US Count Votes: new version of Working Paper (Updated June 26, 2005)

[eomer]

US Count Votes has posted a significant revision of their Working Paper, "Patterns of Exit Poll Discrepancies".

Abstract

The exit polling firm Edison/Mitofsky¹ (E/M), and USCV, agree that the historically unprecedented
discrepancy between the exit polls and the reported vote count for the 2004 U.S. Presidential
election cannot be a result of random sampling error.² This leaves either exit poll error or vote
miscounts as the only two explanations for the exit poll discrepancy.

E/M claimed that the exit poll discrepancy is exclusively a result of “within precinct error” (WPE)³,
and that the entire WPE observed in 2004 could be explained by a hypothetical exit poll completion
rate of 56% among Kerry voters and 50% among Bush voters (herein referred to as “the E/M
hypothetical”).⁴ The E/M hypothetical was widely interpreted by the media and by USCV as a claim
that the 2004 exit poll discrepancy was caused by a pervasive, and on average uniform, shortfall in
Bush voter exit poll response relative to Kerry voter exit poll response that was dubbed the
“reluctant Bush response” (rBr) hypothesis.⁵

A recent clarification by E/M indicates that the “E/M hypothetical” should be interpreted as referring
to hypothetical average (rather than constant average) partisan exit poll response rates.⁶ In this
interpretation, average precinct partisan response rates may vary widely by reported precinct vote
shares, yet all of the reported WPE could be explained by partisan response rates whose average
across the sample is K=0.56 and B=.50.⁷ This interpretation of the “E/M hypothetical” does not
depend on the “rBr hypothesis” of constant average partisan response rates which was shown by
USCV to be inconsistent with the pattern of the exit poll discrepancy.⁸

USCV's analysis shows that even if the “E/M hypothetical,” is interpreted as referring to average,
rather than constant average, partisan exit poll response rates, it is inconsistent with the reported
WPE data. There is no configuration of partisan response rates, however varying across precinct
partisanship categories, with overall averages of K=0.56 and B=0.5 that can produce the reported
values of the actual E/M exit poll data for mean WPE and median WPE.⁹ Thus, neither a "reluctant
Bush responder" (rBr) hypothesis interpretation nor an overall average interpretation of the “E/M
hypothetical” is consistent with the WPE pattern shown by the Edison/Mitofsky exit polling data.

-snip-

http://uscountvotes.org/ucvAnalysis/US/exit-polls/USCV_exit_poll_analysis.pdf

Note: sorry if this is a dupe. I was surprised to find an updated version dated almost a week ago and no corresponding thread (that I could find) in this forum. Maybe the document was posted more recently than the date on it would suggest?

[TruthIsAll]

1. The optimizer confirms: Min alpha 1.1545, R=.535

EXIT POLL RESPONSE OPTIMIZATION MODEL						
7/2/05 10:20 AM						
						
OBJECTIVE:						
Determine minimum weighted K/B (alpha) required to satisfy
partisan precinct WPE's and response rates						
						
Precinct Variable Input (Min, Max) Constraints:
1- Response rates: aggregate R:.53HB,.55B,.52NP,.55K,.56HK
2- Kerry 2-party vote:avg 10%,30%,50%,70%,90% 
3- Alpha (K/B): minimum (1.1545) 
4- WPE:-10.0%HB, -6.1%B, -8.5%NP, -5.9%K,0.3%HK

VOTE	Mil.	Pct (Input)				
Kerry 	59.027	48.76%				
Bush	62.029	51.24%				
Total	121.056					

POLL	Mil.	Pct				
Kerry 	63.127	52.15%				
Bush	57.929	47.85%				
						
Bush-recorded	51.24%					
						
EXIT POLL RESPONSE (input)						
Aggregate	53.5%	R				
Kerry/Bush	1.1545	alpha				
This is minimum alpha required to satisfy 						
the WPE/partisanship constraints						

PROBABILITY						
Bush vote gain	1.59E-11					
 1 in	62,953,509,332					
Kerry > 50% vote	100.0%					
 1 in	1					
						
PARTISAN ALPHA						
Kerry	1.062					
Bush	1.184					
						
RESPONSE INPUT CONSTRAINTS						
1250	Strong Bush		Strong Kerry			
Prcts	40	415	540	165	90	
						
KERRY WIN%						
Min	0%	20%	40%	60%	80%	
Max	20%	40%	60%	80%	100%	
						
RESPONSE						
Min	55.9%	54.9%	51.9%	54.9%	52.9%	
Max	56.1%	55.1%	52.1%	55.1%	53.1%	
						
ALPHA (K/B)						
Min	0.5	0.5	0.5	0.5	0.5	
Max	10.0	10.0	10.0	10.0	10.0	
						
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				
						
		Vote Percentages	Votes (millions)	
	Poll	Vote	Diff	Poll	Vote	Diff 
Kerry	52.15%	48.76%	-3.39%	63.127	59.027	-4.100
Bush	47.85%	51.24%	3.39%	57.929	62.029	4.100
Diff	4.29%	-2.48%	-6.77%	5.199	-3.002	-8.201
						
Pship	HighB	Bush	Even	Kerry	HighK	Total/Avg
Prct	40	415	540	165	90	1250
Vote	3.874	40.191	52.296	15.979	8.716	121.056
Pct	3.2%	33.2%	43.2%	13.2%	7.2%	100.0%
						
RESP.	56.1%	54.9%	51.9%	54.9%	52.9%	53.5%
Dev	2.6%	1.4%	-1.6%	1.4%	-0.6%	0.0%
						
ALPHA						
K/B	1.505	1.153	1.174	1.098	0.996	1.155
Dev	30.4%	-0.2%	1.6%	-4.9%	-13.7%	0.0%
ln a	0.409	0.142	0.160	0.094	-0.004	0.144

VOTE						
Kerry	0.77	16.06	25.62	9.60	6.98	59.027
Pct	19.8%	40.0%	49.0%	60.1%	80.1%	48.76%
						
Bush	3.11	24.13	26.68	6.38	1.74	62.029
Pct	80.2%	60.0%	51.0%	39.9%	19.9%	51.24%
						
RESPONDERS						
Kerry	0.96	17.29	27.84	10.07	6.97	63.127
Pct	24.8%	43.0%	53.2%	63.0%	79.9%	52.15%
						
Bush	2.91	22.90	24.46	5.91	1.75	57.929
Pct	75.2%	57.0%	46.8%	37.0%	20.1%	47.85%
						
REFUSERS						
Kerry	13.4%	36.3%	44.4%	56.5%	80.2%	44.88%
Bush	86.6%	63.7%	55.6%	43.5%	19.8%	55.12%

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

[Bill Bored]

2. WTF is a "constant average" anyway?

I may be dense, but this seems to be almost a contradiction in terms or superfluous language. If there is a constant, there is no need to compute an average, because it's the SAME EVERYWHERE.

For example, there is no need to compute the AVERAGE speed of light in free space; it's a CONSTANT!

I don't even know who first used this term, nor do I care. It's not about who's right or wrong, it's about understanding WTF they're trying to say.

If the numbers vary, then you can take an average, median, standard dev., etc. But if there is a constant, then there is no variance. I don't understand what a "constant average" is!

So what's the definition of a constant average?

I don't mean to digress, I just want this explained in a logical way so it won't further confuse things.

[Stevepol]

3. I think the whole point is that it's not necessary to worry

out little heads about all these technical terms and just move on. Trust us, it was the exit polls and not fraud. The voting machines couldn't have caused this discrepancy. No way.

[Time for change]

4. If that's the only problem you're having in reading this

you're doing a lot better than I am with it.

Can anyone translate this into the kind of English that some reasonable number of people would be able to understand?

[Bill Bored]

6. Seems like they are saying that if 56% of Bush voters

completed the exit polls, and 50% of Kerry voters completed the exit polls, this would not result in the mean and median WPEs that actually occurred in the data. I would assume, the rest of the paper is devoted to proving this.

[TruthIsAll]

9. That is correct; K/B = 56/50 as stated by E-M is not feasible (possible)

Kerry/Bush= 1.12=56/50 as a weighted average response rate across ALL partisanship groupings is IMPOSSIBLE (infeasible). There is no way the partisanship WPE constraints can fall out of that.

In other words, given that:
1) Bush got 51.24% of the 2-party vote,
2) WPE for each of the five partisanship groups is given by E-M,
3) 53.5% is the total weighted average response rate,
4) Response rates by partisanship group vary from .53 (High Bush), to .55,.52,.55 to .56 (High Kerry)

then the E-M conclusion that K/B = 56/50 is impossible.

The optimizer converges to a FEASIBLE solution ONLY if ALL constraints are satisfied.

The K/B alpha ratio must be 1.1545 at MINIMUM, which means that for rBr to be even considered an implausible explanation for the discrepancies, then response bias must be reflected by a MINIMUM 57.73/50 ratio of Kerry responders to Bush responders.

But that's not all. There has been no explanation of why there is no measurable uniform bias across all precincts. They vary wildly.

On the other hand, the numbers tell us where the fraud was concentrated - in partisan Bush and competitive precincts, NOT in Kerry strongholds.

[TruthIsAll]

10. This is what happens if you assume a uniform K/B = 1.154

Check out the effects on partisan WPE's AND the Kerry poll
result (52.53%) required for it to happen.

EXIT POLL RESPONSE OPTIMIZER						
7/2/05 6:09 PM						
						
OBJECTIVE:						
Determine effects of uniform weighted K/B (alpha=1.154)			
on partisan precinct WPE's and response rates	
						
Precinct Variable Input Range (Min, Max) Constraints: 		
1- Response rates: aggregate R:.53HB,.55B,.52NP,.55K,.56HK    
       				
2- Kerry 20% quintile 2-party vote:avg 10%,30%,50%,70%		
3- Alpha (K/B): test for minimum (1.1545) 				
4- WPE:-10.0%HB, -6.1%B, -8.5%NP, -5.9%K,0.3%HK
						
VOTE	Mil.	Pct (Input)				
Kerry 	59.027	48.76%				
Bush	62.029	51.24%				
Total	121.056					

POLL	Mil.	Pct				
Kerry 	63.586	52.53%				
Bush	57.470	47.47%				
						
Bush-recorded	51.24%					
						
EXIT POLL RESPONSE (input)						
Aggregate	53.5%	R				
Kerry/Bush	1.1545	alpha				
This is minimum alpha required to satisfy 						
the WPE/partisanship constraints						

PROBABILITY						
Bush vote gain	7.87E-14					
 1 in	12,704,089,216,842					
Kerry > 50% vote	100.0%					
 1 in	1					
						
PARTISAN ALPHA						
Kerry	1.154					
Bush	1.154					
						
RESPONSE INPUT CONSTRAINTS						
1250	Strong Bush		Strong Kerry			
Prcts	40	415	540	165	90	
						
KERRY WIN%						
Min	0%	20%	40%	60%	80%	
Max	20%	40%	60%	80%	100%	
						
RESPONSE						
Min	55.9%	54.9%	51.9%	54.9%	52.9%	
Max	56.1%	55.1%	52.1%	55.1%	53.1%	
						
ALPHA (K/B)						
Min	1.153	1.153	1.153	1.153	1.153	
Max	1.155	1.155	1.155	1.155	1.155	
						
WPE						
E-M	-10.0%	-6.1%	-8.5%	-5.9%	0.3%	
Min	-50.0%	-50.0%	-50.0%	-50.0%	-50.0%	
Max	50.0%	50.0%	50.0%	50.0%	50.0%	
						
						
		OPTIMIZER OUTPUT SUMMARY				
						
	Vote Pct		Vote (millions)		
	Poll	Vote	Diff	Poll	Vote	Diff 
Kerry	52.53%	48.76%	-3.77%	63.586	59.027	-4.559
Bush	47.47%	51.24%	3.77%	57.470	62.029	4.559
Diff	5.05%	-2.48%	-7.53%	6.116	-3.002	-9.119
						
Pship	HighB	Bush	Even	Kerry	HighK	Total/Avg
Prcts	40	415	540	165	90	1250
Votes	3.874	40.191	52.296	15.979	8.716	121.056
Pct	3.2%	33.2%	43.2%	13.2%	7.2%	100.0%
						
RESP.	56.1%	54.9%	51.9%	54.9%	52.9%	53.5%
Dev	2.6%	1.4%	-1.6%	1.4%	-0.6%	0.0%
						
ALPHA						
K/B	1.154	1.155	1.155	1.154	1.154	1.155
Dev	0.0%	0.0%	0.0%	0.0%	0.0%	0.0%
ln a	0.143	0.144	0.144	0.143	0.143	0.144

VOTE						
Kerry	0.77	16.06	25.62	9.60	6.98	59.027
Pct	19.8%	40.0%	49.0%	60.1%	80.1%	48.76%
						
Bush	3.11	24.13	26.68	6.38	1.74	62.029
Pct	80.2%	60.0%	51.0%	39.9%	19.9%	51.24%
						
RESPONDERS						
Kerry	0.83	17.31	27.60	10.34	7.52	63.586
Pct	21.3%	43.1%	52.8%	64.7%	86.3%	52.53%
						
Bush	3.05	22.88	24.70	5.64	1.20	57.470
Pct	78.7%	56.9%	47.2%	35.3%	13.7%	47.47%
						
REFUSERS						
Kerry	17.8%	36.2%	44.9%	54.4%	73.2%	44.44%
Bush	82.2%	63.8%	55.1%	45.6%	26.8%	55.56%

VOTE DEVIATION						
Kerry	-0.06	-1.24	-1.98	-0.74	-0.54	-4.56
						
WPE						
Calc	-3.0%	-6.2%	-7.6%	-9.3%	-12.3%	-7.532%
E-M	-10.0%	-6.1%	-8.5%	-5.9%	0.3%	-6.774%
Diff	6.95%	-0.1%	0.9%	-3.4%	-12.6%	-0.8%

[TruthIsAll]

12. FOUR GRAPHS WHICH SHOW THE EFFECT OF HYPOTHETICAL UNIFORM K/B=1.154

These are the IMPLAUSIBLE effects of uniform bias (K/B =1.154) on:

Partisan alpha


WPE


Partisan response


Bush percent of Refusers

[TruthIsAll]

17. GRAPH: Sensitivity of Kerry Pct and aggregate WPE to K/B (alpha)

Note: Only alpha > 1.154 is feasible, given -6.77% aggregate WPE.

[Bill Bored]

22. TIA, Does your optimizer

come up with the same mean, median and absolute mean WPEs as the actual data (or are they entered as constants)? I thought this was one of Bruce (but not necessarily Wally) O'Dell's criticisms of your work, whether justified or not. (I'm sure Wally would criticize it too since he has his own optimizers built into his e-votin' machines.) ;)

I agree (I think) that fraud was unlikely in Kerry strongholds, more likely in moderate Kerry and competitive precincts and some moderate Bush, but I don't think there were enough Bush strongholds to make a determination and there weren't enough of them to change the outcome of the election either. I think this part of the analysis (Bush strongholds) is therefore a waste of time.

That said, if you take a Kerry stronghold and REVERSE it, you would get a Bush stronghold, with a hell of a lot of WPE!

Furthermore, if, as I believe, it is possible (and easier) to change the results in every precinct in a given county, which would include the whole range of partisanship, there is no reason to suspect any particular category, which would explain the high WPEs in both directions and the high abs mean WPE. Some Bush-loving precincts could have just been victims of "collateral damage" within a large reversal of results for an entire pro-Kerry county, either on Election Day, or during Early Voting periods (partial reversal) where the chain of custody of ballots and machines is tenuous at best.

In other words, it's a lot easier to commit the fraud more generally than it is to cherry pick individual precincts, so I don't think singling out specific categories of precincts is particularly useful, except in Ohio where ballot order rotations in collocated precincts frequently made it possible to switch Bush and Kerry votes.

[TruthIsAll]

24. BB, a response..

You:
TIA, Does your optimizer come up with the same mean, median and absolute mean WPEs as the actual data (or are they entered as constants)? I thought this was one of Bruce (but not necessarily Wally) O'Dell's criticisms of your work, whether justified or not. (I'm sure Wally would criticize it too since he has his own optimizers built into his e-votin' machines.)

Me:
The optimizer INDEPENDENTLY comes up with the actual partisan WPE’s based on the Bush 2-party Vote: 51.24%.

You:
I agree (I think) that fraud was unlikely in Kerry strongholds, more likely in moderate Kerry and competitive precincts and some moderate Bush, but I don't think there were enough Bush strongholds to make a determination and there weren't enough of them to change the outcome of the election either. I think this part of the analysis (Bush strongholds) is therefore a waste of time. That said, if you take a Kerry stronghold and REVERSE it, you would get a Bush stronghold, with a hell of a lot of WPE!

Me:
Did you read every line of the output? The optimizer shows where and how many votes were, based on the WPE’s and Kerry exit poll response. There were more than enough to change the election in competitive and Bush precincts. Just look at votes the optimizer generates by partisanship. Note that 3.45 million out of the 4.1 million votes switched came from moderate Bush and competitive precincts.

You:
Furthermore, if, as I believe, it is possible (and easier) to change the results in every precinct in a given county, which would include the whole range of partisanship, there is no reason to suspect any particular category, which would explain the high WPEs in both directions and the high abs mean WPE. Some Bush-loving precincts could have just been victims of "collateral damage" within a large reversal of results for an entire pro-Kerry county, either on Election Day, or during Early Voting periods (partial reversal) where the chain of custody of ballots and machines is tenuous at best.

Me: There are NOT high WPE’s in both directions. The high WPE’s are in moderate and Bush strongholds – as are the alphas. In the two Kerry strongholds, alpha= 1.06. In the three others, alpha = 1.18

You:
In other words, it's a lot easier to commit the fraud more generally than it is to cherry pick individual precincts, so I don't think singling out specific categories of precincts is particularly useful, except in Ohio where ballot order rotations in collocated precincts frequently made it possible to switch Bush and Kerry votes.

Me:
I disagree. First of all, you want to go where the bulk of the votes are – and they are not in the High Bush, High Kerry precincts. You want the precincts which are most vulnerable (i.e., competitive and moderate Bush; Repubs for Kerry). There is no need to go to the heavy Dem precincts; disenfranchisement will take care of that.
The beauty of the optimizer is that feasible results are output, and non-feasible rejected, based on a handful of very important inputs.

1. We know the Bush/Kerry 2-party vote.
2. We know the partisanship groupings.
3. We know the partisanship WPE’s.
4. We know the partisanship response rates.

From this, the Optimizer derives the aggregate Kerry poll (vote) of 52.16% necessary to reproduce the given WPE and other constraints. It then calculates the Kerry winning margin, based on 121 million votes (2-party).

In addition, the optimizer derives the MINIMUM aggregate partisanship alpha NECESSARY to satisfy constraints 1-4 above. And alpha = 1.1544, not 1.12. Right away, the foundation of rBr (56/50=1.12 alpha) is undermined. FEASIBLE partisanship alphas fall out of the optimizer as well – and they put the lie to uniform response BIAS, which is what Baiman has been saying all along.

So now Mitofsky says he never claimed uniform bias. That just made it more difficult for Febble to force feed her Fancy Function.

I HAVE DONE NO SUCH FORCE FEEDING. THE OPTIMIZER IS A SEARCH ALGORITHM WHICH DERIVES MATHEMATICALLY FEASIBLE SOLUTIONS TO THE ANALYSIS OF PARTISAN PRECINCT EXIT POLL RESPONSE.

The analysis fully and independently confirms UCSV’s simulation and in addition, gives us further confirmation for all other state and national exit poll analyses which showed Kerry winning.

Finally, I want to congratulate you on your state responder analysis. If there is any way ton adapt it to the Optimizer, I would love to do it.

[TruthIsAll]

26. I forgot to say...

It comes up with the same mean WPE's as per Mitofsky, whuich 
The model does nothing with median WPE.

Here is the calculation
Kv=Kerry vote
Bv=Bush vote
Kp=Kerry poll
Bp=Bush poll

WPE = (Kv-Bv)- (Kp-Bp)

WPE HighB	Bush	Even	Kerry	HighK	Total/Avg
						
Kv-Bv	-60.2%	-20.0%	-2.0%	20.1%	60.1%	-2.48%
Kp-Bp	-50.2%	-13.9%	6.5%	26.0%	59.8%	4.29%
WPE	-10.0%	-6.1%	-8.5%	-5.9%	0.3%	-6.77%

[Bill Bored]

27. All I can say is

there were high WPEs in both directions, meaning +/-. You can see this from the scatter plots and also from the higher abs (unsigned) mean WPEs in the tables.

So any hypothesis for the fraud based on the assumption that these exit polls were correct would also have to explain the high abs mean WPEs.

I said I didn't think "Bush strongholds" alone could determine the outcome of the election, but then you said "competitive and Bush precincts" could. I agree with that, but some of these could have been Kerry precincts and competitive precincts pre-fraud, morphed into competitive and Bush precincts respectively post-fraud. That's what the final distribution looks like.

Febble's function really only looks at WPE in a new way by compensating for the effects of partisanship. It neither proves rBr nor disproves fraud except to show that unlike WPE, alpha was not correlated with precinct partisanship. So what? This is really only controversial if our fraud hypothesis depends on vote shifting being correlated with partisanship. I don't think this has to be the case. (USCV might, but I haven't finished reading the latest version of their paper.)

As far as my state responder analysis, I'd feel better if it worked when actual numbers of votes were taken into account by weighting. It doesn't really, although the trend is still there. I guess we could say that so far we've failed to confirm rBr at the state level, even when population is taken into account. That's something.

[kiwi_expat]

34. E-M Report says high WPEs in both directions in all five elections.

"It is possible to have a large sampling error and no statistical bias. The 2004 problem is the statistical bias. In other years, when the statistical bias was smaller, the overall error has been almost as large as in 2004." - E-M Report page 31.

However, if the reason that the WPE scatter plot has shifted "north" in 2004 (and in 1992?) is fraud, it is not obvious to me why the sampling error PORTION (the non-fraud portion) of the individual WPEs in 2004 would be smaller than the sampling error PORTION in the other years.


Regarding alpha not correlating with precinct partisanship, I just want to remind folks that Febble says that the alpha scatter plot is not inconsistent with vote shifting being correlated with KERRY partisanship. (i.e., if the statistical bias was caused by fraud, the fraud most likely occurred in the Kerry strongholds.)

Cheers.

[TruthIsAll]

42. But aggregate alpha most definitely correlates with precinct partisanship

Just look at the numbers. As you move from Kerry to Bush partisanship, MEAN alpha steadily increases.




Here's a graph of WPE by precinct partisanship. WPE is near zero in High Kerry partisanships. The underlying WPE trend increases as you move to High Bush, but unlike alpha, it's not monotonic.

[kiwi_expat]

45. It's my understanding that the alpha slope is not statistically significant

The reason that Febble is inclined to think that fraud is more likely to have occurred in the KERRY strongholds, is that such fraud, itself, would shift the alpha scatter plot points upwards and towards the middle - which is indeed where they are centered. She has explained it well somewhere - I can try to dig it out.

[TruthIsAll]

50. No, Febble said that USCV focus on High Bush precincts was incorrect.

And she is correct, when you consider the MAGNITUDE of the vote discrepancies.

If you look at the exit poll response optimizer analysis, you will see that 3.45 million of the 4.1 million votes which were switched from Kerry to Bush occurred in competitive and moderate Bush precincts.

You are missing the overall by complicating what appears to be very clear-cut (at least to me).

Kiwi, have you looked at the optimizer input/output?

[kiwi_expat]

52. Febble's post: fraud most likely concentrated in Kerry strongholds.

"However, one remaining hypothesis, to my mind, that might account for massive fraud, but nonetheless be consistent with the observed non-significant slope, is that fraud may have been concentrated, not in Bush strongholds, but in Kerry strongholds - thus counteracting the slope." -Febble
http://www.democraticunderground.com/discuss/duboard.php?az=show_topic&forum=203&topic_id=374482#375572

Febble has suggested that the best precincts to recount to look for fraud are the Kerry strongholds.

I have no doubt that most of the fraud occurred in the competitive precincts, because most of the precincts are in the competitive-range (i.e., in the middle of the scatter plots).

But I believe that Febble's analysis indicates that an individual high-Kerry-partisanship precinct has a higher probability of having fraud than does an individual competitive-range precinct.

Cheers.

[TruthIsAll]

54. Minimum aggregate alpha is 1.155 (optimizer), not 1.12 (E-M)

That would effect the individual partisanship alphas.

You can do the numbers yourself.

[kiwi_expat]

46. My fraud-hypothesis reading of the WPE scatter plot -

As we all know, the rBr-hypothesis reading of the WPE scatter plot is that all the WPEs are caused by +/- sampling error.

It has been suggested that the fraud-hypothesis explanation of the WPE scatter plot must be that all of the WPEs are caused by +/- fraud (i.e., pro-Bush and pro-Kerry fraud). But that is clearly nonsensical.

We have to accept that a good portion of each precinct WPE must be sampling error - as we agree has been the case when the mean WPE is close to zero.

So, using the fraud-hypothesis, the individual precincts' WPEs must be a combination of sampling error and fraud.

But why would sampling-error-plus-fraud have the same "overall error" as just sampling-error? Upon reflection, the answer was obvious:

The absolute values of the WPE "Kerry overestimate" points (shown above the x-axis on the scatter plot) represent the sampling error PLUS (pro-Bush) fraud. The absolute values of "Bush overestimate" points represent sampling error MINUS (pro-Bush) fraud.

[kiwi_expat]

73. I should have said....

Depending upon the hypothesis, the absolute values of the WPE 'Kerry overestimate' points represent the sampling error PLUS (pro-Bush)net sampling bias, or PLUS (pro-Bush)net fraud.

The absolute values of 'Bush overestimate' points represent sampling error MINUS (pro-Bush)net sampling bias, or MINUS (pro-Bush)net fraud.


(Sorry about saying "(pro-Bush)net fraud", implying that there might possibly be pro-Kerry fraud. The pro-Kerry fraud could well be zero.)

[Bill Bored]

48. Right but not necessarily Kerry "strongholds"

Could be moderate Kerry or toss-up precincts too couldn't it?

[kiwi_expat]

49. I'm sure you're right. But strongholds would probably be most affected.

I understand that a downward slope from *true* Kerry-strongholds (if we really knew the correct vote) would produce the post-fraud flat alpha regression line we see.

Of course some of those *true* strongholds would now appear (post-fraud) as middling Kerry-partisanship precincts.

It is all so confusing. :-)

[Bill Bored]

51. Confusing but what we see the most of are moderate Bush precincts.

If these were toss-ups pre fraud, they'd become moderate Bush post fraud.

Or if you had moderate Kerries pre fraud, with their votes reversed post fraud, they would become moderate Bush. This is only too easy to accomplish with paperless DREs or un-inspected VVPATs or VVPBs. If it was done on a county level in moderate Kerry counties, you'd see a bunch of moderate bush precincts, Bush winning the county instead of Kerry, and some collateral damage whereby a few Bush precincts would be switched to Kerry, but not enough to reverse the reversal in the county!

This theory explains the high absolute mean WPEs and Alphas as well as the high distribution of moderate Bush precincts.

Even if we don't know the precincts in question in the exit polls, we could look for ANY reversals in the vote from 2000 to 2004. Note that a reversal in the vote can still be within the margin of error of these polls too.

I think the focus on the 40 Bush strongholds is a waste of time because:
a) there weren't enough of them to swing the popular vote;
b) the bias index would seem to refute the idea that fraud was concentrated therein.

I think we need to find other evidence consistent with the pre and post fraud distribution of precinct partisanship, and also consistent with the bias index which was not correlated with partisanship.

I will say that in Ohio, you had Bush counties that went even more for Bush in 2004! But as you know, Ohio had other ways to switch votes (e.g., by exploiting ballot order) besides DREs and the like. Even ballot order would be easier to exploit in Kerry precincts though!

And we really should look at precinct results in FL!

[TruthIsAll]

53. It is NOT confusing - if you just look at the numbers.

You keep referring to Febble. Good.
Now focus on the optimizer.

Once again. Did you look at the optimizer?
Did you check the vote discrepancies by partisanship?
Do you see where the bulk (3.45mm) of the deviations (4.10mm) are?
Do you have any questions about how the optimizer works?

It's really quite straightforward.
It uses Mitofsky's WPE and response data.
It derives Kerry's vote based on E-M data.
It derives minimum precinct K/B.
It does this using an optimal search algorithm.
No guess work here.

If you believe Mitofsky's WPE/response numbers and the Bush vote count, you must believe the Optimizer's results for what Kerry's numbers had to be to match E-M WPE's, response rates, alphas.
The E-M stated aggregate Alpha of 1.12 (56/50) is not feasible. Start from there.

No scatter plot is necessary.
Don't blind yourself trying to read it.

The results are intuitive.
The heavy votes are not in High Bush/High Kerry.
The heavy votes are in competitive/Moderate Bush.

[kiwi_expat]

55. I agree that the heavy VOTES are not in High Bush/High Kerry precincts.

(See my post #52.) But fraud could be "concentrated" in High Kerry precincts - i.e., the PROPORTION of High Kerry precincts which had fraud could be higher than the proportion of Competitive/Moderate precincts which had fraud. Does your optimizer deal with that question?

My immediate excuse for not looking at your optimizer is that I have the flu.

[TruthIsAll]

56. Well, get over the flu, then.

Believe me, the output from the optimizer is very easy to follow.
As to how the nonlinear search algorithm works, that's another story.

But there is no need to go there. Just know that it works. You can verify the results yourself. Just work top-down, line by line. It won't take you long.

[kiwi_expat]

57. I am very wary of black boxes! :-) :-)

"As to how the nonlinear search algorithm works, that's another story.
But there is no need to go there. Just know that it works." -TIA

Even though I might not be able to understand how the "nonlinear search algorithm" works, some others on DU might be able to understand it, if the code were available. Even I could probably get the gist of it, from the code. (I have done a lot of maintenance programming: problem-solving without understanding the entire program.)

Cheers
kiwi

[TruthIsAll]

58. Just verify the alphas and the wpe's and the vote switches.

You can do that, can't you?

One final point. And this should do it for you.


	Vote Percentages	Votes (millions)	
	Poll	Vote	Diff	Poll	Vote	Diff 
Kerry	52.15%	48.76%	-3.39%	63.127	59.027	-4.100
Bush	47.85%	51.24%	3.39%	57.929	62.029	4.100
Diff	4.29%	-2.48%	-6.77%	5.199	-3.002	-8.201
						
Pship	HighB	Bush	Even	Kerry	HighK	Total/Avg
Prct	40	415	540	165	90	1250
Vote	3.874	40.191	52.296	15.979	8.716	121.056
Pct	3.2%	33.2%	43.2%	13.2%	7.2%	100.0%


Verify the number of votes in each partisanship category (in
millions). 

You will quickly see that the bulk of the votes were in
moderate Bush and competitive precincts: 

NINETY-TWO MILLION OUT OF 121 MILLION VOTES. 
AND 3.45 MILLION OF THE TOTAL 4.10 MILLION VOTES WHICH
DEVIATED TO BUSH WERE IN THE TWO CATEGORIES.

That should tell you all you need to know.

[kiwi_expat]

59. That is where the bulk of the votes are. Where is highest fraud RATE?

Yes, I agree that the bulk of the votes are in the moderate Bush and competitive precincts. I double-checked by looking at the WPE scatterplot (forgive me!).

But I am not talking about where the bulk of the votes or the bulk of the fraud is. I want to know which partisanship category has the highest fraud RATE. Sorry I didn't use the word "rate" before.

Which partisanship category has the most fraud per, say, 100 votes?

Cheers.

[TruthIsAll]

60. WPE and alphas steadily decrease from Bush to Kerry precincts

You have a tendency to miss the overall.

Look close. See a pattern here?

1 alpha from High Bush to High Kerry:
K/B	1.505	1.153	1.174	1.098	0.996

2 WPE from High Bush to High Kerry:
	-10.0%	-6.1%	-8.5%	-5.9%	0.3%


		OPTIMIZER OUTPUT SUMMARY				
						
	Vote Percentages	Votes (millions)	
	Poll	Vote	Diff	Poll	Vote	Diff 
Kerry	52.15%	48.76%	-3.39%	63.127	59.027	-4.100
Bush	47.85%	51.24%	3.39%	57.929	62.029	4.100
Diff	4.29%	-2.48%	-6.77%	5.199	-3.002	-8.201
						
Pship	HighB	Bush	Even	Kerry	HighK	Total/Avg
Prct	40	415	540	165	90	1250
Vote	3.874	40.191	52.296	15.979	8.716	121.056
Pct	3.2%	33.2%	43.2%	13.2%	7.2%	100.0%
						
RESP.	56.1%	54.9%	51.9%	54.9%	52.9%	53.5%
Dev	2.6%	1.4%	-1.6%	1.4%	-0.6%	0.0%
						
ALPHA						
K/B	1.505	1.153	1.174	1.098	0.996	1.155
Dev	30.4%	-0.2%	1.6%	-4.9%	-13.7%	0.0%
ln a	0.409	0.142	0.160	0.094	-0.004	0.144

VOTE						
Kerry	0.77	16.06	25.62	9.60	6.98	59.027
Pct	19.8%	40.0%	49.0%	60.1%	80.1%	48.76%
						
Bush	3.11	24.13	26.68	6.38	1.74	62.029
Pct	80.2%	60.0%	51.0%	39.9%	19.9%	51.24%
						
RESPONDERS						
Kerry	0.96	17.29	27.84	10.07	6.97	63.127
Pct	24.8%	43.0%	53.2%	63.0%	79.9%	52.15%
						
Bush	2.91	22.90	24.46	5.91	1.75	57.929
Pct	75.2%	57.0%	46.8%	37.0%	20.1%	47.85%
						
REFUSERS						
Kerry	13.4%	36.3%	44.4%	56.5%	80.2%	44.88%
Bush	86.6%	63.7%	55.6%	43.5%	19.8%	55.12%

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

[Febble]

61. Someone just requested I respond here

So I will.

Yes, the WPEs have the tendency TIA states.

But alpha does not. And, as I argued in my paper, alpha (actually ln(alpha)) is the more appropriate measure.

This is because the WPE tends to produce an artefactual slope where there is an overall red shift, which in this data there is, as we know, only too well. The point of using ln(alpha) instead of WPE is to get rid of the artefactual slope induced by the WPE, and to see whether there really is an underlying tendency for bias (in count or poll) to be greater as Bush's apparent share of the vote increases.

The statistical test for such a tendency is a linear correlation.

Mitofsky gave us both correlations. The correlation between WPE and Bush's share of the vote has a significant slope. The correlation between ln(alpha) and Bush's share of the vote is zero. There is no slope. Ergo, when you remove the artefact, no slope remains.

I do not know how TIA is computing alpha, but firstly, alpha cannot be computed accurately without precinct level data, and secondly, the statistical significance of any linear trend cannot be computed without a variance term. When the trend between precinct-level computed alphas and Bush's share of the vote is tested for significance, using the variance in the data, it is not significant.

It doesn't disprove fraud. It doesn't prove "rBr".

What it does do is suggest that if there was massive fraud, it must have been concentrated in precincts whose true vote count for Bush was relatively small. This is because of the the bell curve of precinct vote share (more even and moderate precincts than extremes). For those who are interested, look at the WPE_index scatterplot, downloadable here:

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

It plots ln(alpha)(Y axis - called "WPE_index) by Bush's share of the vote (X axis) for every precinct in TIA's analysis (1250 precincts).

The questions is: do a large (or even small) proportion of precincts occupy their current positions on the plot because they have been been "shifted" upwards (more bias) AND rightwards (more Bush) by fraud? Or are they in their "correct" positions by vote count, and have they merely been shifted upwards (more bias) by polling bias? without being shifted rightwards?

Because of the bell curve, if a proportion of precincts owe their current positions to fraud, and if that fraud was randomly distributed throughout the plot, there should be a greater density of biased precincts as you move rightwards, because most of the shifted precincts would move from the bulk in the middle upwards and rightwards, concentrating them at the Bush end of the plot. In other words, there should be a tendency for precincts on the right of the plot to be higher than those on the left. This would give a tendency for ln(alpha) to be higher on the right than on the left, and to generate a positive regression slope. Despite what TIA's optimizer implies, this is not the case. The regression line is flat.

So one might conclude that instead, the whole plot has been simply shifted upwards by polling bias.

However, an alternative conclusion is that the greatest fraud was concentrated in Kerry strongholds. If precincts in the middle should really be lower (less biased) and further to the left (ie. they really should have a higher Kerry vote) but precincts to the right are in their true positions (fewer fraudulent precincts or smaller magnitude fraud) then that would tend to "disguise" the slope, by hoisting the regression line (think of it as a teeter-totter, with the mean as the fulcrum) upwards on the left and levelling it out.

What I am saying therefore, is that if you want to postulate massive fraud, and I agree there is a way that massive fraud could be consistent with the data, you have to postulate that the greatest fraud was in precincts that are more strongly Kerry (but were shifted towards the Bush end), and that less or smaller fraud occurred in the high Bush precincts.

This would be consistent with the data as it actually is.

The trend TIA is inferring (higher alpha in higher Bush precincts), would, if it existed, be consistent with randomly distributed fraud , but in fact that trend does not exist. That is what the correlation of zero tells us. That there is no signficant linear trend.

If the optimizer is giving significantly higher alpha values for the high Bush precincts then it is not doing what the data is doing. Either the values are not correct, or the trend for them to be higher at the high Bush end is not significant.

Lizzie

[TruthIsAll]

62. Aggregate alpha (1.158) = weighted sum of partisan alphas

which satisfies ALL constraints, including WPE

........HBush    Bush   Compet. Kerry   HKerry Total
K/B	1.500	1.187	1.156	1.098	0.996	1.158
Vote	3.874	40.191	52.296	15.979	8.716	121.056
Pct	3.20%	33.20%	43.20%	13.20%	7.20%	100.00%
						
K/B*Pct	0.048	0.394	0.499	0.145	0.072	1.158


EXIT POLL RESPONSE OPTIMIZER	
7/9/05 5:26 AM

Given the data: 						
1- Bush 2-party vote (51.24%)					
2- Exit poll response by Partisanship category			
3- "Within Precinct Error" (WPE) by partisanship
category

Determine the following:						
1- Exit poll results - aggregate and by partisanship
category	
2- K/B (alpha) - aggregate and by partisanship category 		
3- Required Bush percentage of refusers
4. Probability of deviation from exit poll to vote

KEY RESULTS:
1. Kerry won 52.15% of the 2-party vote, 3.39% higher than
his recorded 48.76%,
2. Kerry won 51.63% of the total vote, 3.35% higher than his
recorded 48.28%.
3. Kerry won 63.127mm votes, 4.10 more (6.50%) than his
recorded vote.
4. Bush won 57.929 mm votes, 4.10 less (7.08%) than his
recorded vote.
5. Of the 4.10mm vote shift, 3.45 were in competitive and
moderate Bush precincts.

6. Minimum aggregate alpha (k/b) is 1.155. The 1.12 stated at
AAPOR is NOT feasible.
7. Alpha is 1.062 in Kerry strongholds, compared to 1.214 in
other categories.
8. Alpha steadily increases from High Kerry (1.0)  to High
Bush (1.50) categories, indicating a positive correlation
between vote discrepancy and Bush partisanship.

9.Bush needed 55.13% of refusers for his recorded 51.24%
2-party vote, compared to 47.85% of responders.
						
10. The probability of a 3.39% discrepancy between the exit
poll and vote is 1 in 63 billion.



2004	2-PTY	2-PTY	TOTAL			
VOTE	PCT	VOTE	COUNT			
Kerry 	48.76%	59.027	48.28%			
Bush	51.24%	62.029	50.73%			
Total	100%	121.056	122.267			

EXIT POLL						
Kerry 	52.15%	63.127	51.63%			
Bush	47.85%	57.929	47.38%			
Diff	-3.39%	-4.10	-3.35%			
Diff/K	-6.50%	-6.50%	-6.50%			
Diff/B	-7.08%	-7.08%	-7.08%			
						
PROBABILITY of a 3.39% discrepancy between exit poll and
vote:
 1 in	62,953,509,332				
						
PARTISAN ALPHA						
Kerry strongholds:1.062					
All Other: 1.214					
						

WEIGHTED AVERAGE (AGGREGATE)						
Response - R	53.59%					
K/B - alpha	1.158					

PERCENT OF EXIT POLL REFUSERS						
Kerry 	44.87%					
Bush	55.13%					
						
PARTISANSHIP RANGE CONSTRAINTS						
1250	Strong Bush		Strong Kerry			
Prcts	40	415	540	165	90	
						
KERRY WIN%						
Min	0%	20%	40%	60%	80%	
Max	20%	40%	60%	80%	100%	
						
RESPONSE						
Min	56%	55%	52%	55%	53%	
Max	56%	55%	52%	55%	53%	
						
ALPHA (K/B)						
Min	0	0	0	0	0	
Max	10	10	10	10	10	
						
WPE						
Min	-10.0%	-6.1%	-8.5%	-5.9%	0.3%	
Max	-10.0%	-6.1%	-8.5%	-5.9%	0.3%	
						
E-M	-10.0%	-6.1%	-8.5%	-5.9%	0.3%	
						
						
		OPTIMIZER OUTPUT SUMMARY				
						
	Poll	Count	%Diff	Poll	Count	Diff(mm)
Kerry	52.15%	48.76%	-3.39%	63.127	59.027	-4.100
Bush	47.85%	51.24%	3.39%	57.929	62.029	4.100
Diff	4.29%	-2.48%	-6.77%	5.199	-3.002	-8.201
						
Pship	HighB	Bush	Even	Kerry	HighK	Total
Prcts	40	415	540	165	90	1250
Votes	3.87	40.19	52.30	15.98	8.72	121.056
Pct	3.2%	33.2%	43.2%	13.2%	7.2%	100.0%
						
RESP.	56.0%	55.0%	52.0%	55.0%	53.0%	53.59%
DevAvg	2.4%	1.4%	-1.6%	1.4%	-0.6%	0.0%
						
ALPHA						
K/B	1.500	1.187	1.156	1.098	0.996	1.158
K/ 50B	75.0	59.3	57.8	54.9	49.8	57.9
DevAvg	29.5%	2.5%	-0.2%	-5.2%	-14.0%	0.0%

VOTE						
Kerry	0.77	13.11	28.58	9.59	6.97	59.027
Pct	20.0%	32.6%	54.6%	60.0%	80.0%	48.76%
						
Bush	3.10	27.08	23.72	6.39	1.74	62.029
Pct	80.0%	67.4%	45.4%	40.0%	20.0%	51.24%
						
RESPONDERS						
Kerry	0.97	14.34	30.80	10.06	6.96	63.127
Pct	25.0%	35.7%	58.9%	63.0%	79.9%	52.15%
						
Bush	2.91	25.85	21.49	5.92	1.76	57.929
Pct	75.0%	64.3%	41.1%	37.1%	20.2%	47.85%
						
REFUSERS						
Kerry	13.6%	28.9%	50.0%	56.4%	80.2%	44.87%
Bush	86.4%	71.1%	50.0%	43.6%	19.8%	55.13%

VOTE DEVIATION						
Kerry	-0.19	-1.23	-2.22	-0.47	0.01	-4.10
Pct	-20.0%	-8.5%	-7.2%	-4.7%	0.2%	-6.5%

WPE						
Calc	-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	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%

[Febble]

63. Do you mean weighted mean?

And are your partisan alphas also weighted means?

Or medians?

You are aware that alpha is the numerator of a ratio, therefore the mean is not a good measure of central tendency. Pro-Bush alphas will range from zero to 1 and pro Kerry Alphas will range from 1 to infinity, so means won't mean much, as the model distribution does not have a normal distribution.

But from the little I understand about your optimizer these are not means but "optimized" estimates of some sort. What do you think they are an estimate of?


Each the 1250 precincts has its own alpha, which clearly you can't compute without the precinct level data, so three possibilities that come to mind are:


1. Your optimizer could be estimating the mean of these precinct alphas in each category (which would have the problems I mention above)

2. It could be estimating the median alphas (which would be more valid, I think)

3. Or it could be estimating the alpha corresponding to the mean ln(alpha) or arctan(alpha,) both of which will have a normal distribution, which would also be valid.

I agree the true mean alpha for these 1250 precincts (which do not include precincts in which there were a substantial proportion of absentee ballots) is greater than 1.12. 1.15 seems reasonable.

However, I do not agree that the real, valid, aggregate values show the linear trend you describe. I think the most valid of the three is the third (ln(alpha) or arctan(alpha) and we know, from Mitofsky's plot, that there is no significant trend for ln(alpha) to vary linearly with Bush's share of the vote. As I say, this does not rule out fraud, but it does imply that if it is large, it must be greater at the Kerry end of the plot. Which, as I think I've said before, seems more in line with your earlier evidence.

Lizzie

[TruthIsAll]

64. The optimizer determined the partisan alphas AND the wtd mean aggregate

The aggregate total alpha is the weighted (by votes) sum of the individual partisanship alphas.

They are NOT calculated mean alphas. These are NOT calculated median alphas. These are much more than that.

These are alphas derived by the optimizer which satisfy the input partisan alpha range constraints (0 to 10) AND the E-M partisan WPE constraints AND the E-M individual partisanship responses (R) AND the Bush 51.24% two party vote - ALL of which were provided by E-M.

If you can come up with a set of alternative partisan alphas which will satisfy ALL the above constraints, please do so.

I would love to see it.

[TruthIsAll]

65. Febble, if you would like to have me run the optimizer for your input

range constraints to test your own scenarios, I would gladly do so.

Let's determine if they are mathematically FEASIBLE (ie POSSIBLE).
If they are, then we will judge whether they are PLAUSIBLE (ie REALISTIC).

[Febble]

66. How can it be the weighted "sum"?

Alpha is a ratio.

If you sum your five ratios you will get a number that is larger than any of them. If you do a weighted sum you will get an even larger number.

What you have produced is the weighted mean. You multiplied alpha in each category by the number of precincts in each category, summed the total and divided that by the total number of precincts. At least, I did that with your five alphas and got the answer you got.

But what I don't know is what your category values are. Are they estimates of the mean alpha for the precincts in that category, or what? Those five alphas must represent some measure of central tendency for alpha for each group.

Lizzie

Edit:

I think you must have weighted by numbers of precincts, not by numbers of votes, because I take it you do not have the vote totals for each precinct or category, unless it is hidden somewhere in the E-M report where I have not found it. In any case, precinct Ns seem to give your answer.

[TruthIsAll]

67. I weighted the precincts and estimated the votes...

which of course, is an approximation, since I am assuming that average votes/precinct are fairly constant across partisanships. Of course I do not have the actual votes to work with, but the results must be very close to reality.

So, too, the optimizer-derived partisan alphas must be very close to the actual means. The optimizer determines feasible alphas which satisfy all the constraints.

[Febble]

68. Well, the actual means,

as I've said, would be a poor measure of the central tendency of alpha within the precinct categories. Median would be be more valid, and closer to the answer you would get if you took the mean of either the log or arctangent of the alphas, and then derived the alpha value that corresponded to those means, which I think IS valid.

And you say "the results must be very close to reality" - why must they be? Do you have confidence limits? Because the point is that in order to establish whether there really is a significant trend for alpha (or ln(alpha))to be higher as Bush's share of the vote increases, you have to know the the confidence limits of your means, which involves some estimate of the variance.

And the answer is, because Mitofsky did it, and I have verified it, is that there is no such trend.

ln(alpha) does not significantly increase as Bush's share of the vote increases.

You are about right in your estimate of the mean alpha, but not about that trend. There are many reasons why the aggregate values might give an illusory trend, but the fact is that it isn't there.

So, as I say, if the red-shift was due to fraud, fraud must have been greater in precincts that at started at the Kerry side of the plot, and shifted towards Bush - but not too far, otherwise there would be a slope. And there isn't one, not in the actual data.

Plenty of red-shift, but no slope.

[TruthIsAll]

69. Disagree with the model? Well, then stress-test it....

The optimization satisfies all the constraints.
Do you agree with the input constraints?
Do you disagree with the outputs?

Here's your chance. Put it to the test.

Which input parameters do you wish to change?
WPE? They are given by Mitofsky.
The Bush 2-party vote? That's given.
Response rates? Given by Mitofsky.

You talk about arc tangent transforms or ln transforms or variances. But what about the algebra; what about constrained optimization; what about feasibility. Let's talk about real numbers; the total vote; vote percentages; the derived Kerry/Bush exit poll percentages which cause the deviations.

The model calculates deviations by precinct category. The fact is, and you have not been able to dispute this, it generates very plausible results.

Do you disagree with the assumption that votes/precinct are equal across partisanship? If you agree with that assumption, then you should agree that the significant vote deviations occurred in moderate Bush and non-partisan precinct groupings and that deviations in High Bush and partisan Kerry precincts are minor when compared to the total.

The vote deviations from the FINAL EXIT POLL increase in one direction - from Kerry to Bush partisanship. That is obvious from the data. Alpha (yes, Febble, I know it's a ratio) increases in the same direction - from Kerry to Bush. These measures confirm one another and make intuitive sense.

You would expect that Reluctant Republican responders who wanted to vote for Kerry would reside in Bush and competitive precincts - not in heavily democratic minority precincts.

The results of the optimizer make a lot of sense. They independently confirm the National Exit Poll at the 12:22am timeline of 13047 respondents. Kerry was the winner by 5 million votes - and that's a minimum. The 5 million do not include those disenfranchised who came to vote but never did due to the long lines and few machines, or those who never came to vote because their registrations were lost or invalidated.

[Febble]

70. I cannot possibly comment

on your optimizer uinless I know how it works.

The E-M report gives mean, median and absolute WPEs, so any model has to match those, but you say your optimizer is not a model. So I have no views. I do not know what it is trying to do. All I know is that the alphas that you output are not the alphas in the data, and that the alphas in the data do not vary significantly with increase in Bush's vote.

I am talking about real numbers. Real numbers are exactly what I am talking about.

If you want the algebra, here it is: The alpha for a precinct is:

Kerry responses/Kerry votes divided by Bush responses/Bush votes. I call Kerry responses/Kerry votes "Kp" and Bush response/Bush votes "Bp", i.e. Kp for Kerry participation rate and Bp for Bush participation rate. So alpha = Kp/Bp.

If you want the mean alpha for a category you cannot simply take the mean of all the precinct alphas, because, as you say, it is a ratio, and therefore does not have a normal distribution. What you do is you take the arctangent of the ratios, take the mean of the arctans, then the tangent of the mean. Mean alpha = the tangent of the mean arctan(Kp/Bp).

Obviously you can't do this because you don't have the data. So you approximate to it. Fair enough, it is the best you can do with what you have got.

But I am talking about real data. Mitofsky has the real data, and he has given the answer for ln(alpha). I also have checked the answer using arctan. There is no significant slope between ln(alpha) or arctan(alpha) and Bush's share of the vote. So yes, I dispute that the means are significantly different between categories. They are different, but not significantly different. It is not an assumption. It is a computation. There is no slope.

So whether the optimizer output makes sense or not makes no difference. In the real numbers there is no slope.

And they have absolutely nothing to do with the timeline. They are simply the raw responses, completely unweighted, that went into the estimate. The weights were what produced the timeline changes. the raw data is the raw data. If there is bias in the raw data it is either bias in the poll or bias in the count.

The unweighted data we are working with show, unambiguously, that there was a massively significant red-shift of something around alpha=1.15 on average, with a huge amount of variance, as can be seen from the plot.

But alpha did not vary with partisanship. It varied with lots of other things, but not with that. Not at zero-order correlation level, anyway.

There is no significant slope.

[TruthIsAll]

71. Febble, you STILL don't understand the optimizer.

You:
I cannot possibly comment on your optimizer unless I know how it works.

Me:
You are sure doing a lot of commenting, though. If you want to know how it works, just replicate the numbers, line by line. You know the formulas.

Do the math. And you will see how it works.

You:
The E-M report gives mean, median and absolute WPEs, so any model has to match those, but you say your optimizer is not a model. So I have no views. I do not know what it is trying to do. All I know is that the alphas that you output are not the alphas in the data, and that the alphas in the data do not vary significantly with increase in Bush's vote.

Me:
It sure is a model. I never said it wasn't. And for you to infer that it is not a model is disingenuous to say the least. It is OBVIOUSLY a model. Have you ever designed a model?

You still do not know what the model is trying to do? That is a very interesting comment. I have given you the goals, assumptions and results of the model. And you still don't know what it is trying to do? Or are you in denial of what the model is coming up with?

The alphas produced are very close to the alphas produced by Ron Baiman. What is YOUR estimate of the mean alpha for each partisan grouping?

You:
I am talking about real numbers. Real numbers are exactly what I am talking about.

If you want the algebra, here it is: The alpha for a precinct is:

Kerry responses/Kerry votes divided by Bush responses/Bush votes. I call Kerry responses/Kerry votes "Kp" and Bush response/Bush votes "Bp", i.e. Kp for Kerry participation rate and Bp for Bush participation rate. So alpha = Kp/Bp.

If you want the mean alpha for a category you cannot simply take the mean of all the precinct alphas, because, as you say, it is a ratio, and therefore does not have a normal distribution. What you do is you take the arctangent of the ratios, take the mean of the arctans, then the tangent of the mean. Mean alpha = the tangent of the mean arctan(Kp/Bp).

Me:
That statement shows a complete lack of understanding on your part. I don't take the mean of all the alphas. For the umpteenth time, the model derives precinct category alphas which satisfy the constraints. You are so caught up in your transforms, that you cannot grasp that basic fact.

Febble, I challenge you to prove that the optimizer's category alphas are not feasible. And I challenge you to provide alternate alphas which satisfy the constraints.


You:
Obviously you can't do this because you don't have the data. So you approximate to it. Fair enough, it is the best you can do with what you have got.

Me:
The model doesn't NEED the data. It has enough aggregate data to come up with feasible results. Now, Febble, how did the model determine that aggregate alpha of 1.12 is incorrect and that it must be at least 1.15. How did the model come up with that result without all the data? How would you prove that 1.12 was too low? In fact, when did you ever show that result?

You:
But I am talking about real data. Mitofsky has the real data, and he has given the answer for ln(alpha). I also have checked the answer using arctan. There is no significant slope between ln(alpha) or arctan(alpha) and Bush's share of the vote. So yes, I dispute that the means are significantly different between categories. They are different, but not significantly different. It is not an assumption. It is a computation. There is no slope.

Me: I am talking about real data.

Febble, prove that the alphas I present are not correct. Prove that they are infeasible. You can't. Show us your alphas. I will gladly insert them into the model. Then we will see if the constraints are satisfied.


You:
So whether the optimizer output makes sense or not makes no difference. In the real numbers there is no slope.

Me: There is no slope?

Oh, so alpha or ln alpha is flat across partisanship groups.
Really? From 1.0 for High Kerry to 1.50 to High Bush is flat?

You:
And they have absolutely nothing to do with the time line. They are simply the raw responses, completely unweighted, that went into the estimate. The weights were what produced the timeline changes. the raw data is the raw data. If there is bias in the raw data it is either bias in the poll or bias in the count.

The unweighted data we are working with show, unambiguously, that there was a massively significant red-shift of something around alpha=1.15 on average, with a huge amount of variance, as can be seen from the plot.

But alpha did not vary with partisanship. It varied with lots of other things, but not with that. Not at zero-order correlation level, anyway.

There is no significant slope.

Me: The optimizer has shown that the only partisan alphas which are feasible do in fact vary with partisanship - from 1.0 to 1.50.

There is no significant slope? A monotonic increase across partisanship groups from 1.0 Kerry to 1.18 to 1.50 Bush is not significant? What slope are you looking at?

[Febble]

74. You did, actually

tell me it wasn't a model. So I didn't want to offend you by calling it a model. It looks like a model to me. Fine, it's a model. I understand models.

But if it's a model, it has to model variance. I don't see it doing that yet. If it can match the absolute WPEs given in the E-M report then it will start to model the variance.

But I now know what the values are in the data. Unless you can tell me what measure of central tendency your alphas represent for each category I cannot tell you how close they are to the ones in the data, although I can, and did, tell you what measures I was using, for your information.

What I can tell you, and your DU readers, is that in the actual data, the mean alphas (defined as the mean corresponding to the mean arctan(alpha) or ln(alpha)) are not significantly different across the range of partisanship.

As your model does not model the variance I am not sure how you can test whether your measures of category alphas are signficantly different from each other. If you do want to model the variance, the best values to incorporate would be the absolute WPEs as they are given in the E-M report. No other measure of variance is given.

But I'll stop commenting now, as I've made it clear what my point was regarding the possible distribution of fraud. If you disagree, that is fine. But my own view is that the pattern of fraud that is consistent with the data is fraud that was greater (either more prevalent or larger in effect) where Kerry's vote was stronger. I can, and have, supported that view with reference to the data.

If you want to continue the conversation, feel free to PM me.

Lizzie

[Febble]

75. Fuller response:

You:
I cannot possibly comment on your optimizer unless I know how it works.

Me:
You are sure doing a lot of commenting, though. If you want to know how it works, just replicate the numbers, line by line. You know the formulas.

Do the math. And you will see how it works.

I do not know the formulas in your optimizer, but I have told you mine. And I do not know what your output alphas represent. Each precinct has a different alpha. If you want an aggregate value for a series of values you have to decide on the most appropriate aggregate. You have not yet said which of the possible options your aggregate represents.

You:
The E-M report gives mean, median and absolute WPEs, so any model has to match those, but you say your optimizer is not a model. So I have no views. I do not know what it is trying to do. All I know is that the alphas that you output are not the alphas in the data, and that the alphas in the data do not vary significantly with increase in Bush's vote.

Me:
It sure is a model. I never said it wasn't. And for you to infer that it is not a model is disingenuous to say the least. It is OBVIOUSLY a model. Have you ever designed a model?

Of course. It is what I do. And if you are happy to call your optimizer a model I am more than happy to call it one too. It is.

You still do not know what the model is trying to do? That is a very interesting comment. I have given you the goals, assumptions and results of the model. And you still don't know what it is trying to do? Or are you in denial of what the model is coming up with?

No, I do not know. I know you are producing some measure of "alpha" for each category. I do not know which aggregate measure it is an estimate of. It matters. As do the confidence intervals. I am certainly not in denial. I want to know.

The alphas produced are very close to the alphas produced by Ron Baiman. What is YOUR estimate of the mean alpha for each partisan grouping?

I am afraid I cannot give you those. I can however, and have, told you directly how they are computed, from the actual data. I have also told you that there is no significant trend for arctan(alpha) or ln(alpha) to increase with Bush's vote share. You can disbelieve me if you choose.

You:
I am talking about real numbers. Real numbers are exactly what I am talking about.

If you want the algebra, here it is: The alpha for a precinct is:

Kerry responses/Kerry votes divided by Bush responses/Bush votes. I call Kerry responses/Kerry votes "Kp" and Bush response/Bush votes "Bp", i.e. Kp for Kerry participation rate and Bp for Bush participation rate. So alpha = Kp/Bp.

If you want the mean alpha for a category you cannot simply take the mean of all the precinct alphas, because, as you say, it is a ratio, and therefore does not have a normal distribution. What you do is you take the arctangent of the ratios, take the mean of the arctans, then the tangent of the mean. Mean alpha = the tangent of the mean arctan(Kp/Bp).

Me:
That statement shows a complete lack of understanding on your part. I don't take the mean of all the alphas. For the umpteenth time, the model derives precinct category alphas which satisfy the constraints. You are so caught up in your transforms, that you cannot grasp that basic fact.

Please be civil. I am not saying you take the mean of all the alphas (although it seems to be how you derived your grand weighted mean). What I am telling you is what I have done. What I am asking you is: what do you think your aggregate alpha will most closely match? The mean? The median? or the measure I suggest? Or some other measure of central tendency?

Febble, I challenge you to prove that the optimizer's category alphas are not feasible. And I challenge you to provide alternate alphas which satisfy the constraints.

They may be feasible given the constraints. All I am saying is that they do not match the means, medians or any of the other aggregates that I have computed from the data. They may match some other aggregate, but unless you suggest what, I cannot tell whether it matches.

You:
Obviously you can't do this because you don't have the data. So you approximate to it. Fair enough, it is the best you can do with what you have got.

Me:
The model doesn't NEED the data. It has enough aggregate data to come up with feasible results. Now, Febble, how did the model determine that aggregate alpha of 1.12 is incorrect and that it must be at least 1.15. How did the model come up with that result without all the data? How would you prove that 1.12 was too low? In fact, when did you ever show that result?

I agree, it is fairly easy to approximate, using my formula (and Ron's, and I assume, your own), the E-M aggregate data, and a guesstimate of the mean partisanship of each precinct. It is fairly clear from the E-M aggregate data that the mean alpha is probably a bit higher than 1.12. And in fact it is.

What you cannot do from the aggregate data is test the hypothesis that alpha trends significantly higher as Bush's share of the counted vote increases. That was an important question, worth testing, that could only be tested on the actual data. And it isn't.


You:
But I am talking about real data. Mitofsky has the real data, and he has given the answer for ln(alpha). I also have checked the answer using arctan. There is no significant slope between ln(alpha) or arctan(alpha) and Bush's share of the vote. So yes, I dispute that the means are significantly different between categories. They are different, but not significantly different. It is not an assumption. It is a computation. There is no slope.

Me: I am talking about real data.

Febble, prove that the alphas I present are not correct. Prove that they are infeasible. You can't. Show us your alphas. I will gladly insert them into the model. Then we will see if the constraints are satisfied.

Can't do that, I'm afraid. I am sure yours are feasible. They just don't happen to correspond to values given by any of the measures of central tendency I have proposed.


You:
So whether the optimizer output makes sense or not makes no difference. In the real numbers there is no slope.

Me: There is no slope?

Oh, so alpha or ln alpha is flat across partisanship groups.
Really? From 1.0 for High Kerry to 1.50 to High Bush is flat?

As I've said, the relevant test is a Pearson correlation coefficient. It is not significant. So yes, the regression line is statistically flat. If you look at the WPE_plot you will see that is the case. You will also see why the category estimates are misleading.


You:
And they have absolutely nothing to do with the time line. They are simply the raw responses, completely unweighted, that went into the estimate. The weights were what produced the timeline changes. the raw data is the raw data. If there is bias in the raw data it is either bias in the poll or bias in the count.

The unweighted data we are working with show, unambiguously, that there was a massively significant red-shift of something around alpha=1.15 on average, with a huge amount of variance, as can be seen from the plot.

But alpha did not vary with partisanship. It varied with lots of other things, but not with that. Not at zero-order correlation level, anyway.

There is no significant slope.


Me: The optimizer has shown that the only partisan alphas which are feasible do in fact vary with partisanship - from 1.0 to 1.50.

There is no significant slope? A monotonic increase across partisanship groups from 1.0 Kerry to 1.18 to 1.50 Bush is not significant?

You cannot tell whether a slope is significant without a measure of the variance. It is why I suggest you produce confidence intervals, using the absolute WPEs, which are the only estimates of variance given in the E-M tables. A small slope can be significant if the variance is small. A large slope can be insignificant if the variance is large. The variance is large.

What slope are you looking at?

I am, of course, looking at the slope in the actual data. The one in the scatterplot. The one I have also, personally computed. The one between ln(alpha) and Bush's share of the vote. And you cannot test the significance of any slope without a measure of the variance.

Any apparent slope given by category aggregates, whether yours, medians, means, whatever, ignores the extra information, apparent in the actual data, but not in the category aggregates, that the low category alpha in the high Kerry category is being leveraged by a few extremely low alphas in precincts where Kerry had around 99% of the vote. There are some very high alphas in that category as well. There are also some very low alphas at the high end of the mod Rep category. The category boundaries are completely arbitrary. If the X axis was divided in four or six, rather than five, you would get very different answers. The question is, is there a significant monotonic trend? The answer is, no. It is a very good example of why aggregate values can be very misleading in statistics, why continous variables should not be arbitrarily categorized, and why variance is critical.

So this is why I say: if you want to know what happened in the exit polls, look at the data. And the data tells you that the slope you would expect from widespread, randomly distributed fraud is not there, even though it may appear to be there from estimates of the aggregate alphas. So a viable alternative is that fraud was greatest at the Kerry end.

But if you want to disbelieve the data, fair enough. You have reason enough to, I suppose. You cannot check it. I can only say, I am honest, I've run the regression myself, and the line is statistically flat. Not only that, but if you exclude precincts where Kerry had more than 95% of the vote (there are no equivalently extreme Bush precincts) the line is not only statistically flat but actually tilts the other way. So even the insignificant slope is actually being leveraged not by anything happening in the higher Bush precincts but by something happening in the highest Kerry precincts.

Or not happening. Which might be interesting.

You and other DUers are welcome to PM me if you want to continue.

Lizzie

Edited for typos. Sorry if there are some left.

[kiwi_expat]

72. HC: Really high vote counts are in Bush partisanship precincts

"Do you disagree with the assumption that votes/precinct are equal across partisanship?" -TIA

I eyeballed the Hamilton County 1013 precincts' vote data to see if any vote/partisanship patterns were obvious.

The most striking pattern is that the really high vote counts (>700) were in the Bush partisanship precincts. (The HC average votes/precinct is 427.)

So, I would be hesitant to assume that votes/precinct are equal across partisanship.

[TruthIsAll]

76. Kiwi, compute average votes/precinct. And....

see how close you come to the total vote.

Also, as far as the 1250 precincts representative precincts are concerned, they are in five groups which sum to 121 million.
That is correct, right?

You can bet your life that the partisanship group totals are very close to reality.

[Febble]

77. The total votes

cast in the 1250 precincts sum to approximately 1.3 million. NOT 121 million. So they represent about 1% of the total vote.

Approximately 100,000 responses were tallied at these 1250 precincts, i.e. about 8% of the total voters in those precincts, and about 0.1% of all voters.

In otherword 99% of voters voted in precincts not included in the poll.

[TruthIsAll]

78. No kidding. Now tell us something we don't know.

That is NOT what I said. I said that I applied the percentage breakdown of the 1250 precincts by partisanship to the TOTAL 2-party vote of 121 million in order to determine the approximate equivalent TOTAL vote by partisanship precinct.

And that should be a VERY GOOD APPROXIMATION. Check the vote deviations per partisanship group.

Now, are you next going to say that the total precinct sample is insufficient to draw conclusions?

[Febble]

80. No, I'm not

The numbers are way large enough to do inferential statistics provided the assumptions of the statistics are met. Where the numbers are sometimes too small is at precinct level.

Regarding the total numbers: it was not clear to me from your post what you thought; what you said was "as far as the 1250 precincts representative precincts are concerned, they are in five groups which sum to 121 million", and this, of course is not literally true. So I just thought the actual numbers might be useful to anyone reading your post.

[TruthIsAll]

79. I refer you to USCV Appendix F: Table 1

"Existing Exit Poll Error explanations for discrepancies
are not consistent with the aggregate data".

These simulation results are not much different than those of
the Optimizer which uses a completely different methodology.

Partisanship Alpha
>>>>>>>> Mean Median
HighKerry 0.98 1.02
Kerry 1.16 1.14
Even 1.19 1.18
Bush 1.15 1.15
HighBush 1.59 1.33

On EDIT: The optimizer
ALPHA...HighB	Bush	Even	Kerry	HighK	Aggregate	
K/B	1.500	1.187	1.156	1.098	0.996	1.158
K/ 50B	75.0	59.3	57.8	54.9	49.8	57.9
DevAvg	29.5%	2.5%	-0.2%	-5.2%	-14.0%	0.0%

[Febble]

81. I am not disputing that both

you and Ron got answers that are quite similar.

But in the actual data there is no significant trend for alpha to be higher as Bush's share of the vote increases. This is partly because of the variance, and partly because you cannot accurately compute alpha from the aggregate WPEs.

Aggregate values can be very misleading.

[TruthIsAll]

82. Not misleading when confirmed using TWO different methods

No, Febble, The Optimizer and Monte Carlo agree.

Those are two very tough nuts for you to crack.

I defer to Ron Baiman TOTALLY as to his expertise as a statistician.

But when it comes to quantitative/financial/operations research analytical modeling...well, let's just say that I've been there, done that...over many years. My clients have always been very satisifed with the results. And we are talking about some very important clients.

[Febble]

83. I am not disputing yours or Ron's skills

I am disputing the quality of the data you have access to.

Back in March, I did the same computations as Ron has done, from the aggregates, and got the same answers.

But the raw data gives different answers.

[TruthIsAll]

84. The challenge still stands: Let's have your input to the model.

Febble, put up or...

Let's see your data.

[Febble]

85. You know I can't do that.

I can only report findings that are in the public domain.

You don't have to trust my skills as a statistician. I am only reporting a correlation coefficient. But obviously I can't prove I am not lying. I'm not, however.

Regarding your model, my only suggestion is that you constrain your model to the absolute WPEs as well as the signed WPEs, as that will give you an approximation to the variance.

[TruthIsAll]

86. The WPE's are constrained to exactly match E-M. And they do.

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%

Just using the numbers they gave us.

The optimizer used these WPE's to prove that E-M stated aggregate alpha of 1.12 is NOT FEASIBLE. HOW OFTEN HAVE YOU REFERRED TO THE 56/50 SPLIT?

Now we know that 1.155 is a MINIMUM AGGREGATE ALPHA -THANKS TO THE OPTIMIZER. YOUR FANCY FUNCTION DIDN'T TELL US THAT, DID IT?

SO IT'S NOT 56 KERRY/50 BUSH, AFTER ALL.
IT'S MORE LIKE 58/50.

Febble, aren't you glad that constrained optimization is helping us disprove rBr?

AND DON'T FORGET: BY ANALYZING THE REAL NUMBERS, WE ALSO KNOW THAT THE FINAL NATIONAL EXIT POLL 43% IS BOGUS; THAT 39.8% IS THE MAXIMUM POSSIBLE PERCENTAGE OF BUSH 2000 VOTERS OF THE TOTAL 122.3 MILLION WHO VOTED IN 2004. BUSH GOT 50.456 MILLION VOTES IN 2000, NOT 52.57 MILLION. THAT'S A FACT.

THAT'S TWO STRIKES..

rBr? Is it still breathing?

[foo_bar]

87. black box quoting

In the words of Warren Mitofsky (the chief conspirator in a plot only non-statisticians can see):

Below is the original statement from page 31 of our January 19 report. Try reading it slowly and stop when you get to the word "hypothetical" in order for its meaning to penetrate.

"While we cannot measure the completion rate by Democratic and Republican voters, hypothetical completion rates of 56% among Kerry voters and 50% among Bush voters overall would account for the entire Within Precinct Error that we observed in 2004."

Clearly, completion rates were not constant. If they were constant then the bias would not vary from precinct to precinct. And it does vary -- quite a bit.

http://www.mysterypollster.com/main/2005/05/aapor_exit_poll.htm

Words do in fact mean things, as Mark L. pointed out:

Yes, the words "hypothetical" and "overall" are important in that sentence. Rather a circumspect sentence to have spawned hundreds of posts denouncing "Mitofsky's rBr theory."

Back to the present:

SO IT'S NOT 56 KERRY/50 BUSH, AFTER ALL.
IT'S MORE LIKE 58/50.

58/50 is an approximation of your "optimizer" (should I capitalize "Optimizer", or do the proper nouns rightfully belong to Microsoft and Excel?) figure, same as 56/50 only imprecise in the other direction. So it seems a bit hypocritical to complain about Mitofsky's choice of precision in an example (clearly hypothetical since it employed whole numbers like 56 and 50, not the strawman's "E-M stated aggregate alpha of 1.12", a factoid that only appears in your posts). The perception is reinforced by one's propensity to round ratios to the nearest eight:

And you are quibbling about the numbers.
87 of 94.
86 of 88.
What the hell is the difference?

http://www.democraticunderground.com/discuss/duboard.php?az=show_topic&forum=104&topic_id=4057474#4080626

(edited spelling error)

[Febble]

89. And I would add

That 56:50 is correct for the full sample. However, when precincts with more than 15% percentage of absentee votes/telephone polled responses are omitted, bringing the sample down to 1250 from the original sample of around 1480 (can't remember offhand) the ratio is somewhat higher. This makes sense if the source of error is in voter selection, which should be more truly random in telephone polls which are drawn by a random number generator, than in face-to-face selection methods at the polling place.

In other words the 1250 sample is an "enriched" sample" and the ratio is slightly higher.

I'd also add that the concept of "alpha" (including its name) was my contribution to the debate.

It is the reason Mitofsky is re-analysing his results.

[TruthIsAll]

90. Since when are telephone polls more accurate than exit polls?

Historical polling comparison (pre vs. post) tells us otherwise.

1) A voter is not undecided AFTER voting.
2) A voter can't change his/her mind AFTER voting.
3) Voters are randomly selected AFTER voting.
4) Voters can't decide to stay home for any reason AFTER voting.

Notice the term "voters".
Not Bush voters.
Voters.

The MoE formula for pre-election telephone polls is 1/sqrt(n), where n is the number polled. Mitofsky MoE table agrees: The MoE is 1.0% for N> 10,000. For the National Exit Poll at 12:22am, N=13047.
According to the formula, the MoE is 0.88% for 13047.

The SAME MoE formula is used for pre-election vs. post-election polls.

Sorry, Febble, in-person exit polls are as accurate as telephone polls.

[Febble]

91. I didn't say they were more accurate

I said sampling was more likely to be random. Or at least that non-random sampling would be less likely to occur at voter selection level.

Anyway, why is not the point. The point is that the 1250 precincts in the subsample had a higher alpha.

They could have had more fraud.

But it doesn't mean the 56:50 is wrong for the full sample.

And I'm not even arguing against fraud at this point TIA. I am simply arguing that to be consistent with the data, it is worth looking at the Kerry end.

[TruthIsAll]

25. Proof that the WPE's are output and match E-M

Take a look at the input range constraints on partisanship
WPE's:
 -10% to +10%. 
Check the CALCulated partisan WPE's at the bottom. 

See the very wide input partisan alpha range constraint band:
from 0.1 to 9.0?
Now check the output partisanship alphas. 

It must be MAGIC! The optimizer came up with exactly the same
WPE's as E-M supplied. And it also produced the required
alphas. 

How? 

Again, the  MODEL had to satisfy ALL constraints: Bush 51.24%,
partisan precinct response rates, minimum feasible aggregate
alpha (1.154). By the way, the model determined that there was
no feasible solution when I input the E-M aggregate alpha of
1.12 (56/50).
By using old-fashioned trial and error, I incremented
aggregate alpha  until it gave me a minimum feasible solution
at 1.1544.

EXIT POLL RESPONSE OPTIMIZER						
7/3/05 4:21 PM						
						
OBJECTIVE:						
Determine partisan K/B (alphas) based on minimum feasible
aggregate K/B (1.1544), Bush 2-party vote (51.24%),  partisan
WPE's and response rates.						
						
Precinct Variable Input Range (Min, Max) Constraints:         
                                         						
1- Response: aggregate (.535);
partisan.53HB,.55B,.52NP,.55K,.56HK                           
                       						
2- Kerry 20% quintile 2-party vote:avg 10%,30%,50%,70%        
                                          						
3- Alpha (K/B):minimum (1.1545)                               
                     						
	
						
VOTE	Mil.	Pct (Input)				
Kerry 	59.027	48.76%				
Bush	62.029	51.24%				
Total	121.056					

POLL	Mil.	Pct				
Kerry 	63.127	52.15%				
Bush	57.929	47.85%				
						
Bush-recorded	51.24%					
						
EXIT POLL RESPONSE (input)						
Aggregate	53.5%	R				
Kerry/Bush	1.1544	alpha				
This is minimum alpha required to satisfy 						
the WPE/partisanship constraints						

PROBABILITY						
Bush vote gain	1.59E-11					
 1 in	62,953,509,332					
Kerry > 50% vote	100.0%					
 1 in	1					
						
PARTISAN ALPHA						
Kerry	1.062					
Bush	1.183					
						
RESPONSE INPUT CONSTRAINTS						
1250	Strong Bush		Strong Kerry			
Prcts	40	415	540	165	90	
						
KERRY WIN%						
Min	0%	20%	40%	60%	80%	
Max	20%	40%	60%	80%	100%	
						
RESPONSE						
Min	55.5%	54.5%	51.5%	54.5%	52.5%	
Max	56.5%	55.5%	52.5%	55.5%	53.5%	
						
ALPHA (K/B)						
Min	0.100	0.100	0.100	0.100	0.100	
Max	9.000	9.000	9.000	9.000	9.000	
						
WPE						
E-M	-10.0%	-6.1%	-8.5%	-5.9%	0.3%	
Min	-10.0%	-10.0%	-10.0%	-10.0%	-10.0%	
Max	10.0%	10.0%	10.0%	10.0%	10.0%	
						
						
		OPTIMIZER OUTPUT SUMMARY				
						
	Vote Percentages			Votes (millions)		
	Poll	Vote	Diff	Poll	Vote	Diff 
Kerry	52.15%	48.76%	-3.39%	63.127	59.027	-4.100
Bush	47.85%	51.24%	3.39%	57.929	62.029	4.100
Diff	4.29%	-2.48%	-6.77%	5.199	-3.002	-8.201
						
Pship	HighB	Bush	Even	Kerry	HighK	Total/Avg
Prcts	40	415	540	165	90	1250
Votes	3.87	40.19	52.30	15.98	8.72	121.056
Pct	3.2%	33.2%	43.2%	13.2%	7.2%	100.0%
						
RESP.	56.5%	54.9%	52.0%	54.7%	52.5%	53.5%
Dev	3.0%	1.4%	-1.5%	1.2%	-1.0%	0.0%
						
ALPHA						
K/B	1.503	1.153	1.174	1.098	0.996	1.154
Dev	30.2%	-0.2%	1.7%	-4.9%	-13.7%	0.0%
ln a	0.407	0.142	0.160	0.094	-0.004	0.144

VOTE						
Kerry	0.77	16.07	25.62	9.59	6.98	59.027
Pct	19.9%	40.0%	49.0%	60.0%	80.0%	48.76%
						
Bush	3.10	24.12	26.68	6.39	1.74	62.029
Pct	80.1%	60.0%	51.0%	40.0%	20.0%	51.24%
						
RESPONDERS						
Kerry	0.96	17.30	27.84	10.07	6.96	63.127
Pct	24.9%	43.0%	53.2%	63.0%	79.9%	52.15%
						
Bush	2.91	22.89	24.46	5.91	1.75	57.929
Pct	75.1%	57.0%	46.8%	37.0%	20.1%	47.85%
						
REFUSERS						
Kerry	13.4%	36.3%	44.4%	56.5%	80.2%	44.87%
Bush	86.6%	63.7%	55.6%	43.5%	19.8%	55.13%

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

[Time for change]

16. I understood that part

Here's the part I'm having the biggest problem with:

"...“E/M hypothetical” should be interpreted as referring to hypothetical average (rather than constant average) partisan exit poll response rates. In this interpretation, average precinct partisan response rates may vary widely by reported precinct vote shares..."

This seems to be the key part because it is apparently E-M's latest interpretation, which USCV says they have refuted. Well, maybe it will be explained in an understandable way in the body of the paper.

[Bill Bored]

20. This may be a veiled reference to the bias index

The index showed that WPE is not correlated with partisanship. Since WPE is a function of Completion Rate, I think they're saying that Mitofsky is saying that "average precinct partisan response rates" (OMG!) are also not correlated to precinct partisanship (vote share).

How does that grab you?

[Time for change]

23. I think you're right

Though I have to say that this almost seems to be purposely written in a way that no one will be able to understand it. I hope the main body of the report is easier to read than the abstract.

[NC Beach Girl]

8. LOL

I was just thinking the same thing...this is pretty tough to get through.

[Bill Bored]

21. Please define "reasonable number!" nt

[eomer]

5. Hey Bill, I'll give it a try.

Following your lead, my answer is not about who is right or
wrong, it's just about WTF the term constant average means.

I believe they mean constant across partisanship bands.  Here
are statistics for two theoretical sets of data, one of which
has constant averages as you span across partisanship bands
and the other of which has varying averages as you span across
partisanship bands:

Constant Average:
                 Bush      Kerry
                 Voter     Voter
               Average   Average
               Response  Response
Bush   Kerry     Rate      Rate
 0.1    0.9       50        56
 0.3    0.7       50        56
 0.5    0.5       50        56
 0.7    0.3       50        56
 0.9    0.1       50        56
Overall           50        56

Varying Averages:
                 Bush      Kerry
                 Voter     Voter
               Average   Average
               Response  Response
Bush   Kerry     Rate      Rate
 0.1    0.9       46        60
 0.3    0.7       48        58
 0.5    0.5       50        56
 0.7    0.3       52        54
 0.9    0.1       54        52
Overall           50        56

The first set of data is what you would expect if there is a
universal and consistent tendency of Bush voters to be more
reluctant than Kerry voters.

The second set of data is what you would expect if there is a
tendency of all voters (Bush and Kerry) to be more reluctant
in enemy territory and less reluctant in friendly territory
combined with an overall tendency of Bush voters to be more
reluctant than Kerry voters.

[Bill Bored]

13. Well, I have to say, the first one is ridiculous!

It doesn't seem likely that exactly 56% of ALL Bush voters would answer the exit polls. I don't think I can prove this off the top of my head, but it certainly seems intuitively obvious and a waste of time to even consider as a possibility. Has anyone seriously suggested this?

The second one is a reasonable hypothesis, but I can also see how you could get mean response rates of 56 and 50 without them being correlated to precinct partisanship. Febble's bias index is not correlated to precinct partisanship either.

Meanwhile, I have shown a NEGATIVE correlation between Kerry Vote % and Completion Rate using the state data (r = -.28, p = .05) and it's significant. So if anything, at the state level, Kerry voters were LESS likely to respond to the exit polls.

[TruthIsAll]

14. No one ever said EXACTLY for anything.

But the KEY POINT is that E-M's WPE's are INCOMPATIBLE with a weighted average 1.12 alpha. Alpha must be at least 1.154.

Absolute constant alpha across all partisanships is of course not to be expected. But Kerry strongholds have a 1.06 alpha; the rest are 1.18.

How does that jibe with rBr?

[LightningFlash]

15. It doesn't at all. That's called the whitewash. n/t

[eomer]

18. Sorry, I wasn't too clear about what I meant.

I meant those statistics as theoretical data designed to illustrate the point, not to represent what you would see in real life. In real life the averages would be only approximately equal, not exactly equal as shown in my post. There would be at least some minimal difference from one partisanship band to another due to random variation even if the underlying tendency of the full population were somehow magically consistent.

I agree that it seems highly unlikely that the response rates would be 50% and 56% respectively and that these rates would be universal and consistent. If USCV were arguing that this hypothetical is or may be the case then it would be worth our time dwelling on it. Since what they are saying is that it is not the case and Mitofsky agrees and we all seem to agree then I guess we just all agree.

It seems to me that in reality response rates are going to vary and be correlated to a number of different factors. I would be very surprised if there wasn't some correlation with most if not all of the following:

1. geographical region of the country
   
2. urban vs. suburban vs. rural
   
3. race and/or ethnicity of voter
   
4. race and/or ethnicity of poll taker
   
5. gender of voter
   
6. gender of poll taker
   
7. age of voter
   
8. age of poll taker
   
9. partisanship of voter
   
10. partisanship of poll taker if it is outwardly apparent (t-shirt,...)
    
11. perceived partisanship of polling organization (if questionnaire has logo of some news organization, for example).
    
12. etc.
    
13. etc.
    
14. etc.
    

Whether the sum of all those correlations adds up to favor Bush or Kerry is a total conjecture unless we have some data to go by. Which leads me to ask you to point me to your work on the negative correlation at the state level. I must have missed it. That is a very important conclusion.


edit: add another factor to the list

[Bill Bored]

19. Here:

Greg,

OK, then why doesn't everybody just drop the term
"constant" and be done with it? This may have been
brought up in that recent 427-some-odd-post thread but I can't
recall the outcome. It's just a mental diversion for them to
keep bringing it up.

I've also seen the term "uniform" which might be a
better way to describe the average (in terms of a
distribution) rather than "constant." Just my 2
cents.

The other thing that drives me crazy is the interchangeable
use of the terms "Completion" and
"Response" to describe the action of a voter who
took the exit poll. It rasies the question of whether all the
questions on the survery were answered or not
("completed") and whether surveys were discarded if
they were not. Just muddies the waters IMO. I think these two
terms are the same and the reason why some questions have a
much lower n-value than others is because they didn't appear
on ALL the bloody surveys! The "How voted in Y2K"
question is one example of this. Drives me absolutely insane!

As to my "work", here's the relevant columns from
the spreadsheet:

	       K Vote	Compl
AllPrecincts		53.2
NationalSurvey		52.8
Alabama 	37.08	58.3
Alaska  	36.17	53.2
Arizona 	45.03	57.3
Arkansas	44.74	60.2
California	55.21	50.5
Colorado	47.35	55.5
Connecticut	55.24	51
D.C.    	90.52	53.5
Delaware	53.82	57.5
Florida 	47.47	49
Georgia 	41.58	63.9
Hawaii  	54.37	53.4
Idaho   	30.71	63.2
Illinois	54.99	51.9
Indiana 	39.46	38.6
Iowa    	49.54	52.6
Kansas  	36.97	64.5
Kentucky	39.99	52.6
Louisiana	42.63	47.8
Maine   	54.48	61.3
Maryland	56.25	59.4
Massachusetts	62.70	56.5
Michigan	51.73	50.2
Minnesota	51.76	45.3
Mississippi	39.91	49.6
Missouri	46.33	47
Montana 	39.51	63
Nebraska	32.53	66.5
Nevada  	48.67	49.1
NewHampshire	50.68	44
NewJersey	53.13	59.7
NewMexico	49.42	56.9
NewYork 	58.79	57.9
NorthCarolina	43.72	52.6
NorthDakota	36.09	63
Ohio	        48.75	44.1
Oklahoma	34.44	53.2
Pennsylvania	51.13	46.8
RhodeIsland	60.48	44.2
SouthCarolina	41.31	59.4
SouthDakota	39.09	42.7
Tennessee	42.78	66.7
Texas	        38.49	58.3
Utah	        27.06	59.6
Vermont  	60.34	53.1
Virginia	45.65	56.4
Washington	53.60	53.8
WestVirginia	43.48	48.7
Wisconsin	50.20	55.3
Wyoming 	29.70	66

	KV vs. Compl	
r Val.	-0.27858576	
t Val.	2.01	
p Val.	0.0501	

There was also a post on Kos brought to my attention by Febble
that came to a similar conclusion way back on Jan 24, using
Kerry Margin instead of Kerry Vote % by state, although for
some reason it was NOT statistically significant. That said,
he didn't post his r values and t values, only r^2:

http://www.dailykos.com/story/2005/1/24/151634/021

There are some problems with my "work" such as the
fact that when the states (Kerry Vote) are weighted by
population (Electoral Vote) the correlation is no longer
significant. Also, if you take out the states with a Kerry
Vote < 31%, you lose significance. I'm not sure if this
matters since the precinct level arguments that I've seen so
far don't weight precincts according to population, nor do
they remove outliers (or do they?).

In any case, the trend seems to be clear at the state level:
there were lower Exit Poll Completion Rates in the states with
the highest Kerry Vote %. On its face, this is the reverse of
the rBr hypothesis, or whatever we're calling it now.

See if my Kerry Vote %s above are right. There are after all
so many versions of them!

[TruthIsAll]

28. GRAPH: Kerry Vote vs. Exit Poll completion rate

There's an inverse relationship: As Kerry's state vote percentage increases, the exit poll completion rate decreases.

Another finding which makes rBr even less plausible..

[LightningFlash]

7. Here's where the "hypothesis" is refuted again...

....According to the reluctant "cluster grouping" analysis, Kerry voters came out in full turnout in the morning or earlier afternoon.

While "Bush" voters especially females, came out in record numbers at night, also refusing to be polled at the exits in order to "properly match" the final results.

....However according to a just released "whitewash" report by the Department of Justice, in one county alone it was Kerry voters mostly all black race, who went to the polls late at night and therefore missed their opportunity, while it was WHITE female voters who went to the polls in the morning to early afternoon.

http://www.bradblog.com/archives/00001512.htm

So which is it, clearly? :shrug: You can't have ONE without the other, can you.....

I'm going to cut to the chase: It was bullshit, that's what it was. ;)

[Time for change]

11. I understood that part

Here's the part I'm having the biggest problem with:

"...“E/M hypothetical” should be interpreted as referring to hypothetical average (rather than constant average) partisan exit poll response rates. In this interpretation, average precinct partisan response rates may vary widely by reported precinct vote shares..."

This seems to be the key part because it is apparently E-M's latest interpretation, which USCV says they have refuted. Well, maybe it will be explained in an understandable way in the body of the paper.

[autorank]

29. And to think US Counts Votes almost fell apart, right here on DU.

Thank God they didn't. This is a very interesting paper. I was struck by these two paragraphs.

“The pattern of high discrepancies in the 80%+Bush vote precincts and the almost zero discrepancies in the 80%+ Kerry precincts, and large mean and media WPE levels of over 5.5^ in more competitive districts, has yet to be explained.” (p.9)

I love academic style at times. The constraint can be almost droll. I have a candidate explanation, election fraud. I understand the reasons for this restraint. USCV needs to be seen as credible and the next step is to explain this curious phenomena. They can see/have seen TIA for some good work in this area.

“We reiterate that almost seven months after the election, no plausible explanation of the 2004 exit poll discrepancy, based on exit polling error, has been provided by E/M. Moreover, the precinct level exit polling and official vote count data that would enable independent investigators to analyze the exits poll discrepancy and to confirm or reject explanations has also not been provided to the public.” (p.11)

Right on target as are their recommendations that this EM/Network consortium data files be open and available to the public.

Why?

1) The elections and the products thereof, all information pertaining to the elections, are public property.

2) As public property, the public has a right to examine them.

3) To argue otherwise is to say that we can outsource our most fundamental right -- to freely choose our leaders through fair and honest elections.

I say anyone who argues for proprietary data on the part of EM-The Network Consortium is an opponent of the American people, government, and the Constitution. I mean this as a categorical statement without any ambiguity.

OPEN THE FILES

PROVIDE THE DATA AND INFORMATION IN A VERIFIABLE FASHION.

STOP THE DECEPTION

THE ELECTIONS BELONG TO THE PEOPLE, NOT THE CORPORATIONS AND THEIR TOADIES!

[Nederland]

30. I must be out of it

What happened to the rest of the authors? USCountVotes papers used to have a dozen authors with Phd's in statistic and mathematics. Now all that is left is Kathy Dopp and Ron Baiman? What happened?

[Time for change]

31. They split apart

I'm not certain as to how many people left, but I know of at least four. Your observation might suggest that Kathy and Ron are the only two left.

[Nederland]

32. Which four are left? (nt)

[Time for change]

33. I need to word that differently

When I said that at least 4 people left, I meant that there are at least 4 people who left the group, not 4 people remaining in the group. But I believe it is much more than that, and your pointing out that only Ron and Kathy signed the paper probably indicates that the group is severely diminished, I would think.

[Fly by night]

35. With every new USCV "update", their credibility drops another notch.

That wouldn't be so bad by itself but it also reflects poorly on the rest of us.

If whoever is left at USCV has anything new to say, then say it. Otherwise, stop revising the same paper over and over again. All that appears to be doing is causing even more respected scientists to remove their endorsement from and participation in USCV. Of course, there are likely other reasons why those scientists no longer want to be connected to USCV. (I can name at least one, but not here.)

[TruthIsAll]

36. Not true, Fly By Night . USCV is updating and refining their analysis.

USCV and Ron Baiman have been consistent all along in their basic premise that rBr is implausible. Ron has provided addtitional statistical analysis to show it is even more implausible in the latest update.

In addition, my Exit Poll Optimizer has confirmed USCV analyses and added a few more insights into the fraud.

One basic result which is common to our independent analyses:
Mimimum weighted alpha (Kerry/Bush respondents) is 1.155, not 1.12 as stated by Mitofsky. That means its 58 Kerry/50 Bush, not 56/50 - and that is even more implausible.

I don't believe you have followed the math.

[Fly by night]

37. It's not about the math -- it's about the perception of inexactitude.

Though I am not sure what the actual count of "revised" USCV reports on the same analysis is now, I can remember at least three (maybe four) versions of the same paper that have now been released by USCV. This gives the impression that either the initial paper was rushed to publication prematurely or that the earlier papers were flawed or that the authors did not fully consider all the issues before publishing the paper (and the first several revisions) or that the reviewers' initial comments were not complete or taken into consideration by the authors or that USCV has nothing better to do than to keep massaging the same analysis ad nauseum.

Believe me, I fully accept the value of the exit poll analyses as supporting evidence for the 2004 election theft. But I was convinced of that fact by Steve Freeman's early papers, not by USCV's later revisions of that initial research. I know that several of you continue to focus on the minutiae of this analysis and more power to you. But since the evidence of the 2004 election theft is available to us from many different directions (some of which, like the EIRS analyses conducted by berniew1 before he was banned from DU, are more easily understood and appreciated by non-mathematicians than the exit poll research), I don't need to hang my hat on this single avenue of analysis. And neither, fortunately, does the election reform community.

Where are all the other researchers who lent their names to earlier versions of the USCV paper(s) and why have they separated themselves from this endeavor and from USCV now? I'd love to know the answers to those questions -- do you know?

[TruthIsAll]

38. Perception of inexactitude? Since when is any statistical analysis

an attempt at exactitude? Statistical analysis seeks to determine the probabilities of events.

And it's ALL about the math.

Any model is a work in progress which incorporates new information as it becomes available. USCV's Baiman responded to the recent Mitofsky AAPOR presentation with an updated analysis. What the hell is wrong with that?

Two researchers who separated from the USCV conclusions included Elizabeth Liddle and Bruce O'Dell, who have posted here. You can read up on what they have said.

Why are you being so hard on USCV? They've done a great job in analyzing the exit poll discrepancies and E-M's misleading statements which were contradicted in their own 77-page report.

[Fly by night]

40. No time to get in a pissing contest with you now. I've been PMed...

... by several readers of this thread who have given me more detailed information about the current dismantling of USCV that confirms my own sad experiences with one of their principals. No need to air it here.

Again, every time a newly revised analysis of the same data gets released, it means less and less to folks like me who are trying to accomplish election reform both nationally and in the Orange State, rather than just write about it. However, my priorities are not shared by everyone, which is fine with me.

[borealowl]

39. It's way too dry....

[borealowl]

41. I mean...who can read this?

The point is that twice in US history (as far as I know...?) exit polls election night showed one candidate ahead well into the evening, but suddenly the election reversed itself. In one case, Gore, who indisputably won the national vote, was well ahead in Florida; in the other, Kerry was well ahead both nationally and in the swing states.

In both cases, George Bush, not known for the high ethics of his previous campaigns, "won" in the end.

In both cases, thousands of voting problems, all tending in one direction and all probably deliberate (in 2004, throwing out Democratic registrations; votes flipped from Kerry to Bush; long lines in Democratic districts, etc.; in 2000, faux felon scam; Palm Beach ballots, optical scan machine feature which rejects double voted ballots turned off in black precincts, etc.) are the other side of the equation.

If you just had the exit poll discrepancy but not the voting problems, okay. If you just had the voting problems but not the exit poll discrepancy, okay. What you've got, however, is a set-up just like the Ukraine, and if you want to see how our media handled that, it shows they've got the vocabulary.

These endless discussions of the slight possibility that there was something wrong with the exit polls, adding up to the SECOND Bush victory defying the models, the polls - they just amount to behaving the way the GOP wants us to behave, instead of taking to the streets the "activists" talk about peer-reviewed papers which nobody ever sees. That's what happened in 2000 too - endless chitchat about chads, and little about the real problem - theft.

Of course, it doesn't help that Gore and Kerry conceded! That marginalizes their supporters, to put it mildly...

[TruthIsAll]

43. Exactly - and welcome to DU.

/

[autorank]

44. This an easy read, comprehensive and full of great analysis.

Of course, that's a bit self serving (as you'll see) but you may find this useful:

http://upload.democraticunderground.com/discuss/duboard.php?az=view_all&address=203x384064

USCV is the academic side of the election fraud statistical effort and TIA's work and the other great DUers who contribiute is the real-time lab producing material for the reality-based community to use on a regular basis.

Both have their roles, and they are lonely ones.

[Bill Bored]

47. No the concessions didn't help.

And in the case of Diebold Op Scans, EVERY ballot reject option is OFF by default. You actually have to turn them on if you want the voters to know that they've undervoted, overvoted, etc. Where'd you hear that they were SELECTIVELY disabled? I've thought this was the case from some time, even in 2004, but haven't heard of any proof.

Welcome to DU!

(Gee for such an obscure forum, we're getting a lot of new posters lately! Maybe we've been linked to externally, or there's some kind of pro-Democracy "Underground" out there.)

[autorank]

88. Pro-democracy underground (USA) v. Secret oligarchical pollocracy (EM)

Scoop Puts the TRUTH in Plain English, New Zealand Style

Isn't it interesting that we have to do all this work here and the other volunteer voting rights activists need to do all their work on BradBlog, SolarBus, etc. WHILE our Democratic National Committee caved in and gave away the truth about Ohio.

That's why I say...

NEW LEADERS FOR A NEW DEMOCRATIC PARTY

[johnmilton]

92. creationism vs. election fraud

I have begun posting for the simple reason that I think
the fraud thesis is sufficiently strong to be taught.

Think about it: the right is pusing to teach creationism in school
science class--what the hell are you going to teach if they mandate lets say a month on creationism. And I have a funny
feeling they don't want comparative mythology . . .

The right understands that you have to teach your pov:
We can't wait to PROVE this 100% anymore than we can
wait to PROVE the truth of global warming, we have to teach it . . .
NOW

[autorank]

93. johnmilton, Welcome to DU!!!

Here's a good resource for any teaching. You are so right, if they can teach pure speculation, why not highly informed speculation about an immanent matter vital to our country.

http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=203&topic_id=384064&mesg_id=384064

ps. I used to have "Furthest from him is best, whom reason hath equaled, force hath made supreme above his equals." Maybe I should put it back.

[johnmilton]

96. teaching the 04 election

Thanks for the shout out!

I believe very strongly that teachers have got to stick their
neck out and teach this issue; as I said in my post to the other DU'er who responded, I am not saying teach the theory as FACT
but to go in to the classroom and have a discussion about what happened, what people think happened, what the students think,
and where we want to go.

I strongly believe that the last election has soured a lot of people
who skew young; I did a poll in two class with 35 out of 50 students saying they believed fraud occurred sufficient to have
altered the outcome of the pres election, this without me teaching the issue at all.

We are at a moment when the belief in the integrity of the system is parting ways big time with what gets voiced by the system.
Teachers have got to facilitate the discussion about that gulf.

And thanks for the Paradise Lost quote; these days I teach
Samson Agonistes and Paradise Regained more often, not that I don't love PL but those other two are shorter and in a way like the 04 election: On one hand there is the message of peace that comes in PR but then there is the story of Samson, Christianity's favorite terrorist . . . and Milton published them together . . .

[autorank]

97. How luck your students are!!!!! Great, I'll read these two again. Thanks

[foo_bar]

94. because the right lives in the Dark Ages, so should we?

the right is pusing to teach creationism in school

Biblical inerrancy isn't a rightwing cause per se, plenty of DUers preach intelligent design when the thread comes around. It takes an archconservative view of science, but then so does any belief that contradicts what we know about this universe. Take the inerrancy of exit polls, a doctrine that provides more satisfying answers than "get off your butt". Why bother with elections if you can divine the results from a skewed sample (or straight from God's mouth)? Why not dispense with free will altogether? I have no problem with this in a religious context: you can believe in Set and the Infallibility of Edison-Mitofsky. But when it passes itself off as science, that's when scientists and their well-wishers have a stake in the argument.

We can't wait to PROVE this 100% anymore than we can
wait to PROVE the truth of global warming

I grok the frog in boiling water analogy, but that's why we should concentrate on the big picture instead of minutiae like "our boiling was Prophecied!". The big picture is we can't or won't audit most of our elections, so there's a big paranoid question mark where the answers should be (the too-big picture is "who watches the watcher?", how do you get a two party monopoly to police itself? I think the solution is the Sherman Antitrust Act, replacing "Standard Oil" with "RNC/DNC". Once the regional baby bells emerge, backburner issues like fair elections and global warming would be addressed on a more cellular level). For now it's "think global, act local" like the bumper sticker says. The internet (or a classroom) is the place for thinking global, but it's helpful to think globally about things that can be acted upon locally (unless you're teaching intro to epistemology).

[johnmilton]

95. framing the 04 election issue

Thanks for the response.

I think both your comments merit serious consideration,
more so than I can give in 1 post.

I agree you can introduce "intelligent design" in a biology class
but then what? Really, this is a real question. After you
introduce the notion you need to place the discussion, and in education since at least the Renaissance, intelligent design is
an argument for god and as such is taken up in either philosophy or theology. I happen to like those courses.
In my opinion, the push to teach intelligent design in a science class is simply a desire to combat evolution combined with a belief that it is more significant to teach something in a science class than in a philosophy class because science is (from this point of view) simply more important than philosophy/theology.

The latter point is true if you go by the numbers, that is by what
courses students are taking, and it is this fact that fuels the fight
to insert what is essentially a discussion about god in a science class.

I understand and share your fear about teaching ideology and non-truths (lets calls it) when talking about the 04 election. I did
not say teach that the polls are right or the election was stolen. I think your approach is a reasonable way to open this can of worms. What I believe is that we have got to open this can and I am not saying that I have the WAY to do it. I am
on this list because i want to share my thoughts with people about how to teach the whole paranoid whatever you want to call it. For this reason I will be teaching the 04 election with Freud on
paranoia, Orwell's 1984, and other stuff. I am happy to have
learned that Mark Crispin Miller's book will be out in time for me to include it (at least a bit of it) and Steve Freeman's book too.