional, or what?

That would be "what." His conclusion is inferential: the simulation "shows no statistically-significant correlation between exit poll bias and partisanship." (By the way, he doesn't use the phrase "constant mean" except to quote it from the USCV paper.)

But his tests are linear tests across the entire data set. So, is there a non-linear relationship that shows up as statistically significantly greater bias in the high-Bush precincts?

Doesn't look like it to me. Looking at the B87 tab in his spreadsheet (and I think I had to add the ln alphas myself), the mean ln alpha for the 40 high-Bush precincts is 0.2672; the standard deviation is 0.6473; the nominal standard error of the mean is 0.1023; a 95% confidence interval with 39 degrees of freedom will be roughly the mean plus-or-minus 2.023 times the standard error, or between about 0.0602 and 0.4742.

Since the overall mean ln alpha for this simulation is 0.1478, it seems pretty clear that O'Dell is right by this test as well: the higher observed biases (or imputed response rates) in the high-Bush precincts aren't statistically significantly higher.

Ironically, the test comes out pretty much the same way if you just use WPE. The mean WPE for the high-Bush precincts is -.1007, compared with -0.0665 overall. The nominal standard error of mean WPE for these precincts is 0.0294. Again, a little bit more than one standard error above the overall mean, well short of statistical significance.

So, is there "constant mean rBr"? Heck if I know. E/M never even said that rBr would have a "constant mean." But O'Dell's conclusions seem fine to me (except he should have said "precinct-level ln alpha," since we know that alpha is skewed).

[LightningFlash]

9. rBr is dead, and eBr is next....

And that doesn't change the fact he refuted himself. There was no uniform response level, and its not possible.

[OnTheOtherHand]

10. well, that's quite a closely reasoned rebuttal. n/t

[tommcintyre]

11. Well... "closely reasoned rebuttals" are good

Example:
http://64.233.179.104/search?q=cache:alxkytkXXoEJ:www.sun-sentinel.com/features/booksmags/sfl-bkmabecosbymay01,0,7701556.column%3Fcoll%3Dsfla-features-books+%22closely+reasoned+rebuttal%22&hl=en
"Easy to shrug off, that is, if you don't actually read the book. Because if you do, you'll find that Michael Eric Dyson has paid Cosby the ultimate compliment one social critic can pay another. He has taken Cosby seriously and mounted a closely reasoned rebuttal."

Another example:
http://www.geocities.com/wakefielduu/sermonfiles/1999-2000/9.19.99.txt
"It is hard to overstate the impact of this book as a forceful and closely reasoned rebuttal to the Calvinism of its day."

So that means you like, and agree with lighteningFlash's concise analysis, I take it? He does have a way with words, doesn't he?

[eomer]

13. LOL

Too late to edit the OP but I should have just said:
mean is not constant --> it is not constant mean, QED.

[OnTheOtherHand]

15. eomer, I'm frustrated

As I already pointed out, "constant mean" is a phrase that O'Dell only uses in quoting the USCV working paper. Edison/Mitofsky never use it. Perhaps O'Dell should simply have said that USCV seemed to be refuting a straw man. Instead, he tried to give a reasonable interpretation of USCV's claim, and to show why he disagreed.

(edit to cranky subject header)

[eomer]

17. I don't believe the theory USCV refuted is a strawman.

I believe it was their best guess of what Mitofsky meant when he threw out a vague conjecture as an explanation for a discrepancy that is of utmost importance to the nation.

If we are having difficulties because we each have a different theory of what the theory is then that is Mitofsky's fault, not USCV's. Mitofsky could remedy this problem by providing the specifics of exactly what his theory is and, while he's at, some kind of a scientific argument rather than just a presumption.

So I guess the scorecard (and timeline) is like this:

• Mitofsky proffered a vague theory, let's call it theory (x) to give it the proper air of mystery.
  
  
• USCV refuted theory (a), a more specific theory that was their best guess at the time of what Mitofsky meant by theory (x).
  
  
• Sometime after USCV's refutation of theory (a) Mitofsky stated that he does not claim theory (a), giving credence to USCV's refutation of it. Mitofsky didn't narrow down what theory (x) really means except to say that it is not theory (a).
  
  
• Next O'Dell refuted the refutation of theory (b), which is his guess of what Mitofsky means by theory (x) now that we know he didn't mean theory (a). But he incorrectly says he refuted USCV's refutation. He didn't, because USCV refuted theory (a), which Mitofsky admits is refutable, while O'Dell proved theory (b).
  
  
• We still do not know whether theory (b) is what Mitofsky meant by theory (x) so O'Dell could be barking up the wrong tree just as much as USCV was.
  

Is that complicated enough? If not we could get Rummie to come give us a talk on "what we know we don't know".


edit to remove response to cranky subject header :-)

[OnTheOtherHand]

20. OK, I think we have reached honorable stalemate

or temporary equilibrium, or -- dunno.

My best characterization of Mitofsky's theory (x) is that for some reason or reasons Kerry voters participated in the exit polls at a higher rate than Bush voters, and that this difference rather than fraud accounts for the discrepancy between the exit polls and the official returns.

I agree that the theory is vague, and I agree that nothing on the record directly supports it (although I think there is circumstantial evidence that it is at least partly true).

I'm not quite sure how USCV got from (x) to (a), which I would characterize as the theory that this difference in participation rates would in no way depend on precinct partisanship. But I think the basic line of reasoning was that differential participation could not explain the observed high mean WPEs in the high-Bush precincts, or at least that "Bush stronghold vote count corruption" was a more plausible explanation. That was (at least part of) the argument as of March. I hesitate to characterize the unique contribution of the May paper.

I think that O'Dell did a pretty good job on arguing that differential participation could explain the observed high mean WPEs in the high-Bush precincts. I think that is true whether we look for a linear association (which I think is reasonable because massive vote shift seems likely to induce one) or a non-linear association.

If O'Dell had said that his analysis supported Mitofsky's response bias theory, then he would be going way beyond the evidence, not to mention the small problem that we are unsure what the theory is! So I think you and I agree about that.

But as I read him, O'Dell said that his work supported Mitofsky's findings presented at AAPOR -- in particular, that an apparent correlation between precinct partisanship and exit poll red-shift could be an artifact of the red-shift measure (WPE as opposed to ln alpha). It seems to me that O'Dell is right about that.

I certainly don't accede in the interpretation that O'Dell argued for "constant mean bias" and yet admitted that it wasn't true.

[TruthIsAll]

23. You are defending on the 1 yard-line. If you think there is evidence...

to support rBr (or eDp for that matter), then please point it out.
We would love to see your circumstantial evidence.

"I agree that the theory is vague, and I agree that nothing on the record directly supports it (although I think there is circumstantial evidence that it is at least partly true)."

Sorry, no sale.

[OnTheOtherHand]

24. actually, that doesn't seem germane to the thread

I am not trying to talk you into rBr. I am not trying to sell rBr. We both have better uses of our time.

You apparently regard rBr as inherently unbelievable; I don't. Not a promising basis for dialogue. (I should probably say, again, that on the other hand, we both do consider fraud inherently believable. Yeah, we have huge disagreements over there, too, but it's not like we're matter and anti-matter.)

But this thread is about whether O'Dell refuted his own argument, not about rBr.

[TruthIsAll]

26. What about rKr? Would you consider that believable?

This is very germane.

Were there Republicans who voted for Kerry and reluctant to say so?
Could it have been a wash: rKr= RBr?
If not, what is your justification for believing rBr and not rKr?

WHERE IS THE CIRCUMSTANTIAL EVIDENCE YOU ALLUDED TO?
ENOUGH OF THE INTERMINABLE SPIN.

WE WANT TO SEE YOUR EVIDENCE.
OR ARE YOU GOING TO JUST QUIT ON US?

YOU TOLD EOMER IT IS ENDING AS A STALEMATE WITHOUT OFFERING ANY SUBSTANCE TO YOUR BELIEF THAT rBr WAS POSSIBLE.

IT'S ABOUT TIME YOU DID SO, DON'T YOU AGREE?

WHERE IS THE BEEF?

[OnTheOtherHand]

28. PLEASE REMOVE YOUR CAPS LOCK

You can perhaps outbluster me, but you cannot convince me, nor can I convince you. So we should find something else to talk about, or leave each other alone, until and unless we find a useful way to engage the issue. That isn't "spin," it's common sense (which I often lack).

But yes, I can imagine rKr. I can imagine any number of ways in which a poll could go wrong.

[TruthIsAll]

31. ok, i will whisper...where is the circumstantial evidence ....

you keep alluding to?

much time and effort has been spent on this forum by you and others in relentless esoteric verbiage which has had the effect of promoting what many of us strongly believe to be an unsubstantiated myth - that the exit polls could very well have been wrong. So far, rBr is the only possible hypothesis which has been put forth to account for the discrepancies -other than fraud.

now, since we are at an apparent stalemate, and since we have shown reams of evidence to substantiate our case that the exit polls were in fact correct (up until the final national 13660 which was matched to the vote) would it be too much to expect you and others of similar persuasion to do likewise?

will you finally make your case and present us with some of that circumstantial evidence?

[Internut]

32. Evidence -

by its nature, there can be no evidence that would point to rBr, for the simple reason that it was not anticipated, and no effort was made to figure out which people refused to participate in the polling. When you ask someone to answer the poll, and he/she refuses and walks away, how exactly do you propose to ascertain which way they voted?

This does not mean that rBr did not happen. Neither does it mean that it happened.

What we do have, though, is researched evidence that in other countries, in their elections, refusal rates did correlate with the voting preferences, enough to skew their exit polls, as well as pre-election polls, and enough to make those polls' accuracy improve when that correlation was taken into account. Again, the fact UK, France and Germany have their "rBr" effect does not "prove" that there was one in the US. But it shows that it is definitely possible.

If you'd like to see that research, I can provide links.

[TruthIsAll]

33. no evidence. forget the l links. thanks for your input. end of story.

/

[OnTheOtherHand]

61. well, heck, _I_ would like to see those links

One of those things I've never gotten around to compiling. I think it would be a public service.

[Internut]

65. Here you go

http://www.alba.org.uk/polls/accuracy.html
http://www.portal.telegraph.co.uk/opinion/main.jhtml?xml=/opinion/2004/11/04/do0405.xml&sSheet=/opinion/2004/11/04/ixop.html
http://www.icmresearch.co.uk/reviews/2002/failure-of-the-polls-1997.htm

The third one is the comprehensive one. It also goes into the research into the "faulty memory" effect - that is, people misreporting to pollsters their past voting behaviors.

[tommcintyre]

34. You don't find rBr "unbelievable" FOR TWO SUCCESSIVE BUSH'S???

You said:
"You apparently regard rBr as inherently unbelievable; I don't." <bolding mine ;) >

OTOH, doesn't it bother you that Mitofsky originally fabricated the idea in 1992 to explain away a questionable exit poll situation for his father?

http://www.democraticunderground.com/discuss/duboard.php?az=show_topic&forum=203&topic_id=371523
"OMG! Mitofsky invented reluctant Bush responder scam for Bush Sr. in '92!

I'm very interested in you thoughts on this.

[OnTheOtherHand]

36. your research is really, really selective

Non-response bias is not something that Mitofsky invented in 1992. It is a well-known concern in survey research.

When survey results vary from (presumed) "real-world facts," survey researchers tend to tick down a list of Things That Go Wrong With Surveys -- selection bias, non-response bias, question wording bias, social desirability effects, etc. In this case, Mitofsky may have been right, he may have been wrong, but I doubt that many -- if any -- public opinion analysts concluded that he was probably covering up a stolen election.

I'm surprised that you link to your previous post, which links to a article posted on votefraud.org, which argues this: "Plissner reveals that the exit pollers are totally dependent on which voters are willing to talk to them, and that Buchanan supporters were more likely to want to talk. In other words, the process is NOT scientific."

You may just be the first person in the world to conclude instead that Mitofsky was covering up for Bush. But I wouldn't exactly say that you have proven it.

[tommcintyre]

37. Maybe your perception is really selective? ;)

from the link:

http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=203&topic_id=377607&mesg_id=378723

The pertinent information:
"- Buchanan vs. Bush in the 1992 NH primaries: Scenario: Buchanan very close in the early exit polls (6 points), amazingly (and mysteriously), Bush pulls it out big (16 points), and Mitofsky explains (spins?) the early exit poll discrepancies:"

"July 18, 2000 NA (Network America) e-wire
What? Buchanan Bias? What’s that?
http://www.votefraud.org/News/2000/7/071800.html

"Mitofsky’s best guess <hypothesis?> is that Buchanan’s voters were prouder of what they had done and, hence, more prone to respond, than Bush’s were. <THE BIRTH OF rBr!>” <Hmm... so voters BOTH to the left (Kerry) AND right (Buchanan) are prouder of voting for their candidate than for the Bushs'?>"

So what part of this don't you understand?

This story has been pubbed here twice, tia has referred/used it in several threads; and, it's on Cannon's blog.

So far as I know (and I've really tried to monitor this), ONLY TWO people have disputed (avoided?) the reality of this - you and Febble. Why is that, you think? ;)

Also, a question: Have you listened to Victoria Collier's interview yet? I'm really interested to see what you will have to say about that.

Her dad (and uncle) started investigating election fraud in 1970 (and wrote the book Votescam); and, as you will hear in the interview, Mitofsky's hands have been "dirty" pretty much since he invented exit polling in the mid 1960's.

So, again, I would love to hear your take on this interview.

Will you listen to it?

http://www.edwardsdavid.com/BushVideos/infidelguy.com_Vote_Scam_a.mp3

If you want to read the book, the first eight chapters of "votescam are here:
http://www.constitution.org/vote


I'll post some more links (from different sources) for you here tomorrow that show, I'm NOT the first person to conclude that Mitofsky's hands are dirty.

[OnTheOtherHand]

40. I don't understand any part of the story, actually

You linked to this:
http://www.votefraud.org/News/2000/7/071800.html

which states:

"WE have video footage showing the exit poller calling in her results on a pay phone from 5 PM to 7 PM, while HUNDREDS of voters walk in and out of a high school where five precincts were voting. AND, they admit only 1 out of 15 voters will even talk to them in the off hours."

and

"Plissner reveals that the exit pollers are totally dependent on which voters are willing to talk to them, and that Buchanan supporters were more likely to want to talk."

And you believe this somehow supports the argument that the 2004 exit polls prove fraud? Oh-kay. (The "1 out of 15" sounds pretty extreme, but hey, it's your source, not mine.)

If Victoria Collier came up with this "Mitofsky-in-the-can" argument before you did, that is an interesting historical footnote, but it has no evidential force. If she actually has any evidence to support it, I would encourage you to make it available.

[tommcintyre]

48. OK, do you understand the following sentence from the link?

Same link:
http://www.votefraud.org/News/2000/7/071800.html

"Mitofsky’s best guess is that Buchanan’s voters were prouder of what they had done and, hence, more prone to respond, than Bush’s were."

(Sixth paragraph, last line - to make it easier for you to find.)

Also, I was still wondering if you would be willing to listen to Victoria Collier's interview. Eventually, I will write up a piece on this and the other "evidence" (quotes for you ;) ) I have found.

Before I do, It would be interesting to have your opposing viewpoint ("pro-Mitofsky's veracity").

So, are you willing to listen to the Victoria Collier interview; and render a critique?

[LightningFlash]

52. No, he doesn't. He doesn't understand it's nothing more than sloppy guess.

He doesn't understand any of that and can't wrap his head around the fact that yes, it was the same type of sloppy guess used twice before and has no relavance. Let's now ignore this strawman which means nothing. So don't keep trying to educate him. He will stick by the fact that yes, its still a sloppy guess which should be correct(When it has proven not to be.)

[OnTheOtherHand]

63. sorry, I did miss this response

First of all, I can't figure out whether you are squarely engaging the point I am trying to make in #40. In my reading, the votefraud.org article basically is arguing that exit polls aren't reliable. If exit polls aren't reliable, then they can't prove that the 2004 election was stolen. I'm not trying to be clever here, I think it's a really basic point.

I do see that Mitofsky made a sort of "rBr argument" about Bush versus Buchanan in 1992. What I don't see is why you assume (if you do assume) that he was lying, or even wrong. Where is your argument?

Pro-Mitofsky's veracity -- first of all, I generally assume that people are honest, absent compelling evidence to the contrary. Second, Mitofsky is widely respected, past president of AAPOR and winner of a Lifetime Achievement Award -- which certainly doesn't prove that he is honest, but seems at least to imply that an awful lot of people think he is. Third, non-response bias is a common theme in survey research, so Mitofsky's suggestion was mainstream, not bizarre.

I have to admit that I did listen to the Collier interview. I almost gave up when she talked about the CIA having killed JFK; is that a crucial part of the theory, or can we set that aside?

If I understood rightly, she argued that the exit polls had never been real, had always been rigged to match the predetermined election returns, but that maybe they were becoming more accurate now. If I got that wrong, please let me know, but I will not sit through that hour again.

[tommcintyre]

72. Glad you listened to it, but I'd give you a "D" for comprehension ;)

"If exit polls aren't reliable, then they can't prove that the 2004 election was stolen."
<Actually, historically, exit polls are VERY reliable. But, I think you know that. If not, I can provide numerous links, etc.>

"I do see that Mitofsky made a sort of "rBr argument" about Bush versus Buchanan in 1992. What I don't see is why you assume (if you do assume) that he was lying, or even wrong."
<Actually, it is NOT "a sort of rBr argument", it is the "mirror image" (looking at it from the opposite side). As a matter of fact, the eKr (exuberant Kerry responder) is the equivalent argument in 2004, is it not?>

Pro-Mitofsky's veracity -- first of all, I generally assume that people are honest, absent compelling evidence to the contrary. <my bolding>
<This approach, although it sound good, is prone to lead to bias either way. A much more reasonable/objective approach is to assume a position of neutrality; and THEN evaluate.>

Second, Mitofsky is widely respected, past president of AAPOR and winner of a Lifetime Achievement Award -- which certainly doesn't prove that he is honest, but seems at least to imply that an awful lot of people think he is.
<As you have pointed out, not too many of them are speaking out about the POSSIBILITY of election fraud either - assent of the majority has never impressed me. Generally, the "truth" has proven to be an independent variable - and must be judged on it's own merits.>

Third, non-response bias is a common theme in survey research, so Mitofsky's suggestion was mainstream, not bizarre.
<A problem of "coincidence" (compounded by all the other "coincidences" of this election). Please point out to me where Mitofsky has used either eWCr (or rWCr - WC= "winning candidate"); and I will back-off somewhat (other mitigating factors would, of course, still need to be considered).>

I have to admit that I did listen to the Collier interview. I almost gave up when she talked about the CIA having killed JFK; is that a crucial part of the theory, or can we set that aside?
<Of course we can set it aside. Again, (IMO) we should separate our bias (what we think of every assertion made by an individual); and focus on what is pertinent. For example, I don't totally "write you off" just because you have made statements that have really "raised my eyebrows" regarding election fraud. (On the contrary, I'm still watching for a definitive statement re: "environmental politics" from you.;) )>

If I understood rightly, she argued that the exit polls had never been real, had always been rigged to match the predetermined election returns, but that maybe they were becoming more accurate now.
<Yes, she argued the US exit polls are generally rigged (fixed\"weighted" to match the election results). And, remember, this is based on over two decades of research by her father and uncle, and corroborated by at least two different sources that I have "stumbled" upon.

She didn't say "they were becoming more accurate now". What she actually said was that exit polls normally can't be trusted; but what tripped them (Mitofsky) up this time was: 1) the scrutiny caused in 2004, because of the 2000 election theft 2) the "leaking" (and consequent pubbing on the internet) of early exit poll results - which hadn't been weighted ("fixed") yet. Ask tia about this if you still "doubt". ;) )>


"I will not sit through that hour again."
Based on your responses, I think you should reconsider; and re-listen AT LEAST one more time - this time with a "neutral point-of-view". It seems you missed such things as her overall corroboration of the following: "Mitofsky uses a number of methods to "cover-up". It seems to range from computer meltdowns, "found votes", withholding the data, "convenient guesses", etc.
;)

[OnTheOtherHand]

114. you are still running from your sources

We can go around about whether "exit polls are very reliable," but the plain implication of the page you linked to is that the interviewing techniques are lousy. Other pages argue that the exit polls may not even be conducted. You may be of the view that the exit poll results are (1) uncannily accurate and (2) not supposed to ever be released. Actually, I can't tell what your view is.

You're the one who ran a whole thread called, "OMG! Mitofsky invented reluctant Bush responder scam for Bush Sr. in '92!" So don't bust my chops on semantics now.

I'm not saying that you have to believe Mitofsky is honest just because I do. But it's generally considered good form to present actual evidence if you think he isn't.

In 1992, the Democrat won; in 2004, the Republican (officially) won; both years there was red shift from the exit polls to the official results. That may be suspicious (at least to you), but it doesn't indicate a bias favoring winning candidates.

http://www.onlinejournal.com/evoting/111402Penn/111402penn.html

"On the next level, we need to consider VNS, the Voter News Service, founded as News Election Service (NES) in 1964. According to James Collier, co-author of Votescam and a 20 year student of computer vote fraud (especially in Florida), this vote projection service monopoly came about as a result of a meeting between representatives of the CIA, the FBI and the powers behind the media, who met in secret and hammered out a deal. If the heads of the major media would acquiesce in the Warren Commission's lame conclusion that Lee Harvey Oswald was the 'lone nut' gunman in the assassination of John F. Kennedy, then they would be given a great deal of power in the selection of the winners in future elections."

I am supposed to just ignore this in assessing the credibility of Victoria Collier's arguments about VNS? Really?

Are you claiming that 2004 is the first time that exit poll results were leaked, and in fact released, before being reweighted to official returns? Surely not, since it obviously wasn't. (I say "in fact released" because cnn.com posted state exit poll results shortly after polls closed. That wasn't a leak, it was fair use of information they had paid for. But the results reported earlier in the day were presumably leaked.)

Tom, if you want to convince someone who doesn't already agree with you, you need to reason from specific facts. It makes no sense to say that I should ignore Victoria Collier's assertions about JFK yet accept everything she asserts about Mitofsky.

[tommcintyre]

125. Sorry... now I have to lower your grade even further ;)

"We can go around about whether "exit polls are very reliable,"
No we can't. That argument is a non-starter - they ARE historically VERY reliable, unless Mitofsky gets a chance to "weight" ("fix") them.

Here, this info may help you (normally, I post only the link, but in this case I will re-post the entire text (in an attempt to discourage "cherry picking". ;)

http://en.wikipedia.org/wiki/Exit_poll
"Exit polls have historically and throughout the world been used as "parallel vote tabulation", as a check against and rough indicator of the degree of fraud in an election. Some examples of this include the Venezuelan recall referendum, 2004, the Ukrainian presidential election, 2004, and the 2004 U.S. presidential election controversy."
snip
"Leaks of exit poll figures for the 2004 presidential election, mainly via the Internet, appeared to indicate a victory for John Kerry. The discrepancies between the exit poll data and the vote count that where outside of the margin of error, coupled with irregularities in the election which seem to explain the discrepancies and what many perceive as evasive tactics by the polling companies, have shed doubt on the legitimacy of that election among political activists and government officials. (See 2004_U.S._presidential_election_controversy, exit polls for more detail.)"
-----------------------
Here, this is what may be causing your confusion:

http://en.wikipedia.org/wiki/2004_U.S._presidential_election_controversy%2C_exit_polls
"Because final published exit polls in America are matched to vote counts, they cannot be used to determine election fraud. However, in the 2004 election, pre-matched exit polls were leaked onto the internet."
snip
""Exit poll data - asking voters which way they voted as they leave the polls - are used around the world as excellent predictors of actual vote counts, usually accurate within a fraction of a point. Exit polls in this election seemed to match the vote tallies, as usual, except in those areas using touchscreen voting machines (like the Diebold Accuvote) or other software or modem-mediated electronic systems (like those from ES&S) with no paper trail - used by approximately one third of voters, many in swing states."
snip
"<in Germany>... people fill in hand-marked ballots, which are hand-counted by civil servants, watched over by volunteer representatives of the political parties. ... It's totally clean, and easily audited. And even though it takes a week or more to count the vote ... the German people know the election results the night the polls close because the news media’s exit polls, for two generations, have never been more than a tenth of a percent off."
------------------------

There are tons more sources like this; so like I said in the previous post, this issue is a non-starter. It's one of the weakest ways to argue against the high probability that this election was stolen. The argument is purely designed in an attempt to create doubt; and it does it very poorly.

Over the last seven months I have found that two groups of people attempt this argument: those who are poorly informed, or those who appear to be trying to mislead - and frankly, aren't very good at it.

"...but the plain implication of the page you linked to is that the interviewing techniques are lousy. <yada, yada, yada...>"
Are you trying to "slip by" an "all or nothing" mentality here? ;) Frankly, I don't care what (your opinion of) the "implications" of this page are - "plain", or otherwise. What is important is that the article reveals the following:

http://www.votefraud.org/News/2000/7/071800.html

"Mitofsky’s best guess is that Buchanan’s voters were prouder of what they had done and, hence, more prone to respond, than Bush’s were."

(Sixth paragraph, last line - to make it easier for you to find.)

"In 1992, the Democrat won; in 2004, the Republican (officially) won; both years there was red shift from the exit polls to the official results. That may be suspicious (at least to you), but it doesn't indicate a bias favoring winning candidates."
There ya go - bad comprehension again? ;) Mitofsky's "invention" for daddy Bush had NOTHING to do with who finally won the election, did it? Mitofsky made it up to "explain away" why the exit poll discrepancies on the GOP primaries (Buchanan vs. Bush) - now isn't that right?

<I think this is the part of your quote that you claim is so damning to the Colliers.> "...this vote projection service monopoly came about as a result of a meeting between representatives of the CIA, the FBI and the powers behind the media,..."
<So you say:>
I am supposed to just ignore this in assessing the credibility of Victoria Collier's arguments about VNS? Really?
Totally ignore? Of course not. But do you really have to adhere to such an "all or nothing", "black or white", "absolutist" mentality? Do you always throw "the baby out with the bathwater" as you are attempting to do here?

Instead, why not try and separate what is less "on the edge" and confirm, or disprove it with other sources?

"...if you want to convince someone who doesn't already agree with you, you need to reason from specific facts."

Here ya go - all the facts you care to consume:

<Lifted from this post:>
http://www.democraticunderground.com/discuss/duboard.php?az=show_topic&forum=203&topic_id=374482

Regarding Mitofsky:
http://www.democraticunderground.com/discuss/duboard.ph...
(I've pasted the most important parts here
"...you revealed some interesting things about Mitofsky:

1) He claimed rBr as fact in his original report. At MOST, this was a hypothesis, he knew that, EVEN his rBr "supporters" call it that. Plain and simple, legitimate professionals NEVER claim a hypothesis as a fact - by definition, it is simply a "guess", or "hunch" that must be investigated further. THIS ACTION ALONE IS A STRONG BREACH OF ETHICS. Thanks for pointing it out.

2) As you also pointed out, he also advanced rBr WITHOUT ANY SUPPORTING EVIDENCE. Again, this is something no ethical professional would dream of doing, in a formal report, as he did. They would wait until the investigation was reasonably done, and ONLY THEN publish their findings. Again, Thanks for pointing it out.

---------------------------
Febble, would you at least listen to this interview? I've read your paper twice, printed it out, and highlighted it and notated it, so this isn't too much to ask, is it?
3. I've listened to Victoria Collier's interview, and highly recommend it

http://www.edwardsdavid.com/BushVideos/infidelguy.com_V...
Also, read the first eight chapters (three more than at the votescam site) of the book here:
http://www.constitution.org/vote /

*<Verification here:>
http://www.votefraud.org/Archive/Write/newhampad.htm
"...early media reports stated that Bush and Buchanan were running neck and neck in the 1992 New Hampshire primary..."

<My original source:>
http://www.votefraud.org/News/2000/7/071800.html
-----------------------------------------------

And this is just the tip of the iceberg. Mitofsky has been at this since the mid-sixties.

I've found a clear pattern of <from another post of mine:>

http://www.democraticunderground.com/discuss/duboard.ph ...
"Mitofsky uses a number of methods to "cover-up". It seems to range from computer meltdowns, "found votes", withholding the data, "convenient guesses", etc.

So what may have happened in 2004? <To "trip up" Mitofsky.>

- More scrutiny because of publicity of the 2000 theft.

- Internet leaks"

<Some other related links:>

Election Night Projections - Cover For Vote Rigging Since 1964?
http://www.ecotalk.org/VoterNewsService.htm

2002 Elections: Republican Voting Machines, Election Irregularities, and "Way-Off" Polling Results
http://www.ecotalk.org/MidtermElections2002.htm

Suspicion Surrounds Voter News Service
http://www.ecotalk.org/VNSclosingdoors.htm

Could the Associated Press (AP) Rig the Election?
http://www.ecotalk.org/AP.htm

Votes Downloaded to AP - gets direct feed from voting tabulating computers
http://www.ecotalk.org/APvotes.htm

<You know, it was YOUR assertions of how "squeaky clean" Mitofsky is (in Kathy Dopp's thread the Thursday before Mitofsky's presentation) that tipped me off to check him out. (I was getting really "bad vibes" by what you were doing in that thread, so I thought I better check him out.) I really didn't have much of an opinion on him, either way, before that (just a mental note to eventually check him out, because of his appearance in the Jan 19th Nightline show). Once I started checking him out that Thursday, the "dirt" just started pouring in!

I'm curious, do all three of you (OTOH and Bruce) generally hold Mitofsky in such high regard? If so, is there a pattern here?>

So... all I have to say now is... GO DO YOUR HOMEWORK! ;)

[OnTheOtherHand]

126. on wikipedia, cherry-picking, and infallible exit polls

wikipedia is not an authoritative source, no matter how much of it you post. Did you notice the "edit" link? For all I know, you may have written that text yourself, although I doubt it. Regardless, the claim that "exit poll data... (are) usually accurate within a fraction of a point" is a stretch at best.

http://pollingreport.co.uk/record.html on British exit polls (since I assume there is no point in discussing the American ones)

In the 1992, 1997, and 2001 British exit polls, the lead errors ranged from 2.6 to 5.2 points. So much for the infallible exit polls.

If you want to tell me what part of Collier's evidence you regard as credible -- or indeed what you regard as evidence -- abstracting from her belief that the CIA killed JFK and then conspired with the major media to cover it up, be my guest.

[tommcintyre]

129. YOU REALLY SHOULD DO YOUR HOMEWORK BEFORE YOU CRITICIZE

C'mon oto, YOU can do better than that!

Did you bother to check the guidelines to making edits before spouting off? Did you read how contributors have to negotiate a neutral point of view (NPOV)? About the restrictions on citing sources? Did you bother to read the edit or history pages of the individual pages? <I use wikipedia here because of the quality checks/sourcing; and the fact that it saves time hunting down ALL this info that I have CONFIRMED elsewhere.>

In fact, DID YOU BOTHER TO DO ANYTHING OTHER THAN JUMP TO A CONVENIENT CONCLUSION?

oto, I honestly think you are really "grasping at straws" lately. (Pulled from all those "straw-men" you have created around here, maybe? ;) )

I think your arguments are honestly getting weaker and weaker. I don't think you can find better arguments, so, maybe, is it time to seriously consider changing your positions?

There is no shame in admitting it when you've made mistakes; so I strongly suggest you consider it, and move on. I think if you did, you could make some good contributions here.

One final point: No, I did NOT write any of the information at wikipedia yet. I will soon (about Mitofsky, rBr and the C/B Election Hearing - "weak spots" that need attention). That's WHY I have started to do MY homework and already know about the things I mentioned above (NPOV, etc.). When I do, you will know, because I will be sure to put my name on it.
----------------------------------

A link to get you started:

http://en.wikipedia.org/wiki/Neutral_point_of_view
Wikipedia:Neutral point of view
"Wikipedia policy is that all articles should be written from a neutral point of view: without bias, representing all views fairly. According to Wikipedia founder Jimbo Wales, NPOV is "absolute and non-negotiable". <1> (http://mail.wikipedia.org/pipermail/wikien-l/2003-November/008096.html)

For guidance on making an article conform to the neutral point of view (NPOV), please see the Wikipedia:NPOV tutorial."
----------------------------------

P.S. I don’t see any point in answering any more of your direct questions until you get a handle on this “bias” (cognitive dissonance?) problem. Until you do, it is CLEARLY and excercise in futility. <Although, I certainly will comment if (IMO) I see you attempting to spin or miss-lead.>

Good luck! I'm really pulling for ya. :D

[OnTheOtherHand]

130. well, in fact, I obviously did: I posted a direct factual refutation

which you ignored. In fact, you seem to have ignored just about everything in my post except the word "wikipedia."

If you believe that everything at Wikipedia is neutral, then you may just believe that there are no personal attacks on DU, because that is "policy" too. But even if we do assume that everything at Wikipedia is "neutral," that does not demonstrate that everything at Wikipedia is accurate, nor that Wikipedia is an authoritative source for information on exit polls.

I gave you a fact about British exit polls. You gave me irrelevancies about NPOV. Hmmmm.

[tommcintyre]

131. Well... "now its back to the misunderstanding and spin"

As RonB said here, and I think this statement epitomizes (and sums up) an overall problem:

http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x374482#378637

We had our moment when you finally came up with a specific point - but now its back to the misunderstanding and spin I guess.

You didn't bother to read the discussion, edit or history tabs for these pages, did you? Or, if you did, you're purposely ignoring them?

As anyone can see, I have painstakingly answered your points in previous posts. Not only here; but in other sub-threads also.

In contrast (IMO), you have mostly ducked, bobbed, weaved, misled, misdirected, "misunderstood", obfuscated... and, oh yes spun like a top.

So (IMO), it is long overdue that I no longer take your questions seriously until you straighten out this problem that I perceive.

C'mon oto, be honest. I've answered much tougher "diversions" (IMO) than you have thrown up in that last post, haven't I? ;)

<Caution: If you bother to answer this last question at all; your answer (one way or another) will be VERY revealing, won't it?>

GOOD LUCK! :D

[OnTheOtherHand]

132. and your response to the British exit poll results was...?

If you can hop around finding second-hand potshots to copy and paste, I am sure you can succinctly explain why the fact that all those British exit polls were off by several points has nothing whatsoever to do with whether exit polls are "usually accurate within a fraction of a point."

And as to whether the U.S. exit polls were designed to detect fraud, even the Wikipedia article doesn't claim that. You can, if you wish. You might want to check it with Victoria Collier first.

[tommcintyre]

133. Like I said, no more going in circles with you - fix your problem first nt

[TruthIsAll]

38. We are still waiting for your evidence...

OTOH, if you cannot cite specific evidence for this "non-response" theory, then its case-closed.

The burden of proof is now on you to prove the election was NOT stolen.

[LightningFlash]

41. That's because there is no evidence of EDR except one example years ago.nt

[OnTheOtherHand]

43. unfortunately, it isn't

TIA, I frankly don't care whether you believe the election was stolen, so I bear no burden of proof.

[autorank]

44. Excuse me?

O, if I may be informal, you said "You may just be the first person in the world to conclude instead that Mitofsky was covering up for Bush. But I wouldn't exactly say that you have pr oven it."

The first person? The posts here are replete with questions about motives. After all, why would the poll results change from 51-48 Kerry-Bush at midnight to Bushs WINS at 1:00am or so. That raises the question. The Mitofsky self-deprecation tour, where he gives speeches and releases to the press on his errors is another "hint."

Now, you may say, why didn't he fix the exit poll in advance. Well, that would presume a before the fact conspiracy, one that would be very difficult to carry out. It's the ad hoc element that gives rise to real questions that cause inquiring minds to want to know.

Here's an analogy. If I'm a big wig in LargeVille and I have a problem son, I'm not going to know when he'll get in trouble. I can't go and "piece off" the entire police force before-the-fact or even know when little Wilbur is going to get in trouble. But as the Grand Poobah, I certainly can, at a moments notice, take care of business after the fact. This is how these things are done.

That establishes the questions and the rationale for a valid poll but a skewed final result and it also gives rise to some questions about the curious after-the-fact self deprecation tour.

[OnTheOtherHand]

46. ah, happy to clarify

I have no idea who was the first person to accuse Mitofsky of covering up for Bush. But I wonder how many cited that particular article (which questions the accuracy of exit polls) as evidence.

Tom keeps citing Victoria Collier. As (or if) I understand her argument, it's that the exit polls may not even be conducted, and almost certainly are rigged to match the also-rigged official returns.

http://www.sweetliberty.org/issues/votescam/stolen.htm

Collier doesn't seem to find it hard to believe that the exit polls are rigged in advance. Neither does Lynn Landes:

http://www.voternewsservice.com/

But, for what it's worth, I do. As for everyone else, I don't even know whom to attempt to rebut any more.

[eomer]

49. How USCV got from (x) to (a)

OTOH said:

"I'm not quite sure how USCV got from (x) to (a), which I would characterize as the theory that this difference in participation rates would in no way depend on precinct partisanship."

Ron Baiman has addressed this question at least once here in this forum. Here is an excerpt from one of his posts:

The point here is that any "operational" or "behaviorally meaningful" definition of "rBr" has to imply that there not be statistically significant differences in bias (K/B) across different partisan groups of precincts. If there are such differences, than these need to be explained by some other factor(s) than simply "universal" rBr.
Again we're back to the need for a real substantive explanation.

The flat linear correlation (if this is the new - or old or whatever -definition of the E-M/Liddle "hypothetical" rBr non-hypothesis) really gives us very little, to nothing, with any explanatory significance for the exit poll discrepancy.

USCV focused on a non-varying mean K/B hypothesis as this could (if it were true) be evidence for an "rbr" hypothesis with real explanatory meaning. In contrast a "flat linear correlation rBr hypthetical" seems to be a relatively meaningless finding.

http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x372464#373121

[OnTheOtherHand]

50. ok, so my response to Ron would be

(1) It is surely an overstatement that any operational definition of differential response rates must entail no statistically significant differences in bias across different partisan groups.

(2) I agree that if such differences exist, they need to be explained.

(3) There appear not to be statistically significant differences; if there are, they almost certainly hinge on the four outliers among high-Bush precincts.

(4) A "flat linear correlation" (with or withour an rBr hypothetical) is not meaningless, because many -- but perhaps not all -- versions of massive vote shift would be likely to induce a positive correlation.

[eomer]

51. Response

(1) It is surely an overstatement that any operational definition of differential response rates must entail no statistically significant differences in bias across different partisan groups.
True, but what if the response rates that would be required to explain the data are implausible in the sense that they fly in the face of what is intuitive?

(2) I agree that if such differences exist, they need to be explained.
We agree.

(3) There appear not to be statistically significant differences; if there are, they almost certainly hinge on the four outliers among high-Bush precincts.
USCV apparently does not agree with you that the outliers can fully explain the difference. Ron also makes the point that you should look at the means and medians and that you should not throw away outliers while doing so unless you have a reason to conclude they are anomalous. That said, I agree it would be interesting to see what the results are with those four outliers taken out.

(4) A "flat linear correlation" (with or withour an rBr hypothetical) is not meaningless, because many -- but perhaps not all -- versions of massive vote shift would be likely to induce a positive correlation.
I don't think Ron meant that a flat linear correlation was meaningless. What I think he meant was meaningless is a presumptive rBr where the only criterion is that it produce a flat linear correlation. In other words, no matter what extreme response rates you have to come up with to explain the data, they can be as extreme and non-intuitive as you would like but as long as they produce a flat linear correlation then we presume that they are the explanation. I think he is saying that an rBr of that type has little or no explanatory value.

[OnTheOtherHand]

53. post without a clever title

First, it might be good to step back. On the disagreement between O'Dell and the May paper, I stand with O'Dell -- so far nothing you've said is moving me off that dime. On issues about how to interpret the exit polls, I have opinions, but there is lots of room for interpretation. I suppose that's one of my opinions; the May paper seems to rule out interpretations that I think cannot be ruled out. But that doesn't mean that I am ruling out your interpretations (not that I know what they are!).

(1) If I believed that a large value of "alpha" in a precinct entailed a large response (completion) rate among Kerry voters, then this argument would have more force for me. I don't think that assumption is tenable. We cannot reliably estimate partisan completion rates after the fact -- we can't sort out the effects of sampling error, selection bias, non-completion bias, and possible fraud. I'm of course willing to consider whether the levels of red shift seem implausibly large, but I'm not willing to accept a hypothetical value of "K," extrapolated from other data upon dubious premises, as independent evidence.

(2) I would only add that from my point of view, the existence of precinct outliers in itself is not a strong indication of fraud. I don't think many folks at AAPOR looked at Mitofsky's slide and thought, "Ah hah! There is the smoking gun!" Nor of course is it a strong indication against fraud.

(3) Since Bruce O'Dell is vice president of USCV, I will not undertake to agree or disagree with USCV, but merely note the absence of consensus. I don't think the high-Bush precincts can sustain the weight that the May paper places on them. But we agree that we would like to know more about them.

(4) I'm not sure I understand this, to know whether I agree with it or not. If the point is that it is quite possible that strong patterns suggesting fraud could occur even with a zero linear correlation, then I agree. But I see no such patterns. "a presumptive rBr" etc. -- that does sound pretty meaningless, and I'm not sure it has anything to do with Edison/Mitofsky.

Do you see strong evidence of fraud, and if so, what is it?

[eomer]

54. Post with a title that claims not to be clever but tries to be nonetheless

First, it might be good to step back. On the disagreement between O'Dell and the May paper, I stand with O'Dell -- so far nothing you've said is moving me off that dime. On issues about how to interpret the exit polls, I have opinions, but there is lots of room for interpretation. I suppose that's one of my opinions; the May paper seems to rule out interpretations that I think cannot be ruled out. But that doesn't mean that I am ruling out your interpretations (not that I know what they are!).
The subject of this thread was whether we can rule out one particular interpretation, which is constant mean rBr. The May paper rules out that interpretation and then says we should look for other interpretations. What other interpretations does the paper rule out?

(1) If I believed that a large value of "alpha" in a precinct entailed a large response (completion) rate among Kerry voters, then this argument would have more force for me. I don't think that assumption is tenable. We cannot reliably estimate partisan completion rates after the fact -- we can't sort out the effects of sampling error, selection bias, non-completion bias, and possible fraud. I'm of course willing to consider whether the levels of red shift seem implausibly large, but I'm not willing to accept a hypothetical value of "K," extrapolated from other data upon dubious premises, as independent evidence.
You make basically the same point as the USCV working paper. When you extrapolate K and B from the data you find that they are not, by themselves, a sufficient explanation. You need to look for other factors, either with or without rBr, to explain the data. That is the main premise of the paper.

(2) I would only add that from my point of view, the existence of precinct outliers in itself is not a strong indication of fraud. I don't think many folks at AAPOR looked at Mitofsky's slide and thought, "Ah hah! There is the smoking gun!" Nor of course is it a strong indication against fraud.
Agreed. The outliers by themselves are not going to tell the story.

(3) Since Bruce O'Dell is vice president of USCV, I will not undertake to agree or disagree with USCV, but merely note the absence of consensus. I don't think the high-Bush precincts can sustain the weight that the May paper places on them. But we agree that we would like to know more about them.
I can't really respond to the part about weight unless you can be more specific about some conclusion you don't agree with. If you're talking about the conclusion that constant mean rBr is ruled out then I think they can sustain that weight. If you mean some other conclusion then let me know which one.

(4) I'm not sure I understand this, to know whether I agree with it or not. If the point is that it is quite possible that strong patterns suggesting fraud could occur even with a zero linear correlation, then I agree. But I see no such patterns. "a presumptive rBr" etc. -- that does sound pretty meaningless, and I'm not sure it has anything to do with Edison/Mitofsky.
That is not quite the point. The point is that an implausible pattern of response rates is not rendered plausible just because it produces a zero linear correlation.

Do you see strong evidence of fraud, and if so, what is it?
The subject of this thread was whether constant mean rBr has been refuted, not whether there is evidence of fraud. If we're going to launch into this new subject there is another thread that would be more appropriate.

But so you won't think I'm dodging the question, yes, I see lots of evidence of fraud. I won't try to list all of it here but will just give a brief high-level summary and I'm sure I'll forget something major, there is just such a mountain of it. Off the top of my head:

1. The only indication that Bush won is the official count. All other indicators, including the exit poll and pre- and post-election polls and popularity polls, tell us that Kerry won.
   
2. The fact that the Republicans were determined to not allow transparency and other safeguards in our election system is strong evidence that they intended to and did steal votes.
   
3. The intentional violation of election rules to avoid a recount in Ohio indicates that there was something there to hide.
   
4. There is evidence of physical tampering with ballots in Ohio. For example, stickers were placed on opscan ballots. Punch card decks had clearly been sorted while there was no step in the observed process that would have caused them to be sorted.
   
5. The fact that most or all Democrats on the BOEs in Ohio are obviously DINOs and that it is intentional and systematic that they be DINOs.
   
6. The fact that Blackwell applied coercion to BOE board members by basically telling them they must do exactly as he said or lose their jobs.
   
7. The fact that rulings on questions of law and regulations in Ohio were handled by way of private, secret phone calls to the SOS, also the co-chair of the Bush-Cheney campaign, rather than through some transparent lawful approach that created a documentary trail.
   
8. Clint Curtis' story shows that there was intent within the Republican party to steal votes.
   
9. The Raymond Lemme "suicide" and the facts surrounding it.
   
10. The revelations in the last couple of weeks from Bob Fitrakis that subcontractors have told him that Blackwell had a link to the statewide tabulators that would allow him to "make corrections" to the county election results.
    
11. The Florida felons list in both 2000 and 2004 are clear evidence of intent to commit fraud.
    
12. The allocation of voting machines in Ohio is evidence of fraud. Probably elsewhere too. At the University of Miami (Florida) the story was not that different from what happened in Ohio.
    
13. The DREs that switched votes from Kerry to Bush.
    
14. The implausible relationship between the presidential vote and the vote for certain down-ticket races in certain counties in Ohio.
    
15. Various incidents of vote totals going up in unlikely amounts and/or proportions, sometimes with suspicious timing such as only in the final tally or when a machine was taken offline.
    
16. The use of a fake terrorist alert to lock down the counting in one county in Ohio.
    
17. Numerous reports of voters being purged from registration lists when they should not have been.
    
18. Numerous reports of dirty tricks such as leafleting neighborhoods with false information about time and place of polling.
    
19. Other dirty tricks such as jamming a GOTV call center.
    
20. The shambles that our election system is in, specifically the fact that safeguards such as a chain of custody for ballots have been broken or discarded.
    
21. The extremism and blind loyalty of so many "neocon" supporters convinces me that many of them would do something dirty without being specifically directed to if they found themselves in a position of opportunity.
    
22. The developing story that large amounts of money have shown up missing in Ohio and that some of those involved have a connection with the election story.
    

If you're not convinced by any one of those by itself, take a look at all of them together and see if that doesn't do it for you.

Maybe you want to respond and tell me whether you see any evidence. Then I think we should call it quits on this tangent and move it to a thread where it is on-topic.


edit: remove a dup in the fraud list

[OnTheOtherHand]

59. I should've said, "post without a USEFUL title"

Moving on... lots of great stuff here, I will try to do it justice while keeping my job.

I think the ordinary-English meaning of the May working paper is that (a) E/M enunciated a hypothesis, commonly called the rBr hypothesis, that entails a constant mean bias (whatever that is), and (b) that the working paper gravely infirms that hypothesis. (The abstract also implies that vote shift provides a parsimonious explanation of the known results.) Do you agree with my reading?

I don't think the May paper lays a glove on what E/M actually said about non-response bias. If the paper was all about ruling out one possible interpretation of what E/M might have meant, then it might have been more productive to begin by asking the principals, "What did you mean?" (And yes, in retrospect, I wish I had thought of that myself.) Otherwise, do we really care whether there is "constant mean bias"? or do we care whether there is evidence pointing to fraud?

(1) Actually, I think that I basically disagree with the working paper, in that it assumes that we can usefully argue in terms of K and B, and I do not. But setting that aside -- well, again, if the paper regards the high-Bush precincts as strongly suggestive evidence, then I just can't agree; but of course that doesn't mean I assert that everything in the results has been explained.

(2) Clinging to a clear point of agreement, I move on.

(3) If the May paper intends its rebuttal of "constant mean bias" to be pertinent to the probability of fraud, then I don't see it. If it is OK to interpret the rebuttal as verging on the formalistic and irrelevant, then we can agree. (I feel that I have been saying the same thing over and over in various words, but only because we seem to come back to it.)

(4) I agree in principle that a zero linear correlation does not render an implausible pattern plausible. I don't find an "implausible" pattern in the E/M data. I don't look at the outliers and say, "Aha! Looks like vote shift!" On the contrary, if the election was stolen through central-tab vote shift, I wouldn't expect the action to show up in the outliers. (I've explained elsewhere why I expect the high-Bush precincts to look ordinary if the outliers are set aside.)

-----

I should have clarified that I intended my question about "evidence of fraud" to be limited (like the rest of the post) to the exit polls. That isn't because I'm trying to ignore all those issues -- I just think (as I gather you do) that it is unfeasible to discuss them all at once. Many of your points focus on Ohio and Florida; I have no trouble believing that the Republicans used corrupt means to try to win Ohio and Florida, or even that they may have succeeded.

Regretfully (because I think John Kerry is a far better man than George W. Bush is), professionally I cannot agree that "The exit poll says that Kerry won." (That's the language from your other post.) Exit polls do not speak in that way. In my view, the exit pollers and networks have learned not to take the raw exit poll data as literally as some of us seem to insist upon doing, and that is why they did not call Ohio or the popular vote for Kerry.

Nor can I agree that the pre-election polls tell us that Kerry won, much as I wanted to believe that at the time. As far as I know, the roundup here is reasonably complete and accurate

http://pollingreport.com/2004.htm

and the average of 14 trial heats shows Bush ahead by about 1.45 percentage points. (The vote projections show Bush ahead by a smaller margin.) The spreadsheet of final state polls available at

http://www.electoral-vote.com
http://www.electoral-vote.com/2004/nov/nov02.xls

shows a dead heat in the electoral college.

(I'm not convinced by the rest of your fraud point #1 either, but those are the most important parts in my mind.)

[anaxarchos]

71. Sorry to pile on here but...

The big problem with the LV pre-election polls you cite is "turnout". I hate to use Blumenthal for anything but it is fitting that the "Democratic pollster" actually support "Democrats" at least once:

http://www.mysterypollster.com/main/2004/11/lessons_likely_.html

Blumenthal (10 days after the election) talking about the "election" of Bush versus the pre-election polls:

"However, that margin of error applies to only to one survey at a time. As Prof. X pointed out, the same logic does not apply to an average of 15 or more polls. In fact, most of the surveys done in the final week (11 of 15) had a “bias” toward Kerry – they showed Kerry ahead or had the margin closer than three points. That result cannot be explained by random variation: My application of the binomial distribution puts the probability of that happening by chance alone (assuming that undecideds broke evenly) at roughly 6%."

Of course, Blumenthal understates the case. We know very well that undecideds most assuredly did not break "evenly", etc. But, it gets worse...

"Some likely voter models worked better than others - When I presented the details on likely voter models, I noticed that the pollsters that used a variant of the Gallup likely voter model showed Bush doing consistently better than other surveys. That difference now looks prescient. The following table shows the results of those using the Gallup likely voter model either in the final week (Gallup, Pew, Newsweek) or in the final two weeks (adds Time and the LA Times). In both cases, the Gallup-model showed a Bush margin closer to the actual result (3.2%+) than the average of the other surveys (0.9%). There were three surveys in the “other” category that correctly forecast Bush’s final three-point margin (notably, Pew, TIPP, ICR), but the other 10 showed Kerry doing slightly better."

The irony here is that Gallup's LV model was heavily criticized pre-election precisely "because it uses a "cutoff" for categories like "young first time voters" and thus fails completely in high turnout elections". Blumenthal himself was one of the critics.

The problem with what you rest your case on is 120 million odd votes...

Of course you could use the RVs - but then, they show the opposite.

Finally, you can't get fired from Bard for ANY political position you take, can you? That sure ain't the Bard I remember...

[OnTheOtherHand]

92. to your last, no...

I meant that I need to do my day job, not that I will lose it for taking unpopular positions. I'm not aware of anyone gaining tenure for their incisive posts on Democratic Underground (or even for the rambling and incoherent ones, alas).

As to the rest -- if we assume that the Gallup LV model was wrong, that still leaves you with a Bush +0.9% average for the others. That suggests a very close race, but I don't see how it supports the claim that the pre-election polls show that Kerry won. (Apologies to eomer if that is a freehand paraphrase, but I hope it is recognizable.)

[anaxarchos]

104. The point was that virtually all LV models break at 120 million votes...

... but forget about it. We would have simply gone down an obscure rathole... I'll answer your obscure rathole question instead.

[OnTheOtherHand]

112. sorry, I didn't mean to skip that point

A lot of us figured before the election that Kerry would pull it out, inter alia because we didn't trust the LV models. And it's an open question whether we should. I just think it's a stretch to insist (in effect) that we know we were right all along. I don't think the pre-election surveys are strong evidence against massive fraud.

[TruthIsAll]

105. OTOH, misrepresenting the facts again.

11 out of the final 18 national polls had Kerry in front.

At least 2/3 of the undecided vote always goes to the challenger.
As a political science prof, you should know that.
That makes Kerry a projected 51.8-48.2 winner of the 2-party vote

Your ignorance of pre and post-election polling facts is astounding.
As a political science prof, you should know them.

We have to teach you so much at DU.

http://www.geocities.com/electionmodel/

[OnTheOtherHand]

113. it sure would be nice if your links supported your "facts"

"11 out of the final 18 national polls had Kerry in front."

I followed your link, to the pollingreport.com page, which showed once again that 10 of 13 final trial heats and 5 of 7 vote projections (not counting Gallup, which was a tie) favored Bush. If you want to critique your own source, that's fine, but please don't tee off on me.

"At least 2/3 of the undecided vote always goes to the challenger."

I similarly followed through to the Chris Bowers article. Bowers' methodology cannot support your claim, because he aggregates across all polls; he reports no minimum percentage of the undecided vote that "always" goes to the challenger.

Bowers actually says (apparently referring to non-presidential races): "66-34 is where the smart money, the house money, should be. You will win over the long term betting on that sort of a split, but in any given wager you might get burned." And he shows that in non-presidential races, the overall average split was 60:40 in 2000 and 58:42 in 2002-04 (it's not clear what '04 results are included). There may be a compelling reason to expect a 2/3s rule to be infallible in presidential races, only, but I don't see where anyone has given one.

Bowers does conclude that the overall aggregate split in presidential polls is 86:14 in favor of the challenger, but this is based on just 28 surveys from 1976 through 2000, or four per year. It is hard to assess the analysis without the data, but it's clear that his method will give heavy influence to surveys with lots of undecideds, although (as he reports) most of the surveys have very few undecideds.

Gerry Dales' analysis at http://www.dalythoughts.com/Update-05-26-04.htm shows that undecideds in next-to-last Gallup polls (the last ones in which undecideds are not allocated) from 1936 through 2000 more often broke toward the incumbent or incumbent party than toward the challenger. This doesn't necessarily contradict Bowers' result, but it is certainly discrepant data for an objective analyst to consider.

FWIW, the 2004 national exit poll, unweighted, indicates that voters who decided "just today" or "within the last three days" broke Kerry by 56:39. 54:41 for the "just todays."

I feel sort of foolish for actually trying to sort out the evidence on this point, when you apparently can't be bothered. I'm not sure what the bottom line is on undecideds -- it probably varies from poll to poll and from election to election -- but certainly your "At least... always" is unsubstantiated.

[eomer]

55. Taking a step back...

Sorry to post a second response but I was thinking about your post and my response and it struck me that if we really take a step back, here's what we find:

The exit poll says that Kerry won.

[TruthIsAll]

56. Forgive my intrusion. OTOH, I have three simple questions ...

What do you currently see in the E-M data that indicates rBr is a viable hypothesis?

If there is nothing currently in the data to lend legitimacy to rBr, then I ask you: What data would you need to show that rBr is a viable hypothesis?

Finally, what data would you need to see to be convinced that rBr is without foundation?

[OnTheOtherHand]

58. let me try to answer those questions briefly

1) rBr (or, more generally, differential participation) is inherently viable because non-response bias is a common issue. The relationship between "interviewing rate" and WPE seems much more likely to be explained by selection and/or non-response bias than by something in the pattern of fraud. (Do you have an alternative explanation?) So, I consider that strong circumstantial evidence of differential participation, although it doesn't indicate whether differential participation could effectively account for "all" the net discrepancy between the exit polls and the official results.

(Actually, I would be surprised if it did: I am morally certain that some exit poll participants cast provisional ballots that were later thrown out. Bush could have stolen Ohio and/or Florida -- feel free to add other states -- even if he didn't steal the popular vote. If I ever get a chance to stop arguing about the national polls, and if ESI finally produces their work on Ohio, I will try to form clearer opinions there.)

2) Although I don't accept the "if" premise, I agree with everyone else that multivariate analysis would really help us to assess whether differential participation could explain that nationwide discrepancy. I don't think that the measured variables have to account (statistically) for the whole gap, but if they account for almost none of it, I would really wonder. Between the extremes of "whole" and "almost none," it would depend on the specifics.

3) Besides what I just said, I don't have any great ideas right now, but I'm open to them.

There, I managed to be brief at least once.

[tommcintyre]

60. Forgive mine too. Perhaps this post was "lost in the shuffle"?

After-all, this thread is getting more difficult to find one's way around in.

I still am very interested in your reply to this post, should you care to make one.

http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x378368#378949

Thank You.

[OnTheOtherHand]

69. whoops -- see above, #63 n/t

[TruthIsAll]

62. Thanks for being brief. But let us continue....

You say:
1) rBr (or, more generally, differential participation) is inherently viable because non-response bias is a common issue. The relationship between "interviewing rate" and WPE seems much more likely to be explained by selection and/or non-response bias than by something in the pattern of fraud. (Do you have an alternative explanation?) So, I consider that strong circumstantial evidence of differential participation, although it doesn't indicate whether differential participation could effectively account for "all" the net discrepancy between the exit polls and the official results.

I say:
But WPE is essentially just the difference between the exit polls and the vote. You are implicitly assuming that the vote count was correct and you ignore the reams of contrary evidence eomer (22 items) has presented which indicate otherwise.

You say:
(Actually, I would be surprised if it did: I am morally certain that some exit poll participants cast provisional ballots that were later thrown out. Bush could have stolen Ohio and/or Florida -- feel free to add other states -- even if he didn't steal the popular vote. If I ever get a chance to stop arguing about the national polls, and if ESI finally produces their work on Ohio, I will try to form clearer opinions there.)

I say: You are ignoring the fact that Bush exceeded the MoE from the poll to the vote in at least 16 states. This is much more than just about OH and FL. What about NY, MA, PA etc.? And you ignore the fact that ALL 22 states in the eastern time zone deviated from the exit polls to Bush.

You are implicitly assuming that the Final 13660 et poll was correct, yet you still ignore the mathematical FACT that Bush 2000 voters could comprise no more than 39.8% of the 12.3 mm 2004 voters.
The 13660 Nat Exit Poll says that the split was 43%Bush/37% Gore.
This is not only impossible, it flies in the face of rBr.
BTW, at the 13047 time line, which Kerry won, the split was also an IMPOSSIBLE 41%/39%.

You have never responded to my posts in the past on just this topic.
Will you do so now?

You say:
2) Although I don't accept the "if" premise, I agree with everyone else that multivariate analysis would really help us to assess whether differential participation could explain that nationwide discrepancy. I don't think that the measured variables have to account (statistically) for the whole gap, but if they account for almost none of it, I would really wonder. Between the extremes of "whole" and "almost none," it would depend on the specifics.

I say:
I agree, but we need the raw data for that. How can we get it?

You say:
3) Besides what I just said, I don't have any great ideas right now, but I'm open to them.

There, I managed to be brief at least once.

I say:
Ok, will you explain the 43% Bush/37% Gore How voted in 2000 split?

[Internut]

64. "Ok, will you explain the 43% Bush/37% Gore How voted in 2000 split?"

It is common knowledge in polling community that people routinely lie to pollsters about their past voting behavior. You have been told this several times in response to this question.

[mgr]

67. Actually there are other explanations

1. The MOE is not 1% but 6.5%.

2. The category is uninformative for future elections, therefore not reweighted, or taken into account when reweighed.

Mike

[TruthIsAll]

73. You are flailing. You are total denial. 6% MoE in the National exit poll?

You are making a farce out of this.
Are we back to arguing about MoE?

When Mitofsky himself says there is a 1.0% MoE?
I thought we were past that.

Having any kind of rational discussion with you is impossible.

[mgr]

78. Mitofski's design is one thing, what was the outcome is something else.

You've never verified or refuted this. It was cited in several places that Mitfoski reported that the MOE for the 2004 was 6.5%. What you are using is the MOE expected if the WPE was not swamped by bias. I also gave you an equation for how it might be calculated. Did you ever try it?

Mike

[mgr]

86. Link (from a google search - Mitofski MOE-

http://www.mysterypollster.com/main/2004/12/exits_were_the_.html

"...Finally, we know that Edison/Mitofsky reported MoE's associated with their exit polls PRIOR to the election of 3% to 7%."

Mike

[mgr]

115. Let me lay the argument out for you once again

The application of the MoE to a sample population obtained from combining smaller sub samples requires that conditions of the central limit theorum be met. The most critical of these is that the subsamples vary from the parametric population mean (mu)independently (in both directions), and often with a range with a magnitude less that the subsamples' MoE. We know from the January 19 NEP report that the precincts sampled for the 2004 election show a tendency to vary directionally from the anticipated towards Kerry, with an average WPE of 6.5%.

The 1% MoE you cite, is a pre election estimate, should all sampling constraints be met. Those constraints are not stated, but we can infer that a WPE of 3% seems to be one constraint--sampling that mischaracterizes one response when the refusal rate is running 40% would meet this (and this also operates within the range of the MoE with the precinct population).

If the WPE is a mean of 6.5% and directional, we have to widen the MoE to accommodate that difference. It does not matter what produced that directionality. That means that we either log transform the data to normalize it; or the MoE has to be increased. I provided one equation to do this:

For n precincts MoE (total population)= sum(MoE N1/n1+ MoE N2/n2+...MoE Nn/Nn)

If you were to apply the equation, the MoE for the entire sample (n)would approximate the average WPE.

Mike

[mgr]

68. Actually one more.

The Democrats stole the election from Bush <smile>.

Mike

[TruthIsAll]

70. mgr, what doEs it take for you to understand the problem?

FORGET the poll.
The 43% is IMPOSSIBLE mathematically.

It has NOTHING to do with polling.
It has EVERYTHING to do with the MAXIMUM possible number of Bush and Gore voters who could have voted in 2004.

Can you wrap your mind around that simple fact?
How many times do I have to do the numbers for you?

Bush 2000 voters: 50.456 mm
Bush 2000 voters still alive: 48.69
122.3 million voted in 2004.

48.69/122.3 = 39.8%

What is it about simple arithmetic you don't understand?

[mgr]

79. Not if the dems stole the vote.

That post was tongue in cheek. But maybe you don't register the sardonic, in that its the same problem that plagues Hout's paper--e.g. that the opposite argument can be made applying the same assumptions.

Mike

[TruthIsAll]

83. Enough of this. I deal in facts. You avoid the facts by being sardonic.

I'm not sophisicated as you are in sophistry.

THIS IS A SIMPLE ALGEBRA PROBLEM - NO, ITS ARITMETIC ACTUALLY.
WHEN WILL YOU AND YOUR BUDDIES START TO DEAL IN FACTS?

MUST I LEAD YOU BY THE HAND, LIKE YOUR 3RD GRADE MATH TEACHER DID?

[mgr]

88. What level of argumentation are you applying?

Logic, scientific, legal? It is not sophistry, but the fact that your argument's assumptions cut in both directions--What if the 2000 exit poll had it wrong--they undersampled Bush support due to fraud? And the 2004 exit poll is correct--now they got it right? That is the symmetry I am talking about.

What is a mathematical proof? A fact or an argument? What is it when you compare two estimates? I think you engage in sophistry with your distinction.

You have to do better.

Mike

[TruthIsAll]

75. Thanks for joining in Internut. I bet you can't do simple arithmetic.

Here is the same challenge I put to OTOH.

I want you to calculate the MAXIMUM PERCENTAGE OF BUSH 2000 VOTERS OF THE 122.3 MILLION WHO COULD HAVE VOTED IN 2004.

HINT:
BUSH GOT 50.456 MM VOTES IN 2000.
APPROXIMATELY 48.7 MM ARE STILL ALIVE.
NOW CALCULATE THE MAXIMUM PERCENTAGE OF BUSH 2000 VOTERS.

AND WHILE YOU ARE AT IT, DO THE SAME FOR GORE 2000 VOTERS.
GORE GOT 50.999 MM VOTES

THAT IS THE TOTAL MATH EXAM.
WILL YOU PASS OR FAIL?
GOOD LUCK!

[Internut]

76. You keep saying it is impossible -

yes, it is impossible that that many people ACTUALLY VOTED for Bush in 2000. But it is not impossible that that many people TOLD POLLSTERS they voted for Bush vs Gore. What people say and what actually happened are two different things. You refuse to understand it.

[TruthIsAll]

80. I am NOT disputing that. YOU fail to understand that in this case

we KNOW the maximum.

AT LEAST YOU ADMIT THAT 43%/37% IS IMPOSSIBLE.

WHY TALK ABOUT POLLSTERS LYING WHEN YOU HAVE THE REAL NUMBERS I JUST GAVE YOU?

THIS HAS NOTHING TO DO WITH THE POLL.
GET OFF OF THAT STRAWMAN.

NOW LET'S TAKE IT TO THE NEXT STEP.

WOULD YOU ALSO AGREE THAT BY USING THIS 43%/37% DEMOGRAPHIC WEIGHTING, THE BUSH VOTE PERCENTAGE IS ALSO IMPOSSIBLE - BY DEFINITION?

[Internut]

81. I don't know how I can explain it clearer -

I suspect that no matter how well it is explained, you will still choose not to understand.

1. It is a well known fact that when queried about their past voting, people routinely misreport it to pollsters.
2. Because of that, any numbers that relate to past voting patterns that are revealed in polls are unreliable.
3. Because of that, those numbers would not be used in order to weigh the exit poll numbers. You don't use bogus numbers for weighting purposes.

To recap - the 43%/37% Bush/Gore voting in 2000 that shows up in the 2004 exit poll is obviously incorrect, since Gore and Bush were roughly equal in popular vote in 2000. Yet this does not mean that the people who answered the question "whom did you vote for in 2000" did NOT answer it in the 43%/37% ratio.

[TruthIsAll]

85. You are too much. Denial is not to strong a word. Let me repeat.

We are not talking about the poll.
We are talking about the mathematical reality.

FORGET THE DAMN POLL FOR A MINUTE.
Just focus on the actual numbers.

You have the actuals numbers, why not use them?
Do the math.
The longer you deny the facts, the more credibility you will lose.

Do the math.
Then we will proceed to the next step.

Do you dare, Internut?
Do you dare to do the simple arithmetic?

Jeez, your mind is sure a closed trap.

[Internut]

87. Where do you take the "43%/37%" from?

You take it from the poll. Then how can we "forget the poll" if that is where you take that number from?

[OnTheOtherHand]

66. let's see how fast I can do this

1) No, I'm not assuming that the vote count is correct. I'm saying that it makes sense for selection and non-response bias to be higher in precincts with higher interviewing rates (the N in "every Nth voter"), because a higher rate gives more discretion to interviewers.

I see little in Eomer's list that sheds light on the national popular vote. I disagree with his assertion that the pre-election polls warrant a strong expectation that Kerry would win. I think it is quite possible that Bush "stole" Ohio and Florida whether or not he stole the national popular vote; I do not think that an argument for any one manifestation of fraud (or vote suppression) constitutes an argument for all manifestations of fraud.

I am "ignoring" the MoEs because they do not and cannot reflect the possible extent of differential participation -- they cannot test whether "rBr" or anything like it is plausible.

I don't assume that the final poll was correct; if indeed the raw data were wrong, there is no a priori reason to expect that the final weightings would yield true results for anything else. But I will come back to the Bush/Gore question in 3).

2) As I think you know, I see real confidentiality issues in a full release of the raw data. We can call for the release of multivariate results, on the assumption that E/M will release accurate results, just as we generally assume that the results in the January report were accurate even if the interpretations were not. We can also call for the release of "blurred" data like ESI used in Ohio, but I have no way of assessing whether "blurring" yields usable data -- for now, I'm dependent on Fritz Scheuren's belief that it does.

3) Again, if we assume that the raw data were wrong, there is no particular reason to assume that the Bush/Gore results should be accurate. However, in the 2002 National Election Study (which interviewed both voters and non-voters), 51.6% of respondents reported having voted for George Bush in 2000, and 44.3% reported voting for Al Gore, among those who reported a 2000 presidential vote. I therefore invite you to explain why you think that the 2004 exit poll result proves fraud but the 2002 NES result proves nothing in particular.

[TruthIsAll]

74. Forget polling. Deal with the facts.

Answer ONE simple question.

Would you calculate the MAXIMUM number of Bush 2000 voters who could have voted in 2004?

Right here.
Right now.

[OnTheOtherHand]

84. "Forget polling"?! That will save some time, yup. n/t

[autorank]

89. Lets wait on confidentiality. What about the 43/37 question. That's

the issue. Obviously, TIA's not saying "forget polling" altogether. You are, IMHO, using this petit rhetorical trick to avoid the statistical question at hand.

We can pursue confidentiality later, althought it's important.

The statistical question is essential and I'd like to see it hashed out.

[Internut]

101. Since the 43/37 numbers came from the poll,

it is kinda hard to "forget polling" when answering it, wouldn't you say?

[TruthIsAll]

106. You said 43/37 came from the poll; I said the weights are an artifice to

to match the vote.

Are you aware, Internut, that Mitofsky himself admits to having done just that?

[Internut]

107. Do you have any evidence whatsoever

that the "43/37" numbers were the "weights" vs. resulting from other weights being applied to the polling numbers?

Since, as I have explained and showed to you several times, the "past voting behavior" results in polls are notoriously inaccurate, and Mitofsky and other pollsters know that fact, you would think it would be extremely irresponsible of them to use those results, even partly, as weights for the poll adjustments, wouldn't you?

And it is not I who say that the "43/37" come from the poll. It is the poll that says that "43/37" as well as all those other numbers come from the poll. See the poll: http://www.cnn.com/ELECTION/2004/pages/results/states/US/P/00/epolls.0.html

If those numbers did not come from the poll, please explain where you obtained them.

[autorank]

77. Full release of the "raw data" is both POSSIBLE & MANDATORY!

You said, "As I think you know, I see real confidentiality issues in a full release of the raw data."

This is a fundamental issue that your side keeps harping on. One of you said, well the networks and Roper own the data, therefore Mitofsky can't release it. Then I saw something over at the Election Science Institute indicating that indeed Mitofsky did release some data to them, maybe not the "raw data" but some.

The more important issue here is the falss criteria you set us as an excuse for NOT releasing the necessary data. I responded to you on this previously and you didn't reply. So let me try again:

--Raw data with names and "identifying" factors does not need to be released. The data elements that are essential don't come near names. Who cares what their names are? Nobody. The real information is demographic, candidate voted for, etc.
--The data released needs to be verfied as real. Let's assume that Mitofsky et al can verify that the data has not been altered in any way from the point it was collected, the release of the data can happen in a simple two step process: (a) people who are somehow designated as responsible parties can reveiw the data to assure (as much as possible) that it is what it is and (b) the elements of the raw data that are (i) essential and (ii) reveal no confidential identifying information can the be released as a data set.

In this process, the American public can be satisfied since a major source of justified suspicion in the 2004 election can be resolved and there will no longer be a need to argue over this particular point.

The argument on confidentiality is, you must admit, a PURE RED HERRING, a TRANSPARENT STRAW MAN, a FALSE ARGUMENT used to keep the American people in the dark about their fundamental right - the right to vote and know that the vote is counted fairly.

There are precedents forthis all over like hospital audits, financial audits of major institutions, etc. where the presimption is that there are ways to review "confidential data" as in the case of health data that protect confidentiality while serving a public need to know that institutions are operating safely and honestly.

So please put this argument to rest. It's a built in deal killer that will keep the debate going on forever and you know it.

[OnTheOtherHand]

82. confidentiality

I apologize if I dropped a response from you -- it wasn't intentional. Or I may not have realized that the post was intended to be substantive.

No, I do not admit -- or agree -- that the issue is a pure red herring. And I find the implication that I somehow have a vested interest in not having access to data bizarre at best.

Datasets with demographic information have already been released, as was the case also for past exit polls. Simply culling the demographic data now would not protect confidentiality.

(The ownership issue is more tractable, although I assume it is true that Mitofsky doesn't have carte blanche to do whatever he pleases. I don't know how that was handled in the ESI/Ohio case.)

I'm predisposed to believe that "blurring" as ESI did (or rather, I believe, had E/M do) in Ohio can work. But the technical papers haven't been released, and I don't know. Also, ESI reported that it found no evidence of fraud in the Ohio exit polls -- but again, I have no way of assessing the analysis at this time.

That result suggests to me that releasing the data wouldn't actually accomplish anything. A handful of people would be convinced by the new data, one way or another; the rest of the world would go on believing whatever they were predisposed to believe.

[autorank]

90. We'll deal with this later. How about TIA's question and 43/37.

No appologies necessary.

[OnTheOtherHand]

91. Please read my post #66, especially point 3. n/t

[autorank]

93. Why not deal with it here. Narrow things down, have a real record.

These references back and forth are tedious and difficult to follow, not to mention a distraction.

I truly want to see the debate carried forward here in real time.

Seriously, just "cut and paste" if that's what you want to do but "see my post" is not dealing with the question at hand.

[OnTheOtherHand]

95. huh?

If I post some information, and the person to whom I posted it (TIA) shows no signs of having read it, and you ask the same question, and I direct you to the post, how am I not addressing the question or not carrying the debate forward in real time?

'rank, you have about another week before I leave for a month, and then if you wish you can celebrate my absence. But in the meantime, if you actually want to know anything from me, you will have to make better use of my time. For better or for worse, I frankly don't have time to be lectured on the fine points of DU etiquette as you interpret them.

[autorank]

96. "O", I celebrate your presence. "T" why don't you re-ask the question.

Then we'll have a chance for some real "debate." That's your purpose here isn't it, to help the progressive/Democratic cause by cleaning up any "fuzzy" or illogical thinking.

The request was made in all sincerity.

I want to see the outcome.

Thanks.

[OnTheOtherHand]

97. or try a response to #66 other than "forget polling" n/t

[autorank]

98. Any one who cares to read TIA's answer in context knows that

he didn't mean "forget polling." He's occupied with polling results daily. He wanted to direct you back to his question.

He said, and I quote:

"Answer ONE simple question.

Would you calculate the MAXIMUM number of Bush 2000 voters who could have voted in 2004?

Right here.
Right now." TruthIsAll Mon Jun-20-05 03:48 PM

That was the last question asked in the dialog. Recalling my experience in interscholastic debate, a debate revolves around arguments which start and follow a consistent thread and also question/answer periods (by far the most fun part of that activity) where opponents ask each other questions and get answers. He asked you a question, "ONE simiple question." You won't answer it. He's waiting I suspect to see if that happens then the dialog will continue. But you can answer the question so why don't you?

[OnTheOtherHand]

99. Did he respond to my point #3? I think not. n/t

[autorank]

100. The debate and thread are unbalanced. If you answer his question,

he'll surely show up and the debate will continue. He's that kind of guy and so are you. He asked the last question, I repeated it for you. That's where you are.

Let's restore balance, in a symmetrical sense, to the thread.

[OnTheOtherHand]

102. with due (and sincere) respect, I disagree

I sense that you really believe what you are saying. It makes no sense to me at all.

AFAIK I have done a lot more work at TIA's behest than he has at mine. I waded through his optimizer several times, made a few helpful points, I give him props for acknowledging them.

But if he isn't willing to think about the similarity between the 43/37 result from the 2004 exit poll that exercises him so much, and the 7-point gap in the 2002 National Election Study (not an exit poll, by the way, but a very carefully conducted in-person survey) -- if he insists on ignoring it -- then I see no reason to jump through his hoops.

[autorank]

103. I'm sorry, I didn't mean to trouble you. Signing off for now...

Contact the DNC and Tell Them to PREVENT Election Fraud

NEW LEADERS FOR A NEW DEMOCRATIC PARTY

[TruthIsAll]

111. Lets get right to it. The 43/37 was NOT a result of polling...

Are you aware that Mitofsky has always matched to the final vote?
He did this time, as well.

Even though in the process of adjusting the weights to achieve the match, he had to use totally IMPOSSIBLE numbers.

So 43/37 was an artifice to match the Bush vote.

Or will you still persist in obfuscation (you are very good at that, by the way) by saying that it was just a "polling" result.

EVEN IF IT WAS A POLLING RESULT, IT FLIES IN THE FACE OF REALITY, THE REALITY BEING THE ACTUAL NUMBER OF BUSH/GORE 2000 VOTERS (STILL ALIVE) WHO COULD HAVE VOTED IN 2004.

OTOH, let's face it. All your prior verbiage falls apart when you cannot offer a viable explanation for this fact. This is NOT just another an anomaly. It is a MATHEMATICAL FACT WHICH YOU CAN'T REFUTE.
THERE IS A ZERO MOE IN THIS FACT.

AND SO, TOO, YOUR CREDIBILITY IS ZERO IF YOU REFUSE TO CONFRONT IT.

YOUR AVOIDANCE OF THE ISSUE IS NO DIFFERENT THAN IF I ASKED YOU TO STIPULATE 1+1=2 AND YOU FAILED TO STIPULATE THAT AS A FACT.

WILL YOU STIPULATE THAT 43/37 WAS AN ABSOLUTE IMPOSSIBILITY?

RIGHT HERE?
RIGHT NOW?

CAN WE MOVE OFF THIS DIME AND ON TO BIGGER AND BETTER THINGS?
OR ARE YOU THROWING IN THE TOWEL NOW?

[anaxarchos]

108. Let me try... (point #3 that is)...

Here's the history of the NES presidential response from 1948 to 2000:

http://www.umich.edu/%7Enes/nesguide/toptable/tab9a_1.htm

All in all, it's not too shabby for a post-election telephone poll with a 36 minute interview and a rather complex sampling method.

If your argument is that the 2002 response illustrates a generic bandwagon effect, I sure don't see it.

More, the off years are a little screwy on the partisanship questions, 2002 in particular.

But, hey, don't take my word for it. Look at some of the other partisanship categories...

http://www.umich.edu/%7Enes/nesguide/gd-index.htm#2

Now, your turn...

[Internut]

109. You do realize that OTOH's question was about

polling for the responder's past (4 years ago) vote, while the link you gave in response asks for the responder's current (a few minutes (days?) ago) vote. Apples and oranges.

[anaxarchos]

110. Nope....

I wasn't paying attention. This particular rathole is even more frustrating than the others... But a bandwagon is more likely the closer it is to the election... ain't it? I think I'll just stand on it.

It will allow me to bond with a lot of the other "talk" on this thread.

Lemme see... how do I lead with this crock o' shit?

[anaxarchos]

116. Lemme try this again... awake this time.

The answer is obvious... 2002 is a Mid Term election in which the Republicans out pointed the Democrats by 6 to 8 points (depending on how you calculate it). The sample is different. That's kinda what I was alluding to above when I pointed to the partisanship spikes for mid-terms...

Take a look at the 1996 NES. The opposite effect is discernible (Perot doubles his vote!) for precisely the same reasons.

BTW, NES post-election is days-to-weeks, not hours-to-days... Thus the "bandwagon" discussion is pertinent.

You just weren't arguing it :-)

[LightningFlash]

117. "Lemme" try this again....Excellent verbage!!! n/t

:yourock:

:hi:

[Internut]

118. You can't give a "days to weeks" historical voting question

poll and compare it to the "whom did you vote for in 2000" asked in 2004. You have to compare apples to apples.

It is a well known fact that people routinely misreport their historical voting behavior. That's my point. TIA for some reason thinks that Mitofsky invented the 43/37 split and used it to weigh the other variables in the final exit poll. That is absurd. No pollster in his right mind would use the notoriously unreliable historical voting question to weigh anything else in the poll.

[anaxarchos]

120. Now it's your turn to be asleep...

Read what I wrote again... (I am still sheepish and therefore uncharacteristically generous because of last night's mistake).

[Internut]

122. I did, I showed you the poll that asked, years after the fact,

whom people voted for, and it showed that people routinely lie to pollsters when answering that question.

To answer that, you show me the poll that asks, days or weeks after the vote, whom people voted for. That is not the same thing.

If you'd like other evidence of such misreporting, here is a paper by UK's ICM Research

http://www.icmresearch.co.uk/reviews/2002/failure-of-the-polls-1997.htm

skip to the "Past Vote: Faulty Memories or Faulty Samples?" section.

[OnTheOtherHand]

121. umm, not quite that simple

What do you mean when you say "the sample is different"? The NES is a national survey, not an exit poll, and the 2000 vote question was asked in the pre-election wave. There seems to be no difference between reported 2002 voters and non-voters in their response.

Similar result in the 2004 NES, again in the pre-election wave, although the gap is only 4 points. (These are both unweighted results, by the way, just in case weighting is of the devil. The weighted gaps are wider.)

[anaxarchos]

123. Actually, I think it is...

(First, an observation without comment: In this context there is a big difference between 4 points and 7 points)

I'm not saying anything earth shattering. If we create a series of NES General Election / Mid-Term Election pairs looking backward from 2004, we find that on the "How did you vote in the last general election" question, we see a pattern:

1) The variance from actual vote counts that you note in your "question #3" is not uncommon but is also not universal. As an example, for '96, it is as much as 9 points, for '92 only 3 points, and for '88 it is nearly dead nuts on, etc...

2) When the skew above exists, it is most often worse in the Mid-Term poll than in the GE poll, i.e. responders are "more accurate" 4 years after the election than they are after 2.

3) When the skew above exists, it does not always favor the winner. For '96, the variance is at the expense of the Republican Loser but favors Perot more than the winner. For '92, it favors the winner but at Perot's expense, not that of the Republican loser.

What does this mean? I dunno... but the NES results don't really reflect the consistency your argument implies nor does it support Mr. Nut's "notoriously unreliable". My look wasn't exactly exhaustive but...

So the next thing worth looking at is the NES itself and particularly the MT vs GE results to see if we can account for part of the above. Luckily, we have the guidebook for the CDF:

http://www.umich.edu/~nes/nesguide/nesguide.htm

Take a look at the vote count, political involvement, and partisanship sections. I dunno again, but the MT samples look more "politically active", to me. Just one example: The GE polls consistently exaggerate turnout (not unexpected) as do the MTs but the MT exaggeration is both less consistent and more radical.

If I were ambitious, I would see which way the Perot voters broke in the mid-terms and invent an eMr bias theory ("exuberant Midtermelectionparticipants response" - It's a good year for that sort of thing). Since I am not, it is enough to say that there is at least some possibility that at least part of your "question #3" could result from factors intrinsic to the NES MTs. If I take a step back, it is easy to be sympathetic to NES' problem.

Bottom Line: You probably can't hang your hat on your Q3, let alone the sombrero that Mr. Nut wants to put up there...

[OnTheOtherHand]

124. you think your post offers a "simple" argument?

(In response to your observation without comment, I offer my own: my reporting of unweighted results was conservative, if not downright misleading. Those weights aren't constructed to try to match electoral results -- at least, the PIs don't seem to think so! -- so there is no obvious reason not to apply them.)

Obviously I am not in a position to pound the table and insist that 43/37 is spot on, although I think it might be.

However, I don't think Internut and I were the ones pounding the table. TIA has insisted, if I understand him rightly, that this six-point gap in the 2004 exit poll offers positive mathematical proof of fraud, and that gaps of similar magnitude in NES surveys do not even merit acknowledgement, much less explanation -- "forget polling." I do not regard his argument in this regard as serious, or even as an argument.

[Internut]

127. I don't know about your definition of "notoriously unreliable" -

but when I see the same statistic sometimes 9 points off, sometimes 3 points off, sometimes dead on - that pretty much is "notoriously unreliable".

[LightningFlash]

134. Logic.....We treat it like its a bad thing. :) n/t

[TruthIsAll]

12. OTOH, the aggregate response rate was 53% - not 56%

"That is a large aggregate difference in response rate.
It is an inevitable consequence of simulating a set of 40
precincts with Mean WPE = -10, Median WPE = -5.8 and an
overall response rate of 56%. A small number of precincts with
very large negative WPE are required".

So the input error was (56-53)/53 = 5.7%. 
What effect does this have on the standard errors you cite?

At the reported average 53% response rate:

Categ.	HighB	Bush	Even	Kerry	HighK		Total/Avg	Std dev
Prcts	40	415	540	165	90		1250	
Kerry	16.1%	41.0%	47.1%	63.7%	81.6%		48.77%	
								
ALPHA	1.63	1.15	1.18	1.09	1.00		1.16	0.22
ln(a)	0.49	0.14	0.17	0.09	0.00		0.15	0.17
AvgDev	40%	-1%	2%	-6%	-14%		0%	18.79%

Weighted average alpha = 1.16.
alpha Std deviation = .22

High Bush precinct alpha = 1.63
The difference 0.47 = 1.63-1.16  is more than DOUBLE the alpha
standard deviation of 0.22.
Very Significant.

High Bush precinct ln(alpha) = 0.49
The difference .32 = 0.49-.17 is almost DOUBLE the ln alpha
standard deviation of .17.
Very Significant.

What effect does this have on the standard errors you cite?

[OnTheOtherHand]

14. sorry, no sale

First of all, you need to take the "overall response rate" in context (and, to your credit, you did quote the context). O'Dell was referring to the 40 high-Bush precincts, and the reported completion rate there was indeed 56%. There are probably some mistakes in the paper, but that doesn't seem to be one of them. Even if it were, I doubt it would appreciably affect the standard errors, although I welcome empirical evaluation of the issue.

The average (mean) ln alpha will not equal the ln of the mean alpha. For instance, imagine 2 precincts:

alpha = 2.0, ln alpha = 0.693
alpha = 0.5, ln alpha = -0.693

mean alpha = 1.25 (the ln of 1.25 is 0.223)
mean ln alpha = 0.0

Since the point of ln is precisely to remove the skew in alpha, it doesn't do much good to use ln after calculating the mean.

So, your optimizer analysis has no effect on the actual standard errors or t scores.

[TruthIsAll]

16. The use of the term "overall" response rate...

implies the total weighted average.

But yes, the optimizer confirms that high-Bush response = 56.6%

[TruthIsAll]

18. ln mean alpha =.148 vs. mean ln alpha =.212

ALPHA	2.04	1.15	1.14	1.08	1	1.16
ln(a)	0.71	0.14	0.13	0.08	0	0.148
AvgDev	76%	-1%	-1%	-7%	-14%	0%
						
ln mean alpha= .148=ln (2.04+1.15+1.14+1.08+1)/5) = ln( 1.16)
and
mean ln (alpha) = 0.212 = (.71+.14+.13+.08+0)/5			

[TruthIsAll]

19. From original post: wtd alpha =1.155; ln a = .144; mean (ln a) = .162

ALPHA	1.52	1.15	1.18	1.09	1	1.155
ln(a)	0.42	0.14	0.16	0.09	0	0.144
AvgDev	76%	-1%	-1%	-7%	-14%	0%
						

Ln (wtd alpha) = ln(1.155)					
0.144						
						
mean [ln (alpha)] = (.42+.14+.16+.09+0)/5
0.162						

[OnTheOtherHand]

21. these are optimizer results?

Note that the O'Dell simulator allows different precincts in a given partisanship range to have different alphas. If you aren't reflecting upon that variation from precinct to precinct, I don't see how you can refute the arguments in #8 above.

I will not always be able to respond to numbers without arguments. You may think that they speak for themselves, but honest people might reasonably disagree about what they say.

[TruthIsAll]

22. The numbers do speak for themselves....

No one says that you must respond...but I would think that you would want to.

I am looking at the limited average partisanship data as provided by Mitofsky.

Don't you find it interesting that the model's output mirrors those of the simulator - and provide even MORE insight?

As a quantitative analyst, you should focus on the model - forget everything else. If there is just ONE model input parameter or result which you disagree with please point it out.

You did once before - and I incorporated the fix..

Is there anything at all in the model that you disagree with?

[OnTheOtherHand]

25. the lack of individual precincts, perhaps?

Honestly, I'm not trying to duck anything. As far as I can tell, you, I, O'Dell, Baiman, and anyone else who has looked substantially agree about the general magnitude of the mean alpha.

I don't think your model yields a standard error of alpha or ln alpha within a partisanship category. So I don't see how it engages O'Dell's results at all.

[TruthIsAll]

27. OTOH, let's frame the discussion this way...

What does the fact that the optimization returns similar results as the simulation (and MUCH more information), WITHOUT UTILIZING INDIVIDUAL PRECINCT DATA, tell you?

[OnTheOtherHand]

29. that it conveys no information about variance btw. precincts. n/t

[TruthIsAll]

30. OTOH, with all due respect, that is not an answer.

Of course we both know that precinct variance is not an optimizer driver. But OTOH, the model returns values very close to those of the simulation, for which variance is obviously a major driver.

So if I may repeat myself: What does that tell you?

[OnTheOtherHand]

35. ok...

it tells me that you are probably quite close to the median alpha for each partisanship category, but that you have no evidence about the variance of alpha -- and you therefore have no basis for statistical inference.

[LightningFlash]

39. And that's not evidence for the opposite--You can't guess variance.

There is no variance because there's no precinct level data released including variance. Therefore there is no variance, your point is refuted.

You have not come up with the common sense that your argument "refutes" itself. So because it refutes itself, stop beating the strawman then. There is no linked evidence for "EDR(Exuberant Democratic Response)". and therefore EDR along with RBR, is now nearly dead.

[OnTheOtherHand]

42. "There is no variance"

1) The mean absolute WPEs were released back in January

2) The scatterplots depict the variance

3) Even if we knew nothing about the variance, that wouldn't refute my point, unless you have in mind some inferential statistical test that doesn't require knowledge of the variance

[TruthIsAll]

45. The scatterplots depict squat.

They mean zilch.
And you know it.
They tell us NOTHING.
They tell us NOTHING about INDIVIDUAL precincts.
We cannot VERIFY the data source.
It is presented as an article of faith.

They could have been created by a random-number generator for all we know.

You are spending your valuable time connecting the dots.
You are drawing inferences from points which have NO meaning.
They have no back-up; they have no context.

As a politcal science prof, you should be DEMANDING the REAL data.
You should NOT be relying on this suspect art.
It makes your argument suspect.

But since you apparently have an agenda to keep the facts from view, you cannot do otherwise.

You have too much invested in the position you have staked out.

[OnTheOtherHand]

47. think about this

If they faked the scatterplot, why not fake the January report? why not fake the "real data"?

The rest of your post apparently offers no substance to which to respond.

[TruthIsAll]

57. For the same reason they will not release the data... n/t

.

[autorank]

94. They won't release the data because they are afraid of the TRUTH

Who goes down (a) if the data is released and (b) it shows election fraud? Well, the Networks for one, they commissioned and paid for the NEPs. Also Roper. To a lesser degree EM since they're the small fry (significant as they are as a "knowledge company") compared to the networks.

The news media is rated just above "car salesmen" in a recent public opinion survey on "respected" professions. I think they'd be rated below "Known thieves and liars" if this one blew up in their fact.

[LightningFlash]

119. And of course, Tom Feeney will say..."Lemme" see what I can do...

....So until they release the data, I guess we're at a crossroads!!!!

:wow:

:hi:

[LightningFlash]

135. Bringing back the truth.